SYSTEMS AND METHODS FOR CONTROLLING VARIABLE REFRIGERANT FLOW SYSTEMS AND EQUIPMENT USING ARTIFICIAL INTELLIGENCE MODELS

Info

Publication number: 20230116964
Type: Application
Filed: Mar 12, 2020
Publication Date: Apr 20, 2023
Applicant: Johnson Controls Tyco IP Holdings LLP (Milwaukee, WI)
Inventors: ZHONGYUE SUN (Milwaukee, WI), LIMING YANG (Mequon, WI), ROBERT D. TURNEY (Watertown, WI), LUMING WANG (Wuxi), BO FAN (Wuxi City), ZHIGANG WU (Suzhou)
Application Number: 17/910,240

Abstract

An oil management controller for heating, ventilation, or air conditioning (HVAC) equipment. The controller includes a processing circuit. The processing circuit is configured to analyze operating data for the HVAC equipment using a machine learning model to predict a variable state or condition of oil used by the HVAC equipment. The processing circuit is configured to identify an oil deficiency based on the variable state or condition of the oil. The processing circuit is configured to automatically initiate a corrective action responsive to identifying the oil deficiency.

Description

Description

BACKGROUND

The present disclosure relates generally to the field of operating building equipment and more particularly to using artificial intelligence to predict states of the building equipment.

For building equipment (e.g., heating, ventilation, or air conditioning (HVAC) equipment) to operate effectively and to minimize degradation of the building equipment, various operating conditions of the building equipment should be monitored and accounted for. However, traditional building systems leave many operating conditions unmonitored which can lead to rapid degradation of the building equipment and increased costs over time.

SUMMARY

One embodiment of the present disclosure is an oil management controller for heating, ventilation, or air conditioning (HVAC) equipment. The controller includes a processing circuit. The processing circuit is configured to analyze operating data for the HVAC equipment using a machine learning model to predict a variable state or condition of oil used by the HVAC equipment. The processing circuit is configured to identify an oil deficiency based on the variable state or condition of the oil. The processing circuit is configured to automatically initiate a corrective action responsive to identifying the oil deficiency.

In some embodiments, the variable state or condition of the oil is an amount of the oil in the HVAC equipment. Identifying the oil deficiency includes determining that the amount of the oil in the HVAC equipment is less than a threshold amount. The corrective action includes providing more oil to the HVAC equipment to increase the amount of the oil in the HVAC equipment.

In some embodiments, the variable state or condition of the oil is a viscosity of an oil-refrigerant mixture in the HVAC equipment. Identifying the oil deficiency includes determining that the viscosity of the oil-refrigerant mixture is less than a threshold viscosity. The corrective action includes providing more oil to the HVAC equipment to increase the viscosity of the oil-refrigerant mixture in the HVAC equipment.

In some embodiments, the HVAC equipment is operable at different operating speeds. The variable state or condition of the oil is a viscosity of an oil-refrigerant mixture in the HVAC equipment. The corrective action includes setting an upper limit on an operating speed of the HVAC equipment based on the viscosity of the oil-refrigerant mixture.

In some embodiments, the HVAC equipment is coupled to a refrigerant loop that circulates an oil-refrigerant mixture between the HVAC equipment and one or more other devices coupled to the refrigerant loop. The corrective action includes operating the HVAC equipment to cause the oil-refrigerant mixture to circulate within the refrigerant loop and thereby return the oil from the one or more other devices to the HVAC equipment.

In some embodiments, the machine learning model is a convolutional neural network (CNN) model having an input layer, one or more hidden layers, and an output layer. Analyzing the operating data includes providing the operating data as inputs to the input layer of the CNN model and obtaining a prediction of the variable state or condition of the oil at the output layer of the CNN model.

In some embodiments, the machine learning model is a recurrent neural network (RNN) model. Analyzing the operating data includes providing a time series of values of the operating data as an input to the RNN model and obtaining a prediction of the variable state or condition of the oil as an output of the RNN model.

In some embodiments, the processing circuit configured to generate the machine learning model using a set of training data obtained from a simulation model.

Another embodiment of the present disclosure is a method for operating heating, ventilation, or air conditioning (HVAC) equipment using a machine learning model. The method includes obtaining training data indicating conditions affecting oil used by the HVAC equipment and a variable state or condition of the oil. The method includes generating the machine learning model by performing a training process based on the training data. The machine learning model is trained to predict the variable state or condition of the oil based on the conditions affecting the oil. The method includes using the machine learning model to predict whether the variable state or condition of the oil violates a threshold. The method includes automatically initiating a corrective action responsive to predicting that the variable state or condition of the oil violates the threshold.

In some embodiments, the variable state or condition of the oil is an amount of the oil in the HVAC equipment. The threshold is a minimum threshold for the amount of the oil in the HVAC equipment. The corrective action includes providing more oil to the HVAC equipment to increase the amount of the oil in the HVAC equipment.

In some embodiments, the variable state or condition of the oil is a viscosity of an oil-refrigerant mixture in the HVAC equipment. The threshold is a minimum threshold for the viscosity of the oil-refrigerant mixture. The corrective action includes providing more oil to the HVAC equipment to increase the viscosity of the oil-refrigerant mixture in the HVAC equipment.

In some embodiments, the HVAC equipment is operable at different operating speeds. The variable state or condition of the oil is a viscosity of an oil-refrigerant mixture in the HVAC equipment. The corrective action includes setting an upper limit on an operating speed of the HVAC equipment based on the viscosity of the oil-refrigerant mixture.

In some embodiments, the HVAC equipment is coupled to a refrigerant loop that circulates an oil-refrigerant mixture between the HVAC equipment and one or more other devices coupled to the refrigerant loop. The corrective action includes operating the HVAC equipment to cause the oil-refrigerant mixture to circulate within the refrigerant loop and thereby return the oil from the one or more other devices to the HVAC equipment.

In some embodiments, the machine learning model is a convolutional neural network (CNN) model having an input layer, one or more hidden layers, and an output layer. Using the machine learning model to determine if the variable state or condition of the oil violates the constraint includes providing operating data as inputs to the input layer of the CNN model and obtaining a prediction of the variable state or condition of the oil at the output layer of the CNN model.

In some embodiments, the machine learning model is a recurrent neural network (RNN) model. Using the machine learning model to determine if the variable state or condition of the oil violates the constraint includes providing a time series of values of operating data as an input to the RNN model and obtaining a prediction of the variable state or condition of the oil as an output of the RNN model.

In some embodiments, obtaining the training data includes obtaining a simulation model that simulates operation of the HVAC equipment and changes to the variable state or condition of the oil over time. Obtaining the training data includes executing the simulation model to generate the training data.

Another embodiment of the present disclosure is an environmental control system for a building. The system includes heating, ventilation, or air conditioning (HVAC) equipment operable to affect an environmental condition of the building. The system includes a controller including a processing circuit. The processing circuit is configured to analyze operating data for the HVAC equipment using a machine learning model to predict a variable state or condition of oil used by the HVAC equipment. The processing circuit is configured to identify an oil deficiency based on the variable state or condition of the oil. The processing circuit is configured to automatically initiate a corrective action responsive to identifying the oil deficiency.

In some embodiments, the variable state or condition of the oil is an amount of the oil in the HVAC equipment. Identifying the oil deficiency includes determining that the amount of the oil in the HVAC equipment is less than a threshold amount. The corrective action includes providing more oil to the HVAC equipment to increase the amount of the oil in the HVAC equipment.

In some embodiments, the variable state or condition of the oil is a viscosity of an oil-refrigerant mixture in the HVAC equipment. Identifying the oil deficiency includes determining that the viscosity of the oil-refrigerant mixture is less than a threshold viscosity. The corrective action includes providing more oil to the HVAC equipment to increase the viscosity of the oil-refrigerant mixture in the HVAC equipment.

In some embodiments, the HVAC equipment is operable at different operating speeds. The variable state or condition of the oil is a viscosity of an oil-refrigerant mixture in the HVAC equipment. The corrective action includes setting an upper limit on an operating speed of the HVAC equipment based on the viscosity of the oil-refrigerant mixture.

Another embodiment of the present disclosure is a controller for operating a motor of a compressor in a heating, ventilation, or air conditioning (HVAC) system, according to some embodiments. The control includes a processing circuit. The processing circuit is configured to obtain a machine learning model that predicts amplitude setpoints for an electric current provided to the motor. The amplitude setpoints affect vibrations of the motor. The processing circuit is configured to analyze operating data for the motor using the machine learning model to predict an amplitude setpoint for the electric current. The processing circuit is configured to operate the motor based on the amplitude setpoint.

In some embodiments, operating the motor based on the amplitude setpoint includes providing the amplitude setpoint to an inverter. The inverter is configured to provide the electric current to the motor.

In some embodiments, inputs to the machine learning model include at least one of an axial error between a direct axis and a quadrature axis of the motor, a frequency of the electric current, a real noise level associated with the motor, or a required noise level associated with the motor.

In some embodiments, the machine learning model is trained to learn a correlation between the operating data and an amount of noise produced by the motor. The amount of noise is used as a proxy for predicting the vibrations of the motor.

In some embodiments, the processing circuit further configured to generate the machine learning model using a set of training data obtained from a simulation model.

some embodiments, the machine learning model is a recurrent neural network (RNN) model. Analyzing the operating data includes providing a time series of values of the operating data as an input to the RNN model and obtaining a prediction of the amplitude setpoint as an output of the RNN model.

In some embodiments, the electric current is an alternating current (AC). A frequency of the AC affects a rotational speed of the motor and an amplitude of the AC affects a torque applied by the motor.

Another embodiment of the present disclosure is a controller for predicting faults of a variable refrigerant flow (VRF) system. The controller includes a processing circuit. The processing circuit is configured to analyze operating data for the VRF system using a machine learning model to predict a fault classification for the VRF system. The fault classification identifies a fault condition affecting the VRF system. The processing circuit is configured to identify a VRF device of the VRF system associated with the fault condition. The processing circuit is configured to automatically initiate a corrective action to address the fault condition responsive to identifying the device and the fault condition.

In some embodiments, the fault classification includes a severity metric associated with the fault condition. The severity metric indicates an influence of the fault condition on the VRF system. The corrective action is determined based on the severity metric.

In some embodiments, the machine learning model is a recurrent neural network (RNN) model. Analyzing the operating data includes providing a time series of values of the operating data as an input to the RNN model and obtaining a prediction of the fault classification as an output of the RNN model.

In some embodiments, the processing circuit further configured to generate the machine learning model using a set of training data obtained from a simulation model.

In some embodiments, the fault classification identifies multiple fault conditions affecting the VRF system. The multiple fault conditions are associated with multiple VRF devices of the VRF system.

In some embodiments, the fault condition is at least one of leakage of a refrigerant, frosting of an outdoor unit, clogging of an indoor fan, clogging of an indoor filter, clogging of a heat exchanger, clogging of an outdoor fan, demagnetization of a motor, or leakage of oil from a compressor.

Another embodiment of the present disclosure is a method for operating a motor of a compressor in a heating, ventilation, or air conditioning (HVAC) system, according to some embodiments. The method includes obtaining a machine learning model that predicts amplitude setpoints for an electric current provided to the motor. The amplitude setpoints affect vibrations of the motor. The method includes analyzing operating data for the motor using the machine learning model to predict an amplitude setpoint for the electric current. The method includes operating the motor based on the amplitude setpoint.

In some embodiments, operating the motor based on the amplitude setpoint includes providing the amplitude setpoint to an inverter. The inverter is configured to provide the electric current to the motor.

In some embodiments, inputs to the machine learning model include at least one of an axial error between a direct axis and a quadrature axis of the motor, a frequency of the electric current, a real noise level associated with the motor, or a required noise level associated with the motor.

In some embodiments, the machine learning model is trained to learn a correlation between the operating data and an amount of noise produced by the motor. The amount of noise is used as a proxy for predicting the vibrations of the motor.

In some embodiments, the method includes generating the machine learning model using a set of training data obtained from a simulation model.

In some embodiments, the machine learning model is a recurrent neural network (RNN) model. Analyzing the operating data includes providing a time series of values of the operating data as an input to the RNN model and obtaining a prediction of the amplitude setpoint as an output of the RNN model.

In some embodiments, the electric current is an alternating current (AC). A frequency of the AC affects a rotational speed of the motor and an amplitude of the AC affects a torque applied by the motor.

Those skilled in the art will appreciate that the summary is illustrative only and is not intended to be in any way limiting. Other aspects, inventive features, and advantages of the devices and/or processes described herein, as defined solely by the claims, will become apparent in the detailed description set forth herein and taken in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

Various objects, aspects, features, and advantages of the disclosure will become more apparent and better understood by referring to the detailed description taken in conjunction with the accompanying drawings, in which like reference characters identify corresponding elements throughout. In the drawings, like reference numbers generally indicate identical, functionally similar, and/or structurally similar elements.

FIG. 1 is a drawing of a building equipped with a HVAC system, according to some embodiments.

FIG. 2 is a block diagram of a waterside system which can be used to serve the heating or cooling loads of the building of FIG. 1, according to some embodiments.

FIG. 3 is a block diagram of an airside system which can be used to serve the heating or cooling loads of the building of FIG. 1, according to some embodiments.

FIG. 4 is a block diagram of a building management system (BMS) which can be used to monitor and control the building of FIG. 1, according to some embodiments.

FIG. 5 is a block diagram of another BMS which can be used to monitor and control the building of FIG. 1, according to some embodiments.

FIGS. 6A-6B are drawings of a variable refrigerant flow (VRF) system having one or more outdoor VRF units and multiple indoor VRF units, according to some embodiments.

FIG. 7A is a schematic diagram of a VRF system, according to some embodiments.

FIG. 7B is a block diagram of a VRF system, according to some embodiments.

FIG. 8 is a block diagram of a controller for predicting characteristics of oil, according to some embodiments.

FIG. 9A is an illustration of a recurrent neural network (RNN) structure, according to some embodiments.

FIG. 9B is an illustration of a neural network (NN) architecture, according to some embodiments.

FIG. 10 is a flow diagram of a process for monitoring oil characteristics using an AI model, according to some embodiments.

FIG. 11A is a graph illustrating changes in RMSE based on a number of iterations in an example model training process for an artificial intelligence (AI) model, according to some embodiments.

FIG. 11B is a graph illustrating changes in loss based on the number of iterations associated with the AI model of FIG. 11A, according to some embodiments.

FIG. 12A is a graph illustrating predictions of an oil level of a compressor generated by the AI model of FIG. 11A, according to some embodiments.

FIG. 12B is a graph illustrating predictions of an oil level of an accumulator generated by the AI model of FIG. 11A, according to some embodiments.

FIG. 12C is a graph illustrating predictions of a viscosity of oil generated by the AI model of FIG. 11A, according to some embodiments.

FIG. 13A is a block diagram of a VRF system including a compressor vibration controller, according to some embodiments.

FIG. 13B is a block diagram of the VRF system of FIG. 13A in greater detail, according to some embodiments.

FIG. 14A is a graph illustrating Kirchhoff's law, according to some embodiments.

FIG. 14B is a graph illustrating an αβ conversion for a motor, according to some embodiments.

FIG. 14C is a graph illustrating a direct-quadrature (dq) conversion for a motor, according to some embodiments.

FIG. 15 is a graph illustrating a relationship between crank angle and torque for different types of compressors, according to some embodiments

FIG. 16 is a block diagram of the compressor vibration controller of FIG. 13A in greater detail, according to some embodiments.

FIG. 17A is an illustration of a neural network for predicting values of a current to provide to a compressor motor, according to some embodiments.

FIG. 17B is an illustration of another neural network for predicting values of a current to provide to a compressor motor, according to some embodiments.

FIG. 18 is a flow diagram of a process for predicting an AC signal amplitude to provide to a compressor using an AI model, according to some embodiments.

FIG. 19 is a block diagram of a VRF system, according to some embodiments.

FIG. 20 is a block diagram of a VRF fault controller for predicting faults in a VRF system, according to some embodiments.

FIG. 21 is an illustration of a neural network for predicting a fault classification of a VRF system, according to some embodiments.

FIG. 22 is a flow diagram of a process for predicting a fault classification for a VRF system using an AI model, according to some embodiments.

FIG. 23 is a block diagram of a motor efficiency controller, according to some embodiments.

FIG. 24 is a flow diagram of a process for predicting an efficiency of a motor in a VRF system using an AI model, according to some embodiments.

FIG. 25A is an illustration of a recurrent neural network structure, according to some embodiments.

FIG. 25B is an illustration of a long short-term memory model structure, according to some embodiments.

DETAILED DESCRIPTION Overview

Referring generally to the FIGURES, systems and methods for utilizing artificial intelligence (AI) in predicting characteristics of variable refrigerant flow (VRF) systems of a building and operating VRF systems and VRF system components are shown, according to some embodiments. In particular, the present disclosure utilizes AI to predict characteristics of oil used in VRF systems as well as for predicting various states and characteristics of motors and compressors in VRF systems.

It should be appreciated, however, that the systems and methods described herein are not limited to VRF systems. Rather, VRF systems are shown and described for sake of example only as one potential implementation of the present disclosure. The systems and methods described herein can be applied to a variety of systems (e.g., other environmental control systems) that require oil to be provided to equipment, as well as other types of systems that include compressors, motors, any type of equipment that uses oil, and/or any type of equipment that may experience vibration or faults during operation. For example, the systems and methods described herein can be applied to a variety of heating, ventilation, or air conditioning (HVAC) systems and devices (e.g., various air conditioning equipment, variable air volume (VAV) systems, residential air conditioning (RAC) systems, etc.).

As referred to herein, AI and AI models can be used to describe a variety of different models that can be used in predicting states and other information associated with devices in VRF systems. In some embodiments, recurrent neural network (RNN) models are utilized for generating predictions. RNNs are a class of artificial neural networks where connections between nodes form a directed graph along a temporal sequence. More specifically, long short-term memory (LSTM) models may be utilized in generating predictions. LSTMs are a specific type of artificial RNN architecture that are used primarily for deep learning. LSTMs can classify and process entire sequences of time-series data and can make predictions based on said time-series data. Advantageously, LSTMs can account for lags of unknown duration between important events in a time series. In some embodiments, other types of AI models such as convolutional neural networks (CNNs) are utilized in generating predictions. Accordingly, it should be appreciated that various types of AI models can be utilized in generating predictions.

As defined herein, a characteristic of oil, which is used interchangeably herein with the term “oil characteristic,” can refer to a particular property of the oil. In other words, an oil characteristic may be a variable state or condition of the oil. Oil characteristics (i.e., variable states or conditions of the oil) of a VRF system may include, for example, oil levels in one or more compressors of the VRF system, an oil level in an accumulator of the VRF system, a viscosity of the oil, etc. In some embodiments, a viscosity of an oil-refrigerant mixture is estimated instead of or in addition to the viscosity purely of the oil. In this case, the oil-refrigerant mixture may be outputted by compressors of the VRF system as a result of oil getting integrated into compressed refrigerant typically outputted by the compressors. As the VRF system is operated, the characteristics of the oil may change over time, thereby leading to changes in operation of building devices (e.g., compressors, valves, oil separators, etc.) in the VRF system. For example, if a compressor using the oil operates at a higher speed, a level of oil in the compressor may decrease and a viscosity of the oil may decrease due to a higher operating temperature of the compressor.

Specifically with regard to a VRF system, traditional time-based oil return systems may periodically interrupt heating and/or cooling provided by the VRF system and may lower overall efficiency of the VRF system. In a refrigeration cycle, oil drift and other associated issues may result in a specific indoor heat exchanger or an outdoor heat exchanger being utilized for an extended period of time which is often detrimental to overall operating conditions of the VRF systems. Further, if VRF devices (e.g., compressors) fail to receive enough oil, the VRF devices may be at a higher risk of failure. While the traditional oil return systems can mitigate failure and other catastrophic issues associated with inaccurate oil return, there is often a significant impact on efficiency of the VRF systems. For example, in periodic oil return, after transitioning to the oil return state, a compressor of the VRF system may need to decrease an operating speed and/or restart, thereby resulting in efficiency loss for the heating/cooling system.

As described in detail below, problems associated with traditional oil return systems (e.g., time-based oil return) can be addressed through utilization of AI. AI can be used to predict an oil level in various building devices and viscosity of the oil. Based on said predictions, the AI can determine an optimal time to perform oil return when a VRF system is in cooling/heating operation. Advantageously, the predictions performed by the AI can be made without the use of an oil sensor. Not having to utilize oil sensors to detect oil states (e.g., oil level, oil viscosity, etc.) can reduce costs as fewer components need to be purchased and maintained.

In some embodiments, AI models can be leveraged to manage vibrations associated with compressors of a VRF system. Excessive vibrations of the compressors can lead to rapid degradation and thereby increased costs over a time period due to higher operational costs, maintenance costs, and replacement costs. To manage the vibrations, a particular AI model (e.g., an RNN model) can be trained to predict target currents associated with a direct axis (D-axis) and a quadrature axis (Q-axis). Prediction of the target currents can be based on inputs associated with the compressors such as, for example, a frequency provided by an inverter, a real noise level, a q-axis feedback current, an axial error between axes, etc. Using the predicted D-axis current and Q-axis current, a correlated vibration of the compressors can be determined. If predicted currents and/or predicted vibrations are too high (e.g., the predicted currents are higher than thresholds on the current values), operation of the compressors can be modified to avoid excessive vibrations and thereby avoid rapid degradation of the compressors. Specifically, a constant input current and speed control can be implemented to reduce vibrations.

In some embodiments, AI models can be leveraged to predict fault conditions of compressors in a VRF system. Fault conditions can refer to operating states of the compressors that are beyond preferred operating conditions. For example, fault conditions in a VRF system can include refrigerant leakage, frosting of an outdoor unit, clogging of an indoor fan, a dirty indoor filter, a dirty heat exchanger, a dirty outdoor fan, motor demagnetization, compressor oil leakage, and other conditions that result in imperfect efficiency of compressors. To predict conditions, an AI model (e.g., an RNN model) can be trained to map values of inputs associated with compressors to a fault classification. A fault classification can include an indication of what fault conditions (if any) are identified for a VRF system based on a set of input data. For example, the AI model may utilize inputs such as a compressor speed, an ambient temperature, a discharge temperature, a suction pressure, a discharge pressure, an indoor fan mode, an outdoor fan step, etc. to generate a fault classification. Based on the fault classification, various corrective actions can be initiated in response. As defined herein, a corrective action can refer to any action taken to address a fault and/or some undesirable condition. For example, corrective actions may include scheduling maintenance, alerting a building operator to a fault, disabling certain devices (e.g., certain compressors), operating specific devices, etc. In this way, faults affecting the compressors can be addressed quickly and efficiently.

In some embodiments, AI models are utilized to predict an efficiency of a motor in a VRF system (or other system). Motor efficiency can directly impact costs over a time period, in particular if a motor is operated at a high level of inefficiency. As efficiency of a motor decreases, costs may increase due to additional resources (e.g., electricity, water, etc.) being consumed to generate a desired output. Moreover, the motor may degrade at a quicker rate as a result of needing to operate at a more intensive operating state (e.g., at a higher rotation per minute) to generate a desired output. In some embodiments, a current efficiency of a motor can be represented as a percentage of how efficient the motor is in comparison to a maximum efficiency value when the motor was originally installed. The motor efficiency can be determined based on an amount of resources required as input to result in a specific change. For example, efficiency of an electric motor in a compressor may be defined by an amount of electricity required to compress a predefined amount of a gas (e.g., air). In the example, a 50% efficiency of the electric motor may indicate that the motor is half as efficient as it was when originally installed and thereby requires twice as much electricity to compress a predefined amount of a gas. To predict an efficiency of a motor, an AI model can be trained to map a variety of inputs (e.g., an amount of power consumed by the motor, an ambient temperature, a rotations per minute (RPM) value of the motor, etc.) to an efficiency value of the motor.

It should be noted that while AI models used to predict vibrations, fault conditions, and motor efficiency are described separately, the present disclosure also contemplates one or more aggregate AI models that predicts one or more of the aforementioned prediction targets. For example, a single AI model may be trained to predict currents associated with vibrations and fault conditions that may be affecting compressors. These and other features of the present disclosure are discussed in detail below.

Building HVAC Systems and Building Management Systems

Referring now to FIGS. 1-5, several building management systems (BMS) and HVAC systems in which the systems and methods of the present disclosure can be implemented are shown, according to some embodiments. In brief overview, FIG. 1 shows a building 10 equipped with a HVAC system 100. FIG. 2 is a block diagram of a waterside system 200 which can be used to serve building 10. FIG. 3 is a block diagram of an airside system 300 which can be used to serve building 10. FIG. 4 is a block diagram of a BMS which can be used to monitor and control building 10. FIG. 5 is a block diagram of another BMS which can be used to monitor and control building 10.

Building and HVAC System

Referring particularly to FIG. 1, a perspective view of a building 10 is shown. Building 10 is served by a BMS. A BMS is, in general, a system of devices configured to control, monitor, and manage equipment in or around a building or building area. A BMS can include, for example, a HVAC system, a security system, a lighting system, a fire alerting system, any other system that is capable of managing building functions or devices, or any combination thereof.

The BMS that serves building 10 includes a HVAC system 100. HVAC system 100 can include a plurality of HVAC devices (e.g., heaters, chillers, air handling units, pumps, fans, thermal energy storage, etc.) configured to provide heating, cooling, ventilation, or other services for building 10. For example, HVAC system 100 is shown to include a waterside system 120 and an airside system 130. Waterside system 120 may provide a heated or chilled fluid to an air handling unit of airside system 130. Airside system 130 may use the heated or chilled fluid to heat or cool an airflow provided to building 10. An exemplary waterside system and airside system which can be used in HVAC system 100 are described in greater detail with reference to FIGS. 2-3.

HVAC system 100 is shown to include a chiller 102, a boiler 104, and a rooftop air handling unit (AHU) 106. Waterside system 120 may use boiler 104 and chiller 102 to heat or cool a working fluid (e.g., water, glycol, etc.) and may circulate the working fluid to AHU 106. In various embodiments, the HVAC devices of waterside system 120 can be located in or around building 10 (as shown in FIG. 1) or at an offsite location such as a central plant (e.g., a chiller plant, a steam plant, a heat plant, etc.). The working fluid can be heated in boiler 104 or cooled in chiller 102, depending on whether heating or cooling is required in building 10. Boiler 104 may add heat to the circulated fluid, for example, by burning a combustible material (e.g., natural gas) or using an electric heating element. Chiller 102 may place the circulated fluid in a heat exchange relationship with another fluid (e.g., a refrigerant) in a heat exchanger (e.g., an evaporator) to absorb heat from the circulated fluid. The working fluid from chiller 102 and/or boiler 104 can be transported to AHU 106 via piping 108.

AHU 106 may place the working fluid in a heat exchange relationship with an airflow passing through AHU 106 (e.g., via one or more stages of cooling coils and/or heating coils). The airflow can be, for example, outside air, return air from within building 10, or a combination of both. AHU 106 may transfer heat between the airflow and the working fluid to provide heating or cooling for the airflow. For example, AHU 106 can include one or more fans or blowers configured to pass the airflow over or through a heat exchanger containing the working fluid. The working fluid may then return to chiller 102 or boiler 104 via piping 110.

Airside system 130 may deliver the airflow supplied by AHU 106 (i.e., the supply airflow) to building 10 via air supply ducts 112 and may provide return air from building 10 to AHU 106 via air return ducts 114. In some embodiments, airside system 130 includes multiple variable air volume (VAV) units 116. For example, airside system 130 is shown to include a separate VAV unit 116 on each floor or zone of building 10. VAV units 116 can include dampers or other flow control elements that can be operated to control an amount of the supply airflow provided to individual zones of building 10. In other embodiments, airside system 130 delivers the supply airflow into one or more zones of building 10 (e.g., via supply ducts 112) without using intermediate VAV units 116 or other flow control elements. AHU 106 can include various sensors (e.g., temperature sensors, pressure sensors, etc.) configured to measure attributes of the supply airflow. AHU 106 may receive input from sensors located within AHU 106 and/or within the building zone and may adjust the flow rate, temperature, or other attributes of the supply airflow through AHU 106 to achieve setpoint conditions for the building zone.

Waterside System

Referring now to FIG. 2, a block diagram of a waterside system 200 is shown, according to some embodiments. In various embodiments, waterside system 200 may supplement or replace waterside system 120 in HVAC system 100 or can be implemented separate from HVAC system 100. When implemented in HVAC system 100, waterside system 200 can include a subset of the HVAC devices in HVAC system 100 (e.g., boiler 104, chiller 102, pumps, valves, etc.) and may operate to supply a heated or chilled fluid to AHU 106. The HVAC devices of waterside system 200 can be located within building 10 (e.g., as components of waterside system 120) or at an offsite location such as a central plant.

In FIG. 2, waterside system 200 is shown as a central plant having a plurality of subplants 202-212. Subplants 202-212 are shown to include a heater subplant 202, a heat recovery chiller subplant 204, a chiller subplant 206, a cooling tower subplant 208, a hot thermal energy storage (TES) subplant 210, and a cold thermal energy storage (TES) subplant 212. Subplants 202-212 consume resources (e.g., water, natural gas, electricity, etc.) from utilities to serve thermal energy loads (e.g., hot water, cold water, heating, cooling, etc.) of a building or campus. For example, heater subplant 202 can be configured to heat water in a hot water loop 214 that circulates the hot water between heater subplant 202 and building 10. Chiller subplant 206 can be configured to chill water in a cold water loop 216 that circulates the cold water between chiller subplant 206 building 10. Heat recovery chiller subplant 204 can be configured to transfer heat from cold water loop 216 to hot water loop 214 to provide additional heating for the hot water and additional cooling for the cold water. Condenser water loop 218 may absorb heat from the cold water in chiller subplant 206 and reject the absorbed heat in cooling tower subplant 208 or transfer the absorbed heat to hot water loop 214. Hot TES subplant 210 and cold TES subplant 212 may store hot and cold thermal energy, respectively, for subsequent use.

Hot water loop 214 and cold water loop 216 may deliver the heated and/or chilled water to air handlers located on the rooftop of building 10 (e.g., AHU 106) or to individual floors or zones of building 10 (e.g., VAV units 116). The air handlers push air past heat exchangers (e.g., heating coils or cooling coils) through which the water flows to provide heating or cooling for the air. The heated or cooled air can be delivered to individual zones of building 10 to serve thermal energy loads of building 10. The water then returns to subplants 202-212 to receive further heating or cooling.

Although subplants 202-212 are shown and described as heating and cooling water for circulation to a building, it is understood that any other type of working fluid (e.g., glycol, CO2, etc.) can be used in place of or in addition to water to serve thermal energy loads. In other embodiments, subplants 202-212 may provide heating and/or cooling directly to the building or campus without requiring an intermediate heat transfer fluid. These and other variations to waterside system 200 are within the teachings of the present disclosure.

Each of subplants 202-212 can include a variety of equipment configured to facilitate the functions of the subplant. For example, heater subplant 202 is shown to include a plurality of heating elements 220 (e.g., boilers, electric heaters, etc.) configured to add heat to the hot water in hot water loop 214. Heater subplant 202 is also shown to include several pumps 222 and 224 configured to circulate the hot water in hot water loop 214 and to control the flow rate of the hot water through individual heating elements 220. Chiller subplant 206 is shown to include a plurality of chillers 232 configured to remove heat from the cold water in cold water loop 216. Chiller subplant 206 is also shown to include several pumps 234 and 236 configured to circulate the cold water in cold water loop 216 and to control the flow rate of the cold water through individual chillers 232.

Heat recovery chiller subplant 204 is shown to include a plurality of heat recovery heat exchangers 226 (e.g., refrigeration circuits) configured to transfer heat from cold water loop 216 to hot water loop 214. Heat recovery chiller subplant 204 is also shown to include several pumps 228 and 230 configured to circulate the hot water and/or cold water through heat recovery heat exchangers 226 and to control the flow rate of the water through individual heat recovery heat exchangers 226. Cooling tower subplant 208 is shown to include a plurality of cooling towers 238 configured to remove heat from the condenser water in condenser water loop 218. Cooling tower subplant 208 is also shown to include several pumps 240 configured to circulate the condenser water in condenser water loop 218 and to control the flow rate of the condenser water through individual cooling towers 238.

Hot TES subplant 210 is shown to include a hot TES tank 242 configured to store the hot water for later use. Hot TES subplant 210 may also include one or more pumps or valves configured to control the flow rate of the hot water into or out of hot TES tank 242. Cold TES subplant 212 is shown to include cold TES tanks 244 configured to store the cold water for later use. Cold TES subplant 212 may also include one or more pumps or valves configured to control the flow rate of the cold water into or out of cold TES tanks 244.

In some embodiments, one or more of the pumps in waterside system 200 (e.g., pumps 222, 224, 228, 230, 234, 236, and/or 240) or pipelines in waterside system 200 include an isolation valve associated therewith. Isolation valves can be integrated with the pumps or positioned upstream or downstream of the pumps to control the fluid flows in waterside system 200. In various embodiments, waterside system 200 can include more, fewer, or different types of devices and/or subplants based on the particular configuration of waterside system 200 and the types of loads served by waterside system 200.

Airside System

Referring now to FIG. 3, a block diagram of an airside system 300 is shown, according to some embodiments. In various embodiments, airside system 300 may supplement or replace airside system 130 in HVAC system 100 or can be implemented separate from HVAC system 100. When implemented in HVAC system 100, airside system 300 can include a subset of the HVAC devices in HVAC system 100 (e.g., AHU 106, VAV units 116, ducts 112-114, fans, dampers, etc.) and can be located in or around building 10. Airside system 300 may operate to heat or cool an airflow provided to building 10 using a heated or chilled fluid provided by waterside system 200.

In FIG. 3, airside system 300 is shown to include an economizer-type air handling unit (AHU) 302. Economizer-type AHUs vary the amount of outside air and return air used by the air handling unit for heating or cooling. For example, AHU 302 may receive return air 304 from building zone 306 via return air duct 308 and may deliver supply air 310 to building zone 306 via supply air duct 312. In some embodiments, AHU 302 is a rooftop unit located on the roof of building 10 (e.g., AHU 106 as shown in FIG. 1) or otherwise positioned to receive both return air 304 and outside air 314. AHU 302 can be configured to operate exhaust air damper 316, mixing damper 318, and outside air damper 320 to control an amount of outside air 314 and return air 304 that combine to form supply air 310. Any return air 304 that does not pass through mixing damper 318 can be exhausted from AHU 302 through exhaust damper 316 as exhaust air 322.

Each of dampers 316-320 can be operated by an actuator. For example, exhaust air damper 316 can be operated by actuator 324, mixing damper 318 can be operated by actuator 326, and outside air damper 320 can be operated by actuator 328. Actuators 324-328 may communicate with an AHU controller 330 via a communications link 332. Actuators 324-328 may receive control signals from AHU controller 330 and may provide feedback signals to AHU controller 330. Feedback signals can include, for example, an indication of a current actuator or damper position, an amount of torque or force exerted by the actuator, diagnostic information (e.g., results of diagnostic tests performed by actuators 324-328), status information, commissioning information, configuration settings, calibration data, and/or other types of information or data that can be collected, stored, or used by actuators 324-328. AHU controller 330 can be an economizer controller configured to use one or more control algorithms (e.g., state-based algorithms, extremum seeking control (ESC) algorithms, proportional-integral (PI) control algorithms, proportional-integral-derivative (PID) control algorithms, model predictive control (MPC) algorithms, feedback control algorithms, etc.) to control actuators 324-328.

Still referring to FIG. 3, AHU 302 is shown to include a cooling coil 334, a heating coil 336, and a fan 338 positioned within supply air duct 312. Fan 338 can be configured to force supply air 310 through cooling coil 334 and/or heating coil 336 and provide supply air 310 to building zone 306. AHU controller 330 may communicate with fan 338 via communications link 340 to control a flow rate of supply air 310. In some embodiments, AHU controller 330 controls an amount of heating or cooling applied to supply air 310 by modulating a speed of fan 338.

Cooling coil 334 may receive a chilled fluid from waterside system 200 (e.g., from cold water loop 216) via piping 342 and may return the chilled fluid to waterside system 200 via piping 344. Valve 346 can be positioned along piping 342 or piping 344 to control a flow rate of the chilled fluid through cooling coil 334. In some embodiments, cooling coil 334 includes multiple stages of cooling coils that can be independently activated and deactivated (e.g., by AHU controller 330, by BMS controller 366, etc.) to modulate an amount of cooling applied to supply air 310.

Heating coil 336 may receive a heated fluid from waterside system 200(e.g., from hot water loop 214) via piping 348 and may return the heated fluid to waterside system 200 via piping 350. Valve 352 can be positioned along piping 348 or piping 350 to control a flow rate of the heated fluid through heating coil 336. In some embodiments, heating coil 336 includes multiple stages of heating coils that can be independently activated and deactivated (e.g., by AHU controller 330, by BMS controller 366, etc.) to modulate an amount of heating applied to supply air 310.

Each of valves 346 and 352 can be controlled by an actuator. For example, valve 346 can be controlled by actuator 354 and valve 352 can be controlled by actuator 356. Actuators 354-356 may communicate with AHU controller 330 via communications links 358-360. Actuators 354-356 may receive control signals from AHU controller 330 and may provide feedback signals to controller 330. In some embodiments, AHU controller 330 receives a measurement of the supply air temperature from a temperature sensor 362 positioned in supply air duct 312 (e.g., downstream of cooling coil 334 and/or heating coil 336). AHU controller 330 may also receive a measurement of the temperature of building zone 306 from a temperature sensor 364 located in building zone 306.

In some embodiments, AHU controller 330 operates valves 346 and 352 via actuators 354-356 to modulate an amount of heating or cooling provided to supply air 310 (e.g., to achieve a setpoint temperature for supply air 310 or to maintain the temperature of supply air 310 within a setpoint temperature range). The positions of valves 346 and 352 affect the amount of heating or cooling provided to supply air 310 by cooling coil 334 or heating coil 336 and may correlate with the amount of energy consumed to achieve a desired supply air temperature. AHU 330 may control the temperature of supply air 310 and/or building zone 306 by activating or deactivating coils 334-336, adjusting a speed of fan 338, or a combination of both.

Still referring to FIG. 3, airside system 300 is shown to include a building management system (BMS) controller 366 and a client device 368. BMS controller 366 can include one or more computer systems (e.g., servers, supervisory controllers, subsystem controllers, etc.) that serve as system level controllers, application or data servers, head nodes, or master controllers for airside system 300, waterside system 200, HVAC system 100, and/or other controllable systems that serve building 10. BMS controller 366 may communicate with multiple downstream building systems or subsystems (e.g., HVAC system 100, a security system, a lighting system, waterside system 200, etc.) via a communications link 370 according to like or disparate protocols (e.g., LON, BACnet, etc.). In various embodiments, AHU controller 330 and BMS controller 366 can be separate (as shown in FIG. 3) or integrated. In an integrated implementation, AHU controller 330 can be a software module configured for execution by a processor of BMS controller 366.

In some embodiments, AHU controller 330 receives information from BMS controller 366 (e.g., commands, setpoints, operating boundaries, etc.) and provides information to BMS controller 366 (e.g., temperature measurements, valve or actuator positions, operating statuses, diagnostics, etc.). For example, AHU controller 330 may provide BMS controller 366 with temperature measurements from temperature sensors 362-364, equipment on/off states, equipment operating capacities, and/or any other information that can be used by BMS controller 366 to monitor or control a variable state or condition within building zone 306.

Client device 368 can include one or more human-machine interfaces or client interfaces (e.g., graphical user interfaces, reporting interfaces, text-based computer interfaces, client-facing web services, web servers that provide pages to web clients, etc.) for controlling, viewing, or otherwise interacting with HVAC system 100, its subsystems, and/or devices. Client device 368 can be a computer workstation, a client terminal, a remote or local interface, or any other type of user interface device. Client device 368 can be a stationary terminal or a mobile device. For example, client device 368 can be a desktop computer, a computer server with a user interface, a laptop computer, a tablet, a smartphone, a PDA, or any other type of mobile or non-mobile device. Client device 368 may communicate with BMS controller 366 and/or AHU controller 330 via communications link 372.

Building Management Systems

Referring now to FIG. 4, a block diagram of a building management system (BMS) 400 is shown, according to some embodiments. BMS 400 can be implemented in building 10 to automatically monitor and control various building functions. BMS 400 is shown to include BMS controller 366 and a plurality of building subsystems 428. Building subsystems 428 are shown to include a building electrical subsystem 434, an information communication technology (ICT) subsystem 436, a security subsystem 438, a HVAC subsystem 440, a lighting subsystem 442, a lift/escalators subsystem 432, and a fire safety subsystem 430. In various embodiments, building subsystems 428 can include fewer, additional, or alternative subsystems. For example, building subsystems 428 may also or alternatively include a refrigeration subsystem, an advertising or signage subsystem, a cooking subsystem, a vending subsystem, a printer or copy service subsystem, or any other type of building subsystem that uses controllable equipment and/or sensors to monitor or control building 10. In some embodiments, building subsystems 428 include waterside system 200 and/or airside system 300, as described with reference to FIGS. 2-3.

Each of building subsystems 428 can include any number of devices, controllers, and connections for completing its individual functions and control activities. HVAC subsystem 440 can include many of the same components as HVAC system 100, as described with reference to FIGS. 1-3. For example, HVAC subsystem 440 can include a chiller, a boiler, any number of air handling units, economizers, field controllers, supervisory controllers, actuators, temperature sensors, and other devices for controlling the temperature, humidity, airflow, or other variable conditions within building 10. Lighting subsystem 442 can include any number of light fixtures, ballasts, lighting sensors, dimmers, or other devices configured to controllably adjust the amount of light provided to a building space. Security subsystem 438 can include occupancy sensors, video surveillance cameras, digital video recorders, video processing servers, intrusion detection devices, access control devices and servers, or other security-related devices.

Still referring to FIG. 4, BMS controller 366 is shown to include a communications interface 407 and a BMS interface 409. Interface 407 may facilitate communications between BMS controller 366 and external applications (e.g., monitoring and reporting applications 422, enterprise control applications 426, remote systems and applications 444, applications residing on client devices 448, etc.) for allowing user control, monitoring, and adjustment to BMS controller 366 and/or subsystems 428. Interface 407 may also facilitate communications between BMS controller 366 and client devices 448. BMS interface 409 may facilitate communications between BMS controller 366 and building subsystems 428 (e.g., HVAC, lighting security, lifts, power distribution, business, etc.).

Interfaces 407, 409 can be or include wired or wireless communications interfaces (e.g., jacks, antennas, transmitters, receivers, transceivers, wire terminals, etc.) for conducting data communications with building subsystems 428 or other external systems or devices. In various embodiments, communications via interfaces 407, 409 can be direct (e.g., local wired or wireless communications) or via a communications network 446 (e.g., a WAN, the Internet, a cellular network, etc.). For example, interfaces 407, 409 can include an Ethernet card and port for sending and receiving data via an Ethernet-based communications link or network. In another example, interfaces 407, 409 can include a Wi-Fi transceiver for communicating via a wireless communications network. In another example, one or both of interfaces 407, 409 can include cellular or mobile phone communications transceivers. In one embodiment, communications interface 407 is a power line communications interface and BMS interface 409 is an Ethernet interface. In other embodiments, both communications interface 407 and BMS interface 409 are Ethernet interfaces or are the same Ethernet interface.

Still referring to FIG. 4, BMS controller 366 is shown to include a processing circuit 404 including a processor 406 and memory 408. Processing circuit 404 can be communicably connected to BMS interface 409 and/or communications interface 407 such that processing circuit 404 and the various components thereof can send and receive data via interfaces 407, 409. Processor 406 can be implemented as a general purpose processor, an application specific integrated circuit (ASIC), one or more field programmable gate arrays (FPGAs), a group of processing components, or other suitable electronic processing components.

Memory 408 (e.g., memory, memory unit, storage device, etc.) can include one or more devices (e.g., RAM, ROM, Flash memory, hard disk storage, etc.) for storing data and/or computer code for completing or facilitating the various processes, layers and modules described in the present application. Memory 408 can be or include volatile memory or non-volatile memory. Memory 408 can include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described in the present application. According to some embodiments, memory 408 is communicably connected to processor 406 via processing circuit 404 and includes computer code for executing (e.g., by processing circuit 404 and/or processor 406) one or more processes described herein.

In some embodiments, BMS controller 366 is implemented within a single computer (e.g., one server, one housing, etc.). In various other embodiments BMS controller 366 can be distributed across multiple servers or computers (e.g., that can exist in distributed locations). Further, while FIG. 4 shows applications 422 and 426 as existing outside of BMS controller 366, in some embodiments, applications 422 and 426 can be hosted within BMS controller 366 (e.g., within memory 408).

Still referring to FIG. 4, memory 408 is shown to include an enterprise integration layer 410, an automated measurement and validation (AM&V) layer 412, a demand response (DR) layer 414, a fault detection and diagnostics (FDD) layer 416, an integrated control layer 418, and a building subsystem integration later 420. Layers 410-420 can be configured to receive inputs from building subsystems 428 and other data sources, determine optimal control actions for building subsystems 428 based on the inputs, generate control signals based on the optimal control actions, and provide the generated control signals to building subsystems 428. The following paragraphs describe some of the general functions performed by each of layers 410-420 in BMS 400.

Enterprise integration layer 410 can be configured to serve clients or local applications with information and services to support a variety of enterprise-level applications. For example, enterprise control applications 426 can be configured to provide subsystem-spanning control to a graphical user interface (GUI) or to any number of enterprise-level business applications (e.g., accounting systems, user identification systems, etc.). Enterprise control applications 426 may also or alternatively be configured to provide configuration GUIs for configuring BMS controller 366. In yet other embodiments, enterprise control applications 426 can work with layers 410-420 to optimize building performance (e.g., efficiency, energy use, comfort, or safety) based on inputs received at interface 407 and/or BMS interface 409.

Building subsystem integration layer 420 can be configured to manage communications between BMS controller 366 and building subsystems 428. For example, building subsystem integration layer 420 may receive sensor data and input signals from building subsystems 428 and provide output data and control signals to building subsystems 428. Building subsystem integration layer 420 may also be configured to manage communications between building subsystems 428. Building subsystem integration layer 420 translate communications (e.g., sensor data, input signals, output signals, etc.) across a plurality of multi-vendor/multi-protocol systems.

Demand response layer 414 can be configured to optimize resource usage (e.g., electricity use, natural gas use, water use, etc.) and/or the monetary cost of such resource usage in response to satisfy the demand of building 10. The optimization can be based on time-of-use prices, curtailment signals, energy availability, or other data received from utility providers, distributed energy generation systems 424, from energy storage 427 (e.g., hot TES 242, cold TES 244, etc.), or from other sources. Demand response layer 414 may receive inputs from other layers of BMS controller 366 (e.g., building subsystem integration layer 420, integrated control layer 418, etc.). The inputs received from other layers can include environmental or sensor inputs such as temperature, carbon dioxide levels, relative humidity levels, air quality sensor outputs, occupancy sensor outputs, room schedules, and the like. The inputs may also include inputs such as electrical use (e.g., expressed in kWh), thermal load measurements, pricing information, projected pricing, smoothed pricing, curtailment signals from utilities, and the like.

According to some embodiments, demand response layer 414 includes control logic for responding to the data and signals it receives. These responses can include communicating with the control algorithms in integrated control layer 418, changing control strategies, changing setpoints, or activating/deactivating building equipment or subsystems in a controlled manner. Demand response layer 414 may also include control logic configured to determine when to utilize stored energy. For example, demand response layer 414 may determine to begin using energy from energy storage 427 just prior to the beginning of a peak use hour.

In some embodiments, demand response layer 414 includes a control module configured to actively initiate control actions (e.g., automatically changing setpoints) which minimize energy costs based on one or more inputs representative of or based on demand (e.g., price, a curtailment signal, a demand level, etc.). In some embodiments, demand response layer 414 uses equipment models to determine an optimal set of control actions. The equipment models can include, for example, thermodynamic models describing the inputs, outputs, and/or functions performed by various sets of building equipment. Equipment models may represent collections of building equipment (e.g., subplants, chiller arrays, etc.) or individual devices (e.g., individual chillers, heaters, pumps, etc.).

Demand response layer 414 may further include or draw upon one or more demand response policy definitions (e.g., databases, XML files, etc.). The policy definitions can be edited or adjusted by a user (e.g., via a graphical user interface) so that the control actions initiated in response to demand inputs can be tailored for the user's application, desired comfort level, particular building equipment, or based on other concerns. For example, the demand response policy definitions can specify which equipment can be turned on or off in response to particular demand inputs, how long a system or piece of equipment should be turned off, what setpoints can be changed, what the allowable set point adjustment range is, how long to hold a high demand setpoint before returning to a normally scheduled setpoint, how close to approach capacity limits, which equipment modes to utilize, the energy transfer rates (e.g., the maximum rate, an alarm rate, other rate boundary information, etc.) into and out of energy storage devices (e.g., thermal storage tanks, battery banks, etc.), and when to dispatch on-site generation of energy (e.g., via fuel cells, a motor generator set, etc.).

Integrated control layer 418 can be configured to use the data input or output of building subsystem integration layer 420 and/or demand response later 414 to make control decisions. Due to the subsystem integration provided by building subsystem integration layer 420, integrated control layer 418 can integrate control activities of the subsystems 428 such that the subsystems 428 behave as a single integrated supersystem. In some embodiments, integrated control layer 418 includes control logic that uses inputs and outputs from a plurality of building subsystems to provide greater comfort and energy savings relative to the comfort and energy savings that separate subsystems could provide alone. For example, integrated control layer 418 can be configured to use an input from a first subsystem to make an energy-saving control decision for a second subsystem. Results of these decisions can be communicated back to building subsystem integration layer 420.

Integrated control layer 418 is shown to be logically below demand response layer 414. Integrated control layer 418 can be configured to enhance the effectiveness of demand response layer 414 by enabling building subsystems 428 and their respective control loops to be controlled in coordination with demand response layer 414. This configuration may advantageously reduce disruptive demand response behavior relative to conventional systems. For example, integrated control layer 418 can be configured to assure that a demand response-driven upward adjustment to the setpoint for chilled water temperature (or another component that directly or indirectly affects temperature) does not result in an increase in fan energy (or other energy used to cool a space) that would result in greater total building energy use than was saved at the chiller.

Integrated control layer 418 can be configured to provide feedback to demand response layer 414 so that demand response layer 414 checks that constraints (e.g., temperature, lighting levels, etc.) are properly maintained even while demanded load shedding is in progress. The constraints may also include setpoint or sensed boundaries relating to safety, equipment operating limits and performance, comfort, fire codes, electrical codes, energy codes, and the like. Integrated control layer 418 is also logically below fault detection and diagnostics layer 416 and automated measurement and validation layer 412. Integrated control layer 418 can be configured to provide calculated inputs (e.g., aggregations) to these higher levels based on outputs from more than one building subsystem.

Automated measurement and validation (AM&V) layer 412 can be configured to verify that control strategies commanded by integrated control layer 418 or demand response layer 414 are working properly (e.g., using data aggregated by AM&V layer 412, integrated control layer 418, building subsystem integration layer 420, FDD layer 416, or otherwise). The calculations made by AM&V layer 412 can be based on building system energy models and/or equipment models for individual BMS devices or subsystems. For example, AM&V layer 412 may compare a model-predicted output with an actual output from building subsystems 428 to determine an accuracy of the model.

Fault detection and diagnostics (FDD) layer 416 can be configured to provide on-going fault detection for building subsystems 428, building subsystem devices (i.e., building equipment), and control algorithms used by demand response layer 414 and integrated control layer 418. FDD layer 416 may receive data inputs from integrated control layer 418, directly from one or more building subsystems or devices, or from another data source. FDD layer 416 may automatically diagnose and respond to detected faults. The responses to detected or diagnosed faults can include providing an alert message to a user, a maintenance scheduling system, or a control algorithm configured to attempt to repair the fault or to work-around the fault.

FDD layer 416 can be configured to output a specific identification of the faulty component or cause of the fault (e.g., loose damper linkage) using detailed subsystem inputs available at building subsystem integration layer 420. In other exemplary embodiments, FDD layer 416 is configured to provide “fault” events to integrated control layer 418 which executes control strategies and policies in response to the received fault events. According to some embodiments, FDD layer 416 (or a policy executed by an integrated control engine or business rules engine) may shut-down systems or direct control activities around faulty devices or systems to reduce energy waste, extend equipment life, or assure proper control response.

FDD layer 416 can be configured to store or access a variety of different system data stores (or data points for live data). FDD layer 416 may use some content of the data stores to identify faults at the equipment level (e.g., specific chiller, specific AHU, specific terminal unit, etc.) and other content to identify faults at component or subsystem levels. For example, building subsystems 428 may generate temporal (i.e., time-series) data indicating the performance of BMS 400 and the various components thereof. The data generated by building subsystems 428 can include measured or calculated values that exhibit statistical characteristics and provide information about how the corresponding system or process (e.g., a temperature control process, a flow control process, etc.) is performing in terms of error from its setpoint. These processes can be examined by FDD layer 416 to expose when the system begins to degrade in performance and alert a user to repair the fault before it becomes more severe.

Referring now to FIG. 5, a block diagram of another building management system (BMS) 500 is shown, according to some embodiments. BMS 500 can be used to monitor and control the devices of HVAC system 100, waterside system 200, airside system 300, building subsystems 428, as well as other types of BMS devices (e.g., lighting equipment, security equipment, etc.) and/or HVAC equipment.

BMS 500 provides a system architecture that facilitates automatic equipment discovery and equipment model distribution. Equipment discovery can occur on multiple levels of BMS 500 across multiple different communications busses (e.g., a system bus 554, zone buses 556-560 and 564, sensor/actuator bus 566, etc.) and across multiple different communications protocols. In some embodiments, equipment discovery is accomplished using active node tables, which provide status information for devices connected to each communications bus. For example, each communications bus can be monitored for new devices by monitoring the corresponding active node table for new nodes. When a new device is detected, BMS 500 can begin interacting with the new device (e.g., sending control signals, using data from the device) without user interaction.

Some devices in BMS 500 present themselves to the network using equipment models. An equipment model defines equipment object attributes, view definitions, schedules, trends, and the associated BACnet value objects (e.g., analog value, binary value, multistate value, etc.) that are used for integration with other systems. Some devices in BMS 500 store their own equipment models. Other devices in BMS 500 have equipment models stored externally (e.g., within other devices). For example, a zone coordinator 508 can store the equipment model for a bypass damper 528. In some embodiments, zone coordinator 508 automatically creates the equipment model for bypass damper 528 or other devices on zone bus 558. Other zone coordinators can also create equipment models for devices connected to their zone busses. The equipment model for a device can be created automatically based on the types of data points exposed by the device on the zone bus, device type, and/or other device attributes. Several examples of automatic equipment discovery and equipment model distribution are discussed in greater detail below.

Still referring to FIG. 5, BMS 500 is shown to include a system manager 502; several zone coordinators 506, 508, 510 and 518; and several zone controllers 524, 530, 532, 536, 548, and 550. System manager 502 can monitor data points in BMS 500 and report monitored variables to various monitoring and/or control applications. System manager 502 can communicate with client devices 504 (e.g., user devices, desktop computers, laptop computers, mobile devices, etc.) via a data communications link 574 (e.g., BACnet IP, Ethernet, wired or wireless communications, etc.). System manager 502 can provide a user interface to client devices 504 via data communications link 574. The user interface may allow users to monitor and/or control BMS 500 via client devices 504.

In some embodiments, system manager 502 is connected with zone coordinators 506-510 and 518 via a system bus 554. System manager 502 can be configured to communicate with zone coordinators 506-510 and 518 via system bus 554 using a master-slave token passing (MSTP) protocol or any other communications protocol. System bus 554 can also connect system manager 502 with other devices such as a constant volume (CV) rooftop unit (RTU) 512, an input/output module (TOM) 514, a thermostat controller 516 (e.g., a TEC5000 series thermostat controller), and a network automation engine (NAE) or third-party controller 520. RTU 512 can be configured to communicate directly with system manager 502 and can be connected directly to system bus 554. Other RTUs can communicate with system manager 502 via an intermediate device. For example, a wired input 562 can connect a third-party RTU 542 to thermostat controller 516, which connects to system bus 554.

System manager 502 can provide a user interface for any device containing an equipment model. Devices such as zone coordinators 506-510 and 518 and thermostat controller 516 can provide their equipment models to system manager 502 via system bus 554. In some embodiments, system manager 502 automatically creates equipment models for connected devices that do not contain an equipment model (e.g., IOM 514, third party controller 520, etc.). For example, system manager 502 can create an equipment model for any device that responds to a device tree request. The equipment models created by system manager 502 can be stored within system manager 502. System manager 502 can then provide a user interface for devices that do not contain their own equipment models using the equipment models created by system manager 502. In some embodiments, system manager 502 stores a view definition for each type of equipment connected via system bus 554 and uses the stored view definition to generate a user interface for the equipment.

Each zone coordinator 506-510 and 518 can be connected with one or more of zone controllers 524, 530-532, 536, and 548-550 via zone buses 556, 558, 560, and 564. Zone coordinators 506-510 and 518 can communicate with zone controllers 524, 530-532, 536, and 548-550 via zone busses 556-560 and 564 using a MSTP protocol or any other communications protocol. Zone busses 556-560 and 564 can also connect zone coordinators 506-510 and 518 with other types of devices such as variable air volume (VAV) RTUs 522 and 540, changeover bypass (COBP) RTUs 526 and 552, bypass dampers 528 and 546, and PEAK controllers 534 and 544.

Zone coordinators 506-510 and 518 can be configured to monitor and command various zoning systems. In some embodiments, each zone coordinator 506-510 and 518 monitors and commands a separate zoning system and is connected to the zoning system via a separate zone bus. For example, zone coordinator 506 can be connected to VAV RTU 522 and zone controller 524 via zone bus 556. Zone coordinator 508 can be connected to COBP RTU 526, bypass damper 528, COBP zone controller 530, and VAV zone controller 532 via zone bus 558. Zone coordinator 510 can be connected to PEAK controller 534 and VAV zone controller 536 via zone bus 560. Zone coordinator 518 can be connected to PEAK controller 544, bypass damper 546, COBP zone controller 548, and VAV zone controller 550 via zone bus 564.

A single model of zone coordinator 506-510 and 518 can be configured to handle multiple different types of zoning systems (e.g., a VAV zoning system, a COBP zoning system, etc.). Each zoning system can include a RTU, one or more zone controllers, and/or a bypass damper. For example, zone coordinators 506 and 510 are shown as Verasys VAV engines (VVEs) connected to VAV RTUs 522 and 540, respectively. Zone coordinator 506 is connected directly to VAV RTU 522 via zone bus 556, whereas zone coordinator 510 is connected to a third-party VAV RTU 540 via a wired input 568 provided to PEAK controller 534. Zone coordinators 508 and 518 are shown as Verasys COBP engines (VCEs) connected to COBP RTUs 526 and 552, respectively. Zone coordinator 508 is connected directly to COBP RTU 526 via zone bus 558, whereas zone coordinator 518 is connected to a third-party COBP RTU 552 via a wired input 570 provided to PEAK controller 544.

Zone controllers 524, 530-532, 536, and 548-550 can communicate with individual BMS devices (e.g., sensors, actuators, etc.) via sensor/actuator (SA) busses. For example, VAV zone controller 536 is shown connected to networked sensors 538 via SA bus 566. Zone controller 536 can communicate with networked sensors 538 using a MSTP protocol or any other communications protocol. Although only one SA bus 566 is shown in FIG. 5, it should be understood that each zone controller 524, 530-532, 536, and 548-550 can be connected to a different SA bus. Each SA bus can connect a zone controller with various sensors (e.g., temperature sensors, humidity sensors, pressure sensors, light sensors, occupancy sensors, etc.), actuators (e.g., damper actuators, valve actuators, etc.) and/or other types of controllable equipment (e.g., chillers, heaters, fans, pumps, etc.).

Each zone controller 524, 530-532, 536, and 548-550 can be configured to monitor and control a different building zone. Zone controllers 524, 530-532, 536, and 548-550 can use the inputs and outputs provided via their SA busses to monitor and control various building zones. For example, a zone controller 536 can use a temperature input received from networked sensors 538 via SA bus 566 (e.g., a measured temperature of a building zone) as feedback in a temperature control algorithm. Zone controllers 524, 530-532, 536, and 548-550 can use various types of control algorithms (e.g., state-based algorithms, extremum seeking control (ESC) algorithms, proportional-integral (PI) control algorithms, proportional-integral-derivative (PID) control algorithms, model predictive control (MPC) algorithms, feedback control algorithms, etc.) to control a variable state or condition (e.g., temperature, humidity, airflow, lighting, etc.) in or around building 10.

Variable Refrigerant Flow System

Referring now to FIGS. 6A-6B, a variable refrigerant flow (VRF) system 600 is shown, according to some embodiments. VRF system 600 is shown to include multiple outdoor VRF units 602 and multiple indoor VRF units 604. Outdoor VRF units 602 can be located outside a building and can operate to heat or cool a refrigerant. Outdoor VRF units 602 can consume electricity to convert refrigerant between liquid, gas, and/or super-heated gas phases. Indoor VRF units 604 can be distributed throughout various building zones within a building and can receive the heated or cooled refrigerant from outdoor VRF units 602. Each indoor VRF unit 604 can provide temperature control for the particular building zone in which the indoor VRF unit is located.

A primary advantage of VRF systems is that some indoor VRF units 604 can operate in a cooling mode while other indoor VRF units 604 operate in a heating mode. For example, each of outdoor VRF units 602 and indoor VRF units 604 can operate in a heating mode, a cooling mode, or an off mode. Each building zone can be controlled independently and can have different temperature setpoints. In some embodiments, each building has up to three outdoor VRF units 602 located outside the building (e.g., on a rooftop) and up to 128 indoor VRF units 604 distributed throughout the building (e.g., in various building zones).

Many different configurations exist for VRF system 600. In some embodiments, VRF system 600 is a two-pipe system in which each outdoor VRF unit 602 connects to a single refrigerant return line and a single refrigerant outlet line. In a two-pipe system, all of the outdoor VRF units 602 operate in the same mode since only one of a heated or chilled refrigerant can be provided via the single refrigerant outlet line. In other embodiments, VRF system 600 is a three-pipe system in which each outdoor VRF unit 602 connects to a refrigerant return line, a hot refrigerant outlet line, and a cold refrigerant outlet line. In a three-pipe system, both heating and cooling can be provided simultaneously via dual refrigerant outlet lines.

VRF system 600 can represent an example of a VRF system that may utilize AI to predict oil level and viscosity to ensure components (e.g., outdoor VRF units 602, indoor VRF units 604, etc.) are operating correctly. Specifically, the components may require oil for adequate lubrication to ensure that rapid degradation of the components are avoided. Specifically, VRF system 600 may leverage the systems and methods described throughout FIGS. 7-10 to ensure that all components have an adequate supply of oil, that the oil provided is of an appropriate viscosity, etc.

Referring now to FIG. 7A, an illustration of a VRF system 700 is shown, according to some embodiments. In some embodiments, VRF system 700 is similar to and/or the same as VRF system 600 as described with reference to FIGS. 6A and 6B. More particularly, VRF system 700 can illustrate how oil is utilized in a VRF system. It should be noted that VRF system 700 is shown purely for sake of example of how a VRF system may operate. Components, relationships, and/or other features of VRF system 700 can be customized and configured based on particular implementations. For example, VRF system 700 may include more or fewer compressors 701 than as shown in FIG. 7A.

VRF system 700 is shown to include compressors 701, a heat exchanger 702, a double tube type heat exchanger 703, oil separators 704, and an accumulator 705. To operate properly, compressors 701 may require oil to ensure components of compressors 701 are properly lubricated. Without oil, the components may rapidly degrade and compressors 701 may fail to provide adequate cooling/heating to a zone. Heat exchangers 702 and 703 can transfer heat between fluids (e.g., oil and a refrigerant). Oil separators 704 can separate oil from refrigerant and/or other fluids in VRF system 700. Specifically, during operation, compressors 701 may leak and/or otherwise allow some oil to get mixed into refrigerant outputted by compressors 701. If said oil is not extracted back out of the oil/refrigerant mixture, the oil may be provided to components beyond VRF system 700 (e.g., indoor AHUs) which can result in a rapid loss of oil in VRF system 700. Accordingly, oil separators 704 can distill the oil from the fluid mixtures and provide the oil to accumulator 705 for temporary storage. Accumulator 705 can store the oil and can be accessed as needed to retrieve oil for other components of VRF system 700.

VRF system 700 is also shown to include strainers 706, a distributor 707, reversing valves 708, capillary tubes 709, and micro-computer control expansion valves 710. Strainers 706 can remove impurities (e.g., dirt, debris, etc.) from the oil and/or the oil/refrigerant mixture that may be accidentally integrated with the oil and/or oil/refrigerant mixture during operation of VRF system 700. Impurities can result in poor functioning of building equipment and may affect characteristics of the oil in VRF system 700 (e.g., by increasing or decreasing a viscosity of the oil). Distributor 707 can help in distributing fluids throughout heat exchanger 702. Reversing valves 708 can change a direction of refrigerant flow in VRF system 700 to switch VRF system 700 between heating and cooling modes. Capillary tubes 709 can assist in reducing a temperature of refrigerant in VRF system 700 by affecting a pressure of the refrigerant. Micro-computer control expansion valves 710 can regulate an amount of refrigerant entering components of VRF system 700.

VRF system 700 is also shown to include check valves 711, solenoid valves 712, check joints 713, a stop valve 714 for the liquid line, a stop valve 715 for the gas (low) line, a stop valve 716 for the gas (high/low) line, a refrigerant pressure sensor 717, another refrigerant pressure sensor 718, and high pressure switches 719. Check valves 711 can help ensure that fluid is flowing in a correct direction within VRF system 700 by restricting the fluid from flowing in a direction opposite a desired direction of flow. Solenoid valves 712 can regulate a flow of fluids in VRF system 700. Check joints 713 can help regulate stress on components of VRF system 700. Stop valves 714, 715, and 716 can restrict a flow of fluid in the liquid line, the gas (low) line, and the gas (high/low) line shown in the illustration of FIG. 7A, respectively. With regard to refrigerant pressure sensors 717 and 718, refrigerant pressure sensor 717 may be a high pressure sensor whereas refrigerant pressure sensor 718 may be a low pressure sensor within VRF system 700. As refrigerant returns to compressors 701, high pressure switches 719 can stop the refrigerant from entering compressors 701 if a pressure of the refrigerant is too high or too low in order to prevent damage to compressors 701.

VRF system 700 is also shown to include a variety of thermistors. In VRF system 700, a resistance across a thermistor can be primarily based on a temperature of a connected component. In the illustration of FIG. 7A, VRF system 700 is shown to include thermistors 720-725. Thermistor 720 is associated with an upper side of first compressor 701. Thermistor 721 is associated with an upper side of second compressor 701. Thermistor 722 is associated with a gas side of heat exchanger 702. Thermistor 723 is associated with a liquid side of heat exchanger 702. Thermistor 724 is associated with a subcooler bypass side. Thermistor 725 is associated with an auto charge of refrigerant.

Each pipe in VRF system 700 is also labeled with a corresponding outer diameter OD and a thickness T which are given below in Table A. As should be noted, the material used in VRF system 700 across all piping is C1220T-O.

TABLE A Outer Diameter and Thickness of Piping Mark OD × T Material a 1 3/32 × 0.075 C1220T-O [28.0] × [1.9] b 1 3/32 × 0.063 [28.0] × [1.6] c 1 × 0.071 [25.4] × [1.8] d 1 × 0.047 [25.4] × [1.2] e ⅞ × 0.059 [22.0] × [1.5] f ⅞ × 0.047 [22.0] × [1.2] g ¾ × 0.065 [19.05] × [1.65] h ⅝ × 0.047 [15.88] × [1.2] i ½ × 0.039 [12.7] × [1.0] j ⅜ × 0.031 [9.52] × [0.8] k ¼ × 0.042 [6.35] × [1.07] l ¼ × 0.028 [6.35] × [0.7]

Referring now to FIG. 7B, an illustration of a VRF system 750 is shown, according to some embodiments. In some embodiments, VRF system 750 is similar to and/or the same as VRF system 700 as described with reference to FIG. 7 and/or VRF system 600 as described with reference to FIGS. 6A and 6B. Specifically, VRF system 750 illustrates a flow of oil throughout a VRF system. As with VRF system 700, VRF system 750 is provided purely for sake of example. The components, structure, and/or other characteristics of VRF system 750 can be customized and configured dependent on implementation.

VRF system 750 is shown to include a suction line 752. Suction line 752 can provide refrigerant (e.g., refrigerant vapor) used by one or more devices/systems (e.g., an indoor AHU) to compressors 754. In some embodiments, some oil may be included in fluid provided to compressors 754 by suction line 752. Based on the received refrigerant, compressors 754 can operate to compress the refrigerant into a higher pressure gas and output said higher pressure gas via a discharge line 756. As compressors 754 may require oil to function properly, the compression process performed by compressors 754 may result in some oil getting mixed into the outputted high pressure gas, thereby resulting in an oil/refrigerant mixture.

VRF system 750 is also shown to include an oil separator 758. Based on the received oil/refrigerant mixture, oil separator 758 can operate to separate the oil from the refrigerant. After separation, the refrigerant can be provided to some device/system (e.g., an indoor AHU) via a refrigerant line 762. The separated oil can be provided to an accumulator 760 via an oil line 764. Accumulator 760 can function as a storage container for oil separated by oil separator 758. Accumulator 760 may have some maximum capacity that defines a maximum amount of oil that can be stored in accumulator 760.

Oil stored by accumulator 760 can be provided back to compressors 754 via an oil return line 766. Given that accumulator 760 has some non-zero amount of stored oil, accumulator 760 can provide the oil to any of compressors 754 if a particular compressor 754 requires more oil. In some embodiments, VRF system 750 includes valves 768 that regulate a flow of oil to compressors 754. In this case, valves 768 may prevent too much oil from being provided to compressors 754 and/or otherwise regulate oil being provided to compressors 754.

In some embodiments, a viscosity of the oil in VRF system 750 as well as oil levels in each of compressors 754 and accumulator 760 are estimated/predicted by an AI model. In this case, the AI model may take in inputs such as an operating speed of compressors 754, an ambient temperature near VRF system 750, a discharge temperature and discharge pressure of discharge line 756, a suction pressure of suction line 752, a temperature of the refrigerant vapor and/or a temperature of other gas within VRF system 750, etc. Said inputs may be measured by sensors (e.g., temperature sensors, pressure sensors, etc.) throughout VRF system 750 and/or may be directly provided by components of VRF system 750 (e.g., compressors 754 may directly output their operating speeds).

Based on the inputs, the AI model may predict characteristics of the oil (e.g., oil levels, oil viscosity, etc.) in VRF system 750. If the characteristics of the oil do not meet predefined thresholds (or some other constraitn), one or more corrective action may be initiated to address the failure of the oil characteristics to meet the predefined thresholds. For example, if a viscosity of the oil in VRF system 750 is too high or too low, a corrective action may include introducing new oil to VRF system 750 and/or completely replacing the oil within VRF system 750 at a specific time. As another example, if an oil level in a specific compressor 754 is too low, a corrective may be initiated to operate the specific compressor 754, a specific valve 768, and/or accumulator 760 such that the specific compressor 754 can obtain more oil from accumulator 760 at a specific time that results in a relatively low impact to efficiency of VRF system 750. The AI models and corrective actions that can be initiated are described in greater detail below with reference to FIGS. 8-10.

Systems and Methods for Oil Level and Viscosity Estimation

Referring generally to FIGS. 8-10, systems and methods for estimating and predicting oil characteristics in a VRF system are shown and described, according to some embodiments. It should be appreciated that the description below is described with reference to a VRF system for sake of example only and should not be regarded as limiting. The systems and methods described throughout FIGS. 8-10 can be similarly applied to a variety of systems that utilize oil (e.g., other building systems, vehicle systems, etc.) and are not meant to be limited to VRF systems.

The systems and methods described below can utilize artificial intelligence (AI) models to predict how characteristics of the oil change over time based on a variety of inputs. The AI models utilize can include any appropriate type of AI model. For example, the AI models may be or include long short-term memory (LSTM) models, other types of recurrent neural networks (RNNs), convolutional neural networks (CNNs), etc. A type of AI model to utilize can be selected based on, for example, accuracy of a given AI model, what specific inputs/outputs are of consideration, user preferences, etc. In some embodiments, RNN models such as LSTM models are preferred due to the time-series nature of oil. It should be noted that machine learning models may be referred to herein as synonymous with AI models.

The AI model can be trained to predict certain oil characteristics based on a set of training data. The training data may be provided by a variety of sources. For example, a user may provide a set of inputs including a variety variables that can help the AI model to determine how the oil is being utilized in the VRF system and a corresponding set of outputs based on actual measured operating states of the system or a similar system. In this case, the inputs may include, for example, the inputs may include an operating speed of a compressor, an ambient temperature near the compressors, a discharge temperature of the compressor, a suction pressure of the compressor, a discharge pressure of the compressor, a gas temperature of gas in the compressor, etc. As defined herein, a discharge temperature of a compressor can refer to a temperature measure of a superheated refrigerant vapor in the VRF system, a suction pressure can refer to an intake pressure generated by the compressor during operation, and a discharge pressure can refer to a pressure generated on an output side of the compressor. The outputs may include, for example, oil viscosity, an oil level of a compressor, an oil level of an accumulator, etc. The AI model can then be trained using the inputs and corresponding outputs to predict values of the outputs based on inputs. Of course, said inputs and outputs are given for sake of example and are not meant to be limiting on possible inputs to the AI model.

In some embodiments, a simulation model is utilized to generate the training data used to train the AI model. Training data generated using the simulation model may be used separately or in addition to training data gathered from other sources (e.g., from measured states of an actual system). The simulation model can be constructed to simulate changes in an oil return system over time based on a variety of conditions. In other words, the simulation model can be constructed to digitally mimic operation of an oil return system. States of the simulation model (e.g., oil level, oil viscosity, energy consumption, etc.) can be manipulated to generate training data representing a wide variety of conditions. The simulation model may be executed multiple times to generate training data representing evolution of the system over time under a variety of different loads, using different building devices, under different weather/environmental conditions, at different times, etc. In terms of a VRF system, the simulation model may be, for example, a closed loop functional mock-up unit (FMU) model that may typically be used to operate the VRF system if no AI model is used. Advantageously, by utilizing a simulation model, large amounts of training data can be generated in shorter periods of time as compared to operating an actual system over time to generate training data. Moreover, the simulation model can be executed to generate training data representing fringe scenarios (e.g., dangerously high loads, dangerous operating conditions, device faults, etc.) without subjecting an actual system to conditions that may be dangerous and may disrupt comfort of occupants in a real building.

Once trained based on a set of training data, the AI model can predict oil characteristics based on the inputs. For example, the AI model may predict a viscosity of the oil, an oil level in a compressor, and an oil level in an accumulator. Based on the predicted oil characteristics, a determination can be made whether the predicted oil characteristics adhere to predefined thresholds (and/or other constraints) on values of the characteristics. The predefined thresholds can include any limitations on the values such as, for example, threshold values the values of the characteristics must be above or below, ranges of acceptable values of the characteristics, etc. If the values of the oil characteristics meet the predefined thresholds, the VRF system can continue standard operation. However, if the values of the oil characteristics do not meet the predefined thresholds, a corrective action can be generated and initiated. Corrective actions, as defined herein, can refer to any action taken to address one or more oil characteristics not meeting some predefined constraint(s)/threshold(s). For example, corrective actions may be or include generating and transmitting a notification to a user device, scheduling a technician to perform maintenance on the VRF system and/or to replace the oil, generating control signals and operating building equipment (e.g., VRF devices) based on the control signals, disabling certain building devices, automatically injecting new oil into the VRF system, logging the threshold violation(s) to a database, etc. A corrective action to initiate can be determined based on a variety of factors such as what threshold is violated, an amount the threshold was violated (e.g., a difference between an actual value of the oil characteristic and a threshold value), user preferences, and/or any other applicable consideration. Violation of a threshold on oil characteristics can also indicate a deficiency of the oil in that the oil is deficient of some desired property (e.g., a desired viscosity). As described herein, a violation of a threshold can refer to when a value (e.g., a predicted value) is above the threshold in the case of a maximum value threshold, or is below the threshold in the case of a minimum value threshold.

Referring now to FIG. 8, a block diagram of an oil management controller 800 for predicting characteristics of oil is shown, according to some embodiments. In particular, oil management controller 800 may predict characteristics of oil in a VRF system. However, oil management controller 800 can be applied to a variety of other systems/devices (e.g., other HVAC systems, car systems, etc.) that require oil to properly operate. In some embodiments, oil management controller 800 and/or components therein are incorporated in BMS controller 366 as described with reference to FIGS. 3-4 and/or another controller. In some embodiments, oil management controller 800 is used to operate some and/or all of the VRF systems described throughout FIGS. 7A-7B.

Oil management controller 800 is shown to include a communications interface 808 and a processing circuit 802. Communications interface 808 may include wired or wireless interfaces (e.g., jacks, antennas, transmitters, receivers, transceivers, wire terminals, etc.) for conducting data communications with various systems, devices, or networks. For example, communications interface 808 may include an Ethernet card and port for sending and receiving data via an Ethernet-based communications network and/or a Wi-Fi transceiver for communicating via a wireless communications network. Communications interface 808 may be configured to communicate via local area networks or wide area networks (e.g., the Internet, a building WAN, etc.) and may use a variety of communications protocols (e.g., BACnet, IP, LON, etc.).

Communications interface 808 may be a network interface configured to facilitate electronic data communications between oil management controller 800 and various external systems or devices (e.g., equipment 822, sensors 820, a user device 824, etc.). For example, oil management controller 800 may receive equipment feedback from equipment 822 via communications interface 808.

Processing circuit 802 is shown to include a processor 804 and memory 806. Processor 804 may be a general purpose or specific purpose processor, an application specific integrated circuit (ASIC), one or more field programmable gate arrays (FPGAs), a group of processing components, or other suitable processing components. Processor 804 may be configured to execute computer code or instructions stored in memory 806 or received from other computer readable media (e.g., CDROM, network storage, a remote server, etc.).

Memory 806 may include one or more devices (e.g., memory units, memory devices, storage devices, etc.) for storing data and/or computer code for completing and/or facilitating the various processes described in the present disclosure. Memory 806 may include random access memory (RAM), read-only memory (ROM), hard drive storage, temporary storage, non-volatile memory, flash memory, optical memory, or any other suitable memory for storing software objects and/or computer instructions. Memory 806 may include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described in the present disclosure. Memory 806 may be communicably connected to processor 804 via processing circuit 802 and may include computer code for executing (e.g., by processor 804) one or more processes described herein. In some embodiments, one or more components of memory 806 are part of a singular component. However, each component of memory 806 is shown independently for ease of explanation.

Memory 806 is shown to include a training data collector 810. Training data collector 810 can collect training data used to train an artificial intelligence model from one or more training data sources 818. Specifically, training data collector 810 can obtain training data associated with characteristics of oil in a VRF system. In some embodiments, training data collector 810 transmits queries to training data sources 818 to obtain the training data. In some embodiments, training data collector 810 may passively receive training data from training data sources 818 without needing to actively request the training data.

Training data sources 818 can include any source of data that can store and/or provide training data to training data collector 810. For example, training data sources 818 may be or include a user device (e.g., a laptop, a desktop computer, a mobile device, a tablet, etc.) that can provide a stored training data set to training data collector 810. As another example, training data sources 818 may be or include a database (e.g., a cloud database) that stores data associated with utilizing a standard VRF plant model with additional outputs of oil level, oil viscosity, etc. In said example, the VRF plant model may be the standard model used to operate the VRF system. In this way, the training data can include measurements from the VRF system in actual operation along with measurements of the oil characteristics.

In some embodiments, training data collector 810 utilizes a simulation model to generate some or all of the training data used by model generator 812 to generate an AI model. The simulation model can model how an actual system may operate under various conditions (e.g., weather conditions, heating/cooling loads, device limitations, etc.) and how oil in the system may be consumed and/or otherwise change over time. In this way, training data collector 810 may not need to retrieve training data from training data sources 818 and instead can generate the training data within oil management controller 800. In some embodiments, the simulation model is hosted by a third party controller/device/system (e.g., a cloud computing system) which can provide the training data generated as a result of running the simulation model to oil management controller 800. In any case, the simulation model can be used/executed to generate a variety of training data representing various operating conditions of a system utilizing oil in shorter periods of time as compared to waiting for an actual system to generate training data through operation. Moreover, the simulation model can be executed to generate training data illustrative of fringe scenarios that may be dangerous for an actual system to operate under purely for the sake of generating training data.

Based on the obtained training data, training data collector 810 can combine the collected training data into a training data set and provide the training data set to a model generator 812. Based on the training data set, model generator 812 can generate an AI model that models oil characteristics over time. Specifically, model generator 812 can train the AI model to predict oil characteristics based on specified inputs. For example, model generator 812 may train the AI model to predict values of oil viscosity, oil level in one or more compressors, and oil level in an accumulator based on inputs of compressor speeds, an ambient temperature, a discharge temperature, a suction pressure, a discharge pressure, and a gas temperature.

The AI model generated by model generator 812 can be any of a variety of AI model structures. In some embodiments, the AI model is an RNN model such as an LSTM model. In this case, for the RNN to properly work on the VRF system, an original FMU plant model can be utilized to generate enough simulation data for RNN model to analyze and be trained based on. With training time increasing, the final RNN model may have much closer function as the original plant model. Some advantages of using RNN models in particular is that they may be faster and have a higher stability as compared with the FMU plant model. Further, the trained RNN model may reduce the influence the oil return and improve the efficiency of the VRF system. With particular regard to LSTM models, an LSTM model is an artificial RNN used for deep learning. LSTM models can classify and process entire sequences of time series data and can make predictions even with lags of unknown duration between important events in a time series. An LSTM model generated by model generator 812 may include various structures depending on implementation. For example, an LSTM model generated by model generator 812 may include one sequence input layer, one drop out layer, two fully connected layers, and two LSTM layers.

In some embodiments, the AI model is a CNN model. In this case, the CNN model may include, for example, an input layer, multiple hidden layers (e.g., rectified linear unit layers, pooling layers, fully connected layers, normalization layers, etc.), an output layer, etc. In some embodiments, the AI model follows some other artificial intelligence model architecture. Example architectures of the AI model are described in greater detail below with reference to FIGS. 9A and 9B.

Model generator 812 may utilize a variety of training techniques to generate the AI model. For example, model generator 812 may utilize a stochastic gradient descent with momentum approach, an adaptive moment estimation approach, a root mean square propagation approach, etc. With specific regard to the root mean square propagation approach, model generator 812 may utilize a root mean squared error (RMSE) to measure how accurate model predictions are to the training data provided by training data collector 810. Specifically, model generator 812 may monitor the RMSE over time based on the following equation:

RMSE=√{square root over ((Y_pred,t−Y_test,t)²)}

where Y_pred,tis a previous prediction of the AI model for a variable Y at a time step t, and Y_test,tis an actual value of the variable Y as indicated by the training data at time step t. The calculation of (Y_pred,t-Y_test,t)²can be performed for each time step t=1. . . n where n is a total number of predictions. Each difference can then be averaged together. During the training process, model generator 812 can refine the AI model to reduce the RMSE. Example experiments of training an AI model are described in detail below with reference to FIGS. 11A and 11B.

Model generator 812 can provide the generated AI model to a prediction generator 814. Prediction generator 814 can use the AI model to generate predictions of oil characteristics over time. In order to generate said predictions, prediction generator 814 can operate to obtain values of inputs required by the AI model from a variety of sources. For example, prediction generator 814 may obtain equipment feedback from equipment 822, measured variables from sensors 820, and/or any other appropriate source of input values.

Equipment 822 can be or include any devices that can provide values of inputs needed by the AI model. For example, in a VRF system, equipment 822 may include compressors, a heat exchanger, an accumulator, etc. More particularly, if the AI model requires a compressor speed as an input, equipment 822 may include one or more compressors that can provide an operating speed as equipment feedback to prediction generator 814.

Sensors 820 may be or include a variety of sensors that can measure values of inputs (i.e., variables) that are required by the AI model. For example, sensors 820 may include pressure sensors that measure a suction pressure and/or a discharge pressure. As another example, sensors 820 may include temperature sensors that measure an ambient temperature, a discharge temperature, and/or a gas temperature.

Based on the AI model and the obtained input values, prediction generator 814 can generate oil characteristic predictions by passing the obtained input values through the AI model. As a result of passing the obtained input values through the AI model, the AI model can output values of one or more oil characteristics (e.g., oil viscosity, oil levels in compressors, an oil level in an accumulator, etc.). In this way, characteristics of the oil in the VRF system can be estimated without the need for additional sensors to measure the oil characteristics.

In some embodiments, prediction generator 814 generates predictions regarding oil at multiple stages within a VRF system. For example, prediction generator 814 may generate a prediction of viscosity of an oil-refrigerant mixture outputted by compressors and may predict a separate viscosity of just the oil after the oil is separated from the oil-refrigerant mixture. In said example, predicting the viscosity of the oil-refrigerant mixture and the viscosity of just the oil may be beneficial in determining how the oil may benefit from certain corrective actions (e.g., an amount of oil to add/remove from the VRF system). By generating predictions regarding oil at multiple stages within the VRF system, oil deficiencies can be predicted and tracked over time throughout the VRF system rather than at a single point of the VRF system.

Prediction generator 814 can provide the predictions of the oil characteristics to a corrective action generator 816. Corrective action generator 816 can analyze the predicted oil characteristics to determine if any corrective actions should be initiated and what corrective actions to initiate. As defined above, a corrective action can refer to any action taken to address an oil characteristics not meeting some predefined threshold(s). Corrective actions may include, for example, distributing notifications/alerts to user device 824 to indicate to a user that certain oil characteristics are violating the predefined thresholds, operating equipment 822 in oil return control, disabling equipment 822, automatically scheduling a maintenance activity to be performed on equipment 822, logging the threshold violation in a database, etc. The predefined threshold(s) can be defined by a user, provided by a manufacturer, estimated based on operating states of equipment in the VRF system, etc. For example, a manufacturer may define a range of acceptable oil viscosity levels that compressors should be operated with. In this case, overly viscous oil and/or oil with low viscosity may result in more rapid deterioration of the compressors. As another example, a user may define a minimum oil level threshold defining a lowest acceptable amount of oil in an accumulator. As should be appreciated, thresholds defined for oil characteristics can be obtained from a variety of sources (e.g., manufacturers, users, based on predictions, etc.) and can include a variety of limitation types (e.g., ranges, threshold values, exact values to which characteristics should be equal, etc.).

As a more specific example, consider a scenario where the AI model predicts values of oil characteristics including an oil level in a compressor, an oil level in an accumulator, a viscosity of the oil, and/or a viscosity of an oil-refrigerant mixture flowing through a compressor. In the example, corrective action generator 816 may determine whether the oil levels in the compressor and the accumulator are above a first and second minimum threshold value, respectively. If the oil level in the compressor is low, corrective action generator 816 may determine the compressor should be operated in an oil return control to retrieve oil from an external source (e.g., an indoor VRF system) and/or from the accumulator. As defined herein, oil return control may indicate an operational mode in which the compressor runs at a high speed to bring oil back from an outdoor system. Oil return control can result in high energy consumption and may result in required heating/cooling being temporarily put on hold while the compressor operates to retrieve oil. If the oil level in the accumulator is below the second minimum threshold value, corrective action generator 816 may determine that more oil should be added to the VRF system as the oil level is too low. Addition of oil may result from operating the compressors in oil return control, may include a user manually adding new oil to the VRF system, etc.

Likewise, if corrective action generator 816 determines a viscosity of the oil or of the oil-refrigerant mixture is outside a predefined range, corrective action generator 816 may determine that oil should be added or removed from the VRF system. Specifically, if the viscosity of the oil-refrigerant mixture is above a maximum bound of the range, corrective action generator 816 may determine oil should be removed from the VRF system (i.e., reducing an amount of oil in the oil-refrigerant mixture) to lower the viscosity. If the viscosity of the oil-refrigerant mixture is below a minimum bound of the range, corrective action generator 816 may determine oil should be added to the VRF system (i.e., increasing an amount of oil in the oil-refrigerant mixture) to increase the viscosity. With regard to the oil itself, if the viscosity of the oil is above a maximum bound of the range, corrective action generator 816 may determine oil should be added to the VRF system to lower the viscosity if new oil is expected to be less viscous as compared to existing oil. If the viscosity of the oil is below the minimum bound, corrective action generator 816 may determine oil should be removed to increase the viscosity.

In some embodiments, corrective action generator 816 compares outputs of the AI model over time to determine if certain oil characteristics are approaching a threshold violation and will thereby include a deficiency. In this case, corrective action generator 816 may compare the values of the oil characteristics outputted by the AI model to previous outputted values of the oil characteristics. If a particular oil characteristic is trending towards violating a threshold, corrective action generator 816 may preemptively initiate a corrective action prior to the violation occurring. For example, if a viscosity of the oil is increasing over time and, based on a current trend, will exceed a maximum bound on viscosity within an upcoming time period, corrective action generator 816 may initiate a corrective action prior to the viscosity of the oil exceeding the maximum bound. Advantageously, preemptive initiation of corrective actions can ensure that an amount of time that equipment (e.g., compressors) is operated under conditions associated with violations of oil characteristic thresholds is reduced. Reducing said amount of time can likewise reduce degradation of the equipment, reduce energy consumption, and can provide other benefits.

In some embodiments, corrective action generator 816 predicts times to initiate certain corrective actions to reduce an impact on equipment 822 and/or other devices/systems. For example, corrective action generator 816 may predict a time to initiate oil return control in order to reduce a negative impact on heating/cooling loads required by a building. As described above, operating a compressor under oil return control can result in required heating/cooling being temporarily postponed for a duration of the oil return control. Accordingly, corrective action generator 816 can predict a time when the impact on required heating/cooling may be reduced (e.g., minimized). As another example, if new oil is required to be introduced to the system, corrective action generator 816 can predict a time to temporarily disable equipment 822 such that a new oil can be safely added to the system by an individual.

Corrective action generator 816 can utilize a variety of techniques to predict times at which to initiate certain corrective actions. For example, corrective action generator 816 may track certain variables over time and identify a lower range of values that may result in low amounts of disruption to a system. As a specific example, corrective action generator 816 may identify a range of values associated with low heating/cooling needs such that an impact on environmental conditions within a building will be reduced. Based on the identified range of values, corrective action generator 816 may track actual heating/cooling needs over time and, in response to identifying a time period where actual heating/cooling needs are within the identified range, can initiate a corrective action during said time period. In this way, corrective action generator 816 is effectively predicting a time period where initiating a corrective action results in a low overall impact.

In some embodiments, corrective action generator 816 can operate as a standard equipment controller in the case where no corrective actions are needed (e.g., if oil viscosity and oil levels are at appropriate values). In other words, corrective action generator 816 can generate control signals for equipment 822 in order to operate equipment 822 to affect some variable state or condition (e.g., temperature, humidity, etc.) within a building. In some embodiments, corrective action generator 816 may be configured to set boundary conditions for equipment 822 based on the predictions provided by prediction generator 814. For example, corrective action generator 816 may set a maximum speed of compressors based on a prediction of oil viscosity. In said example, if the viscosity is within an appropriate range and is not approaching a violation of the range, corrective action generator 816 may generate control signals to operate the compressors at a higher speed as the oil viscosity is appropriate.

As a result of initiating some corrective action, any violations of thresholds for oil characteristics can be addressed. This can ensure that an amount of time during which equipment 822 is operating under conditions associated with some oil characteristic violation is reduced (e.g., minimized). Overall, in a VRF system, initiating the corrective actions using predictions based on the AI model can save hardware cost, reduce an influence of oil return, and improve efficiency of the VRF system, among other benefits.

Referring now to FIG. 9A, an illustration of a recurrent neural network (RNN) structure 900 is shown, according to some embodiments. Specifically, RNN structure 900 can illustrate the structure of an RNN model (e.g., an LSTM model) that can be generated and utilized as the AI model described above with reference to FIG. 8.

RNNs are a class of artificial neural networks where connections between nodes form a directed graph along a temporal sequence. In the case of a VRF system, the RNN model represented by RNN structure 900 can be generated using simulation data collected based on an FMU plant model. As training time increases, the RNN model can have closer and closer function with the original plant model for the VRF system.

RNN structure 900 illustrates a condensed network structure and how the condensed network structure can be “unfolded” to illustrate how RNN structure 900 operates over a temporal sequence. Specifically, the condensed (“folded”) structure and the unfolded structure are equivalent, with the unfolded structure illustrating usage of the RNN model over a temporal sequence in greater detail.

RNN structure 900 is shown to include an input represented as x which may be a vector including inputs required by the RNN model. For a VRF system, the input vector x may include, for example, a compressor speed, an ambient temperature, a discharge temperature, a suction pressure, a discharge pressure, a gas temperature, etc. A weight vector U can be applied to x and a result provided to a hidden layer vector h. Similarly, a weight vector V can be applied to a hidden layer vector of a previous time step. Based on the weighted inputs and the weighted values of the previous hidden layer vector, a function can be applied to determine a corresponding output which, after a weight vector W is applied, can result in an output o. This process can be repeated for each time step of a temporal sequence. In other words, a new input vector x_tcan be obtained for a time step t and, based on x_t, a previous state h_t-1, and corresponding weight vectors U, V, and W, an output vector o_tcan be determined for time step t.

As a result of incorporating RNN structure 900 in the RNN model generated and used by oil management controller 800, predictions of the RNN model can be modified over time as a result of previous time steps. As oil characteristics change over time a result of changing conditions (e.g., changing environmental conditions, operating conditions, etc.), utilizing the RNN model in particular can be useful due to the unique ability of the RNN model to account for changes over a temporal sequence, as opposed to being limited by an original training process as some other neural network architectures are.

Referring now to FIG. 9B, an illustration of a neural network (NN) architecture 950 is shown, according to some embodiments. NN architecture 950 can describe a general architecture that may be utilized by the AI model described above with reference to FIG. 8 for a VRF system (e.g., VRF system 600). Specifically, NN architecture 950 can illustrate how a neural network can generate a set of outputs based on a set of inputs related to the VRF system. it should be noted, however, that NN architecture 950 is provided purely for sake of example of a neural network architecture that can be utilized and is not meant to be limiting on neural network architectures that can be utilized by the AI model described with reference to FIG. 8.

NN architecture 950 is shown to receive a compressor speed, an ambient temperature, a discharge temperature, a suction pressure, a discharge pressure, and a gas temperature as inputs. Each input can be associated with a particular input node of an input layer in NN architecture 950. In other words, a number of nodes in the input layer may correspond to a number of actual inputs as a one-to-one relationship. It should be appreciated that the inputs shown in FIG. 9B are provided purely for sake of example. NN architecture 950 can be modified to account for various different inputs depending on implementation. For example, if gas temperature is not accounted for as an input, the input layer may only include five input nodes.

NN architecture 950 is also shown to include a hidden layer including hidden nodes. In NN architecture 950, the hidden layer is shown to include a single layer including a number of hidden nodes that is equivalent to the number of input nodes of the input layer. However, it should be noted that, according to various embodiments, the hidden layer can include one or more layers including varying numbers of hidden nodes that may or may not correspond to a number of input nodes. For example, in a convolutional neural network architecture, NN architecture 950 may include multiple hidden layers (e.g., multiple convolutional layers) that have varying numbers of hidden nodes. Moreover, the nodes of each layer need not necessarily connect to every node of adjacent layers as shown in FIG. 9B.

In NN architecture 950, a weight W can be applied with regard to connections between two nodes. In some embodiments, each connection between nodes includes a particular value for a particular connection. For example, a weight between input node 1 of the input layer and hidden node 1 of the hidden layer may be different from a weight between input node 1 and hidden node 2 of the hidden layer. In some embodiments, various connections between nodes may be associated with the same weight. For example, in an LSTM-specific architecture, the weights associated with connections between input nodes and hidden nodes may be the same.

Based on each weighted value incoming to a particular node, a function can be applied to determine a composite value of the node. For example, for hidden node 1 of NN architecture 950, a function can be applied to the weighted input values incoming to the node to determine a composite value of hidden node 1. Composite values of each node in a particular layer to determine outputs of the particular layer. The outputs of the particular layer can correspond with inputs to a subsequent layer along with weights between the particular layer and the subsequent layer. This process can be repeated for each layer until an output layer is reached.

NN architecture 950 is also shown to include an output layer including output nodes. A number of output nodes in the output layer can correspond to desired outputs of the NN model on a one-to-one basis. With particular regard to the VRF system, the outputs may include oil viscosity, oil-refrigerant mixture viscosity, an oil level of one or more compressors, and an oil level of an accumulator. Accordingly, a first, second, and third output node can correspond with the oil viscosity, the oil level of the one or more compressors, and the oil level of the accumulator, respectively. With regard to NN architecture 950, a composite value of output node 1 can correspond to oil viscosity, a composite value of output node 2 can correspond to the oil level of the one or more compressors, and a composite value of output node 3 can correspond to the oil level of the accumulator. In this way, by simply providing the input values to NN architecture 950, predicted values of the outputs can be generated.

Referring now to FIG. 10, a flow diagram of a process 1000 for monitoring oil characteristics using an AI model is shown, according to some embodiments. Process 1000 can leverage the AI model to predict values of the oil characteristics and can initiate corrective actions if said values do not meet predefined thresholds. While process 1000 is described primarily with reference to a building system (e.g., a VRF system), process 1000 can be applied to a variety of systems that include components/devices that require oil for proper operation. For example, process 1000 can be applied to VRF systems, other HVAC systems, car systems, etc. In some embodiments, some and/or all steps of process 1000 are performed by oil management controller 800 as described with reference to FIG. 8.

Process 1000 is shown to include obtaining training data describing a relationship between conditions affecting oil used by building equipment of a building and characteristics of the oil (step 1002). The building equipment can include a variety of devices that can affect a variable state or condition of the building and utilizes oil for proper operation. For example, the building equipment may include compressors, AHUs, other subplants, etc. The training data can be obtained from a variety of sources. For example, the training data may be obtain via direct input from a user, by accessing a database (e.g., a cloud database) storing historical information associated with operation of the building equipment, using training data provided by a manufacturer of the building equipment, etc. In some embodiments, step 1002 includes generating the training data using a simulation model. If a simulation model is used, the simulation model can generate some and/or all of the training data obtained in step 1002. The simulation model can be structured to account for various aspects of the system including the building equipment such as, for example, how much oil is used by devices of the building equipment during operation, how external weather conditions and/or other ambient conditions affect the system, various heating/cooling loads of the building, etc. During generation of the training data, variables associated with the simulation model can be manipulated to generate training data representing a variety of scenarios. Using the simulation model in step 1002 can result in a greater amount of training data being available in a shorter amount of time as compared to gathering data based on actual operation of the building equipment. Moreover, using the simulation model in step 1002 can help obtain training data describing fringe cases that may not be typically included in training data collected based on actual device operation. In some embodiments, step 1002 is performed by training data collector 810.

Process 1000 is shown to include generating an artificial intelligence (AI) model based on the training data to model the characteristics of the oil (step 1004). The AI model generated in step 1004 can be of a variety of different AI models such as, for example, an RNN model (e.g., an LSTM model), a CNN model, etc. The AI model can be generated to associate the conditions affecting the oil and the oil characteristics themselves. Specifically, the AI model may be trained to associate certain inputs (e.g., compressor speed, ambient temperature, discharge temperature, suction pressure, discharge pressure, gas temperature, etc.) with certain outputs (e.g., oil viscosity, an oil level in a compressor, an oil level in an accumulator, etc.). In some embodiments, step 1004 may include training weights associated with connections between nodes of the AI model to account for relationships between the conditions and the characteristics of the oil. In some embodiments, step 1004 is performed by model generator 812.

Process 1000 is shown to include using the AI model to generate predictions of the oil characteristics over time based on a set of model inputs (step 1006). As described above in step 1004, the AI model can be trained to associate certain inputs with certain outputs. Accordingly, once trained, the AI model can utilize values of the inputs to predict corresponding values of the outputs (i.e., the oil characteristics). In some embodiments, step 1006 is performed by prediction generator 814.

Process 1000 is shown to include determining whether the predictions violate any constraints (step 1008). In some embodiments, the constraints are predefined constraints that define acceptable values of the oil characteristics. For example, the constraints may include threshold values that values of the oil characteristics should be above/below, acceptable ranges of values that the values of the oil characteristics should be within, etc. Specifically, the constraints may be thresholds that should not be violated. As a particular example, a constraint for oil viscosity may be defined as a range of acceptable values that the oil viscosity can be within. In said example, if the predicted oil viscosity is above a maximum value of the range or below a minimum value of the range, a violation may be identified. If the predicted oil characteristics do not violate any constraints and therefore has no deficiencies (step 1008, “NO”), process 1000 may repeat starting at step 1006. In this case, a new set of predictions can be generated for a subsequent time step such that the oil characteristics can be monitored/tracked over time. However, if a constraint violation is identified (step 1008, “YES”), process 1000 may proceed to step 1010. In some embodiments, a single constraint violation will result in process 1000 proceeding to step 1010. In some embodiments, multiple constraint violations (e.g., 2 constraint violations, 3 constraint violations, etc.) may be required for process 1000 to proceed to step 1010. In some embodiments, step 1008 includes at least partially accounting for a severity of particular constraint violations in determining whether to proceed to step 1010. For example, an oil viscosity exceeding a maximum viscosity by 0.001 Pascal-seconds may require some other constraint to also be violated for process 1000 to proceed to step 1010 whereas the oil viscosity exceeding the maximum viscosity by 1 Pascal-second may be independently sufficient for process 1000 to proceed to step 1010. In some embodiments, step 1008 may include predicting whether an oil characteristic will violate a constraint within an upcoming time period, and if so, can cause process 1000 to proceed to step 1010 to preemptively address the anticipated violation. In some embodiments, step 1008 is performed by corrective action generator 816.

Process 1000 is shown to include determining a corrective action based on what oil characteristic(s) violated constraints (step 1010). In other words, the corrective action can be determined to address the particular oil characteristic(s) that are violating one or more constraints/thresholds. For example, if an oil viscosity violates a constraint (e.g., a maximum allowable viscosity), the corrective action determined may be to transmit a notification to a user device to notify the user that servicing of the building equipment may be necessary to reduce the oil viscosity. As another example, if an oil level of a compressor is below a minimum threshold, the corrective action determined in step 1010 may be to operate the compressor in oil return control to retrieve more oil (e.g., from an accumulator, from an indoor VRF system, etc.). In some embodiments, step 1010 includes determining a specific time and/or time period for the corrective action to occur. To determine the specific time and/or time period, step 1010 may include monitoring conditions (e.g., operating conditions of devices, ambient conditions, etc.) of the system to determine a time at which a lowest impact on cost, heating/cooling efficiency, etc. may occur. In some embodiments, if the corrective action is transmitting a notification or if the constraint violation is severe, the determined time and/or time period may be a soonest possible time (e.g., immediately). In some embodiments, step 1010 is performed by corrective action generator 816.

Process 1000 is shown to include initiating the corrective action (step 1012). By initiating the corrective action determined in step 1010, the one or more constraint/threshold violations identified in step 1008 can be addressed. In this way, an overall amount of time during which the one or more constraint/threshold violations are active can be reduced. Reducing an amount of time during which constraints/thresholds are violated can help reduce degradation of the building equipment, reduce costs (e.g., energy costs), and can increase overall safety of the system, among other benefits. In some embodiments, if step 1010 includes determining when the corrective action should be performed, step 1012 can include initiating the corrective action at the determined time. In some embodiments, step 1012 is performed by corrective action generator 816.

Experimental Results

Referring generally to FIGS. 11A-12C, results of an example experiment are shown, according to some embodiments. The example experiment of FIGS. 11A-12C is provided for illustrative purposes only and is not intended to be limiting on the present disclosure, but rather to show the practicality of utilizing an AI model in predicting oil characteristics. The AI model referenced below throughout FIGS. 11A-12C is an RNN model trained for purposes of predicting oil characteristics.

Referring now to FIGS. 11A and 11B, a pair of graphs illustrating results of a training process of an AI model for an example experiment are shown, according to some embodiments. FIG. 11A is shown to include a graph 1100 illustrating changes in RMSE based on a number of iterations of the example training process. FIG. 11B is shown to include a graph 1150 illustrating changes in loss based on the number of iterations. The example training process associated with FIGS. 11A and 11B utilized ten closed loop test cases from a VRF model-based definition (MBD) oil viscosity plant model with approximately 4000 seconds allocated for each test case as training data. To determine accuracy of the AI model, a single test data set from the VRF MBD oil viscosity plant model with approximately 4000 seconds allocated for each test case was used for comparison.

Graph 1100 is shown to include a series 1102. Series 1102 can illustrate how the RMSE associated with the AI model changes as a result of additional iterations of the training process. Specifically, series 1102 illustrates a generally decreasing trend as the number of iterations increases. In other words, increasing the number of iterations can improve accuracy of the AI model. It should be noted that series 1102 represents a smoothed curve of the data points of RMSE collected at each iteration.

Graph 1150 is shown to include a series 1152. Series 1152 can illustrate how a loss associated with the AI model changes over time based on a number of iterations. In this case, loss describes how inaccurate predictions of the AI model are with a loss of 0 indicating a particular prediction is equivalent to actual measurements. As is apparent from series 1152 and series 1102, accuracy of the AI model for predicting oil characteristics increases based on the number of iterations.

Referring generally to FIGS. 12A-12C, graphs illustrating oil characteristic predictions of the AI model trained in the example experiment of FIGS. 11A-11B are shown, according to some embodiments. In the example experiment, the AI model took in inputs of compressor speed, ambient temperature, discharge temperature, suction pressure, discharge pressure, and gas temperature to produce predicted outputs of oil viscosity, an oil level in a compressor, and an oil level of an accumulator.

Referring specifically to FIG. 12A, a graph 1200 illustrating predictions of the AI model regarding an oil level of a compressor is shown, according to some embodiments. Graph 1200 is shown to include a series 1202 illustrating predicted oil levels in the compressor over time. In some embodiments, an objective of operational decisions associated with the compressors may be to maintain a relatively constant value of the oil level in the compressors to ensure steady and reliable operation.

Referring now to FIG. 12B, a graph 1210 illustrating predictions of the AI model regarding an oil level of an accumulator is shown, according to some embodiments. Graph 1210 is shown to include a series 1212 illustrating predicted oil levels in the accumulator over time. Changes in series 1212 may result from, for example, changes in operation of compressors. Specifically, compressors may provide output an oil/refrigerant mixture that is separated such that the oil of the oil/refrigerant mixture is gathered by the accumulator, which may result in predicted increases of series 1212. Alternatively, compressors may operate to retrieve oil from the accumulator, which may result in decreases of series 1212.

Referring now to FIG. 12C, a graph 1220 illustrating predictions of the AI model regarding a viscosity of oil is shown, according to some embodiments. Oil viscosity may change based on external temperature, how compressors utilize the oil, etc. Accordingly, the AI model can predict how the oil viscosity changes over time based on how an overall system utilizing the oil changes.

Utilizing AI for VRF System Management

Referring generally to FIGS. 13A-24, systems and methods for utilizing AI for generating predictions regarding characteristics of VRF systems are shown and described, according to some embodiments. In particular, the systems and methods described below illustrate how AI can be utilized to predict vibrations of compressors in VRF systems, fault conditions in VRF systems, and efficiency of motors within compressors. Based on the predictions of the AI, determinations can be made regarding whether any characteristics of the VRF system and/or components therein are operating beyond preferred operating bounds. As described herein, a corrective action can refer to any action taken to address some predicted characteristic of a VRF system and/or one or more components therein being beyond predefined acceptable bounds (e.g., predefined thresholds). For example, if an AI model predicts that a fault condition exists for the VRF system, a corrective action that automatically schedules maintenance to fix the fault condition may be initiated. As another example, if an AI predicts that a rate of vibrations of a compressor is beyond a predefined maximum threshold (e.g., a maximum number of motion cycles per minute), a corrective action that can be initiated may be to lower an input current being provided to the compressor and/or to temporarily disable to the compressor.

As described herein, the term “AI” can be used to refer to a variety of different types of AI models used for generating predictions. In particular, the systems and methods described herein may leverage neural networks to generate predictions associated with VRF systems. For example, recurrent neural networks (RNNs) such as long short-term memory (LSTM) networks may be used to generate predictions. Example structures of LSTMs are described in greater detail below with reference to FIGS. 25A and 25B. Of course, it should be appreciated that other types of neural networks and/or different AI models can be utilized in generating predictions associated with VRF systems. For example, convolutional neural networks (CNNs), multi-layer perceptrons, etc. can be utilized in generating predictions. In some embodiments, multiple AI models are trained for generating specific predictions and can be tested for an accuracy of predictions relative to a known data set. Based on the determined accuracy of each AI model, an AI model that generates predictions that most closely match the known data set can be selected and utilized.

As described in greater detail below, AI models utilized for generating predictions can be trained based on a variety of different data sources. In some embodiments, the AI models are trained based on real data gathered from one or more VRF systems in operation. In some embodiments, the AI models are trained based on data generated based on a simulated environment. In this case, the simulated environment can be structured to mimic an actual VRF system in operation. Advantageously, using a simulation for generating training data may result in a larger set of training data being generated than may otherwise be generated by a real VRF system in operation. In some embodiments, the AI models are trained based on a mix of training data generated by a simulation, gathered from an actual VRF system, and/or from some other source of training data.

By utilizing AI models to generate predictions associated with VRF systems, issues within the VRF systems can be identified and addressed quickly. Moreover, the AI models can help accurately identify possible issues within the VRF systems before they become a more serious issue (e.g., complete failure of a building device). In this way, the AI models can

Compressor Vibration Prediction

Referring generally to FIGS. 13A-18, systems and methods for managing vibrations of compressors in a VRF system are shown and described, according to some embodiments. However, it should be appreciated that the below description is not necessarily limited to compressors in a VRF system. The systems and methods described below can be applied to a variety of different compressors and/or different building equipment in a variety of systems (e.g., different HVAC systems).

Vibrations of compressors can result in rapid degradation of the compressors. As a frequency and/or intensity of vibrations increases, a compressor may become increasingly susceptible to faults and other issues that may require maintenance and/or complete replacement of the compressor. However, complete elimination of vibrations may be unrealistic as compressors should vibrate to some extent during operation. As such, operation of the compressors should be balanced between reducing vibrations while still fulfilling loads and/or other required needs of the VRF system.

As referred to herein, “compressors” may be used to generally refer to a variety of different types of compressors that can be utilized in VRF systems. For example, VRF systems may include single rotary compressors, twin rotary compressors, scroll compressors, etc. In VRF systems, a compressor may be operated based on an alternating current (AC) provided by an inverter. Specifically, the inverter can change a frequency of the AC provided to the compressor to change a speed at which a motor of the compressor rotates and can change an amplitude of the AC to change a torque applied by the motor.

As the AC provided by the inverter directly affects operation of the compressor, a correlation between characteristics of the AC (e.g., frequency and/or amplitude) and vibrations of the compressor can be identified. For example, a frequency of the vibrations may increase as a frequency and/or intensity (i.e., amplitude) of the AC increases. Based on identified relationships, an AI model can be trained to learn what values of current characteristics (e.g., amplitude) can result in reduced vibrations while still satisfying heating/cooling loads and/or other requirements of the compressor. In some embodiments, and as described in greater detail below, the AI model may specifically be trained to predict appropriate values of AC amplitude to modify a torque applied by the motor in order to manage (e.g., reduce) vibrations.

In some embodiments, the AI model structured for predicting appropriate current values for a compressor utilizes noise as a proxy for vibrations. In this way, the AI model can be trained to predict an appropriate amplitude of the AC provided to the compressor that reduces an amount of noise produced by the compressor. Using noise as a proxy for vibrations may be appropriate as a result of known relationships between noise and vibrations. In general, louder noises may indicate increased vibrations whereas quieter noises may indicate fewer vibrations. More particularly, a relationship between decibels (dBs) and a frequency of vibrations of the compressor in hertz (Hz) may exist. For example, a 1 dB increase in noise may be determined to be correlated with a 1 Hz increase in vibration frequency. Advantageously, using noise as a proxy for vibrations may reduce overall costs as sensors for directly measuring vibrations may be significantly more expensive as compared to sensors for measuring sound (e.g., standard audio sensors).

Referring now to FIG. 13A, a block diagram of a VRF system 1300 is shown, according to some embodiments. As should be appreciated, VRF system 1300, as shown in FIG. 13A, illustrates a portion of a larger VRF system that utilizes compressors to provided heating/cooling to a building. For example, VRF system 1300 may be a subsystem of VRF system 600 as described with reference to FIGS. 6A and 6B.

VRF system 1300 is shown to include a converter 1304 and an inverter 1306. Converter 1304 can receive alternating current (AC) power from an AC power source (not shown) and can convert the AC power to direct current (DC) power. Converter 1304 can provide the converted DC power to inverter 1306. Inverter 1306 may be an electronic modulator that changes a frequency of the received DC signal and outputs an AC signal to a compressor 1302. In particular, inverter 1306 can operate to manipulate the frequency and/or amplitude of the AC signal such that the outputted AC signal results in a motor 1308 of compressor 1302 operating at a specific speed and torque.

In some embodiments, the frequency of the AC signal provided by inverter 1306 is relative to a known amount of a fluid passing through compressor 1302 and/or a known amount of the fluid that is expected to pass through compressor 1302. The frequency of the AC signal can be indicative of how quickly motor 1308 should rotate and/or is rotating, which may be measured in rotations per minute, as to ensure a proper amount of the fluid flows through compressor 1302. The frequency of the AC signal may thereby be determined based on heating/cooling loads of a building and/or other targets. For example, a higher required cooling load for the building may result in a higher frequency of the AC signal, thereby resulting in motor 1308 rotating quicker to pass more fluid through compressor 1302. The frequency of the AC signal may be determined by a controller, by inverter 1306 itself, based on feedback from compressor 1302, etc.

To determine what amplitude (also referred to herein as intensity) of the AC signal to output, inverter 1306 may receive an AC amplitude setpoint from a compressor vibration controller 1310. In some embodiments, compressor vibration controller 1310 utilizes an AI model to predict what amplitude value (e.g., in volts) should be provided to motor 1308 as to ensure compressor 1302 does not degrade too quickly as a result of vibrations. Based on the received AC amplitude setpoint, inverter 1306 can modulate the DC signal received from converter 1304 as to output an AC signal to compressor 1302 that matches (or nearly matches) the AC amplitude setpoint provided by compressor vibration controller 1310 as well as the desired frequency described above. Compressor vibration controller 1310 is described in greater detail below with reference to FIG. 16.

VRF system 1300 is also shown to include compressor 1302. Compressor 1302 can be any type of compressor (e.g., a single rotary compressor, a twin rotary compressor, a scroll compressor, etc.) used to compress a fluid. Compressor 1302 can be structured to intake a fluid (e.g., a gas) via a suction line, compress the fluid, and output the compressed fluid via a discharge line. As described above, compressor 1302 may include motor 1308 that drives components of compressor 1302 to compress the fluid and move the fluid through compressor 1302. Motor 1308 can operate based on the AC signal (electric current) received from inverter 1306. Specifically, the frequency of the AC signal can affect an operating speed of motor 1308.

In traditional systems, the electric current provided by inverter 1306 may be a predetermined, stagnant value that does not account for vibrations of compressor 1302. These traditional systems may result in compressor 1302 becoming quickly degraded as compressor 1302 may be vibrating at dangerous frequencies and/or intensities. However, as described in greater detail below with reference to FIG. 16, compressor vibration controller 1310 can predict an appropriate value for the amplitude of the AC signal provided by inverter 1306 such that vibrations of compressor 1302 can be managed (e.g., reduced). In this way, inverter 1306 can consistently provide an electric current to compressor 1302 that is determined respective of possible vibrations as to increase the longevity of compressor 1302.

Referring now to FIG. 13B, a block diagram of VRF system 1300 in greater detail is shown, according to some embodiments. Motor 1308 may be any type of motor that operates based on an AC signal. For example, motor 1308 may be a single-phase motor that operates based on a single-phase source of power. As another example, and as shown in FIG. 13B, motor 1308 may be a three-phase motor that operates based on a three-phase source of power.

As described above with reference to FIG. 13A, inverter 1306 can be configured to provide an AC signal to motor 1308 based on the AC amplitude setpoint provided by compressor vibration controller 1310. As shown in FIG. 13B, the signal provided to motor 1308 may be a three-phase signal including phases A, B, and C. In this case, compressor vibration controller 1310 may generate a prediction for the three-phase signal needed to properly operate motor 1308. Each of the phases A, B, and C may correspond to a specific axis, namely the A-axis, B-axis, and C-axis, respectively.

In some embodiments, compressor 1302 may receive a DC signal as opposed to an AC signal. In this case, inverter 1306 may or may not be a component of VRF system 1300. For example, converter 1304 may include some and/or all of the functionality of inverter 1306 for modulating a signal to provide to motor 1308. An intensity of the DC signal may be determined by compressor vibration controller 1310 and provided to converter 1304 in order to affect a torque applied by motor 1308. However, a speed at which motor 1308 may be determined based on some other input value other than a frequency of the electric signal. For example, the rotational speed of motor 1308 may be directly provided to motor 1308 by compressor vibration controller 1310 and/or some other computing device. However, compressor 1302/motor 1308 are described herein as receiving an AC signal from inverter 1306 for ease of explanation and clarity.

FIG. 13B further illustrates the direct axis (d-axis) and the quadrature axis (q-axis) of motor 1308. The d-axis and the q-axis may always be at a 90 degree angle relative to one another. In some embodiments, the three-phase current provided by inverter 1306 may be separated for the d-axis and the q-axis. In this case and as described in greater detail below with reference to FIG. 16, compressor vibration controller 1310 may generate a first three-phase current prediction for the d-axis and a second three-phase current prediction for the q-axis. In this way, motor 1308 can be properly operated to compress a received fluid. Calculations of the axial currents made by compressor vibration controller 1310 are described in greater detail below with reference to FIGS. 14A-14C.

The three-phase current signal provided by inverter 1306 is shown to include phases A, B, and C. In some embodiments, compressor vibration controller 1310 is configured to predict appropriate amplitude values for each phase. In some embodiments, compressor vibration controller 1310 is configured to predict a single amplitude value that applies to phases A, B, and C. It should be appreciated that using a single amplitude value that applies to each phase may be less computationally expensive to determine as compared to determining three separate amplitude values.

Referring now to FIGS. 14A-14C, a set of graphs illustrating current conversions are shown, according to some embodiments. FIGS. 14A-14C illustrate relationships between currents that can be used by compressor vibration controller 1310 in generating predictions for what current values to provide to a motor (e.g., motor 1308). Compressor vibration controller 1310 can be configured to perform both forward and reverse dq current conversions. The d-axis and q-axis of a motor can be associated with current values I_dcand I_qc, respectively. In forward current conversion, I_dcand I_qccan be determined by converting a detected U-phase and W-phase observed current values I_ucand I_wcto values on a virtual rotational coordinate dcqc-axis.

Reverse current conversion can relate estimated current values Î_ucand Î_wcused in UVW current capture processing. UVW current capture processing can include inversely transforming the value of the virtual rotational coordinate dcqc-axis to determine Î_ucand Î_wc. A V-phase current Î_vccan additionally be determined via Kirchoff's law, as described in greater detail below. In the UVW current capture processing, I_uand I_wcan be estimated in order to compensate for the motor current information when detection is invalidated in a current detection condition setting processing. Estimated values I_ucand I_wcof I_uand I_wcan be obtained by inversely converting to the UVW based on the value of the dcqc-axis. In addition, I_vcused for calculation formula for I_u. I_westimation can be obtained from Kirchhoff s law.

Referring more particularly to FIG. 14A, a graph 1400 illustrating Kirchhoff's law is shown, according to some embodiments. Graph 1400 is shown to include currents I_v, I_u, and I_wemanating from a point a. Following Kirchoff's law, a sum of the current at point a is 0. Accordingly, Kirchoff's law holds that:

I_v=−I_u−I_w

This relationship can be used in both forward and reverse current conversion as described in greater detail below with reference to FIGS. 14A and 14B.

Referring now to FIG. 14B, a graph 1420 illustrating an αβ conversion is shown, according to some embodiments. In forward current conversion, an αβ conversion can be performed for currents I_αcand I_βcby compressor vibration controller 1310. The αβ conversion can be represented by the following equation:

$[\begin{matrix} I_{α c} \\ I_{β c} \end{matrix}] = \frac{2}{3} [\begin{matrix} 1 & - \frac{1}{2} & - \frac{1}{2} \\ 0 & \frac{\sqrt{3}}{2} & - \frac{\sqrt{3}}{2} \end{matrix}] [\begin{matrix} I_{u} \\ I_{v} \\ I_{w} \end{matrix}]$

If the relationship of I_v=−I_u−I_wis substituted into the above equation by compressor vibration controller 1310, the following relationship can be identified:

$[\begin{matrix} I_{α c} \\ I_{β c} \end{matrix}] = [\begin{matrix} 1 & 0 \\ - \frac{1}{\sqrt{3}} & - \frac{2}{\sqrt{3}} \end{matrix}] [\begin{matrix} I_{u c} \\ I_{w c} \end{matrix}]$

where I_uc=I_uand I_wc=I_w. In this way, I_αcand I_βccan be obtained using Kirchoff's law.

Referring now to FIG. 14C, a graph 1440 illustrating a dq conversion is shown, according to some embodiments. In graph 1440, θ_dccan represent an angle between the fixed coordinate α-axis and the virtual rotational coordinate dc-axis. In forward current conversion, the dq conversion can be represented by the following equation where I_dcand I_qccan be obtained by substituting I_αcand I_βc:

$[\begin{matrix} I_{d c} \\ I_{q c} \end{matrix}] = [\begin{matrix} \cos (θ_{d c}) & \sin (θ_{d c}) \\ - \sin (θ_{d c}) & \cos (θ_{d c}) \end{matrix}] [\begin{matrix} I_{α c} \\ I_{β c} \end{matrix}]$

Values of I_dcand I_qccan be filtered and used for control of a compressor motor (e.g., motor 1308). To filter I_dcand I_qc, a filter time constant (e.g., in milliseconds, in seconds, etc.) of the filter can be switched according to a rotational speed of the motor. In some embodiments, the rotational speed is divided into three regions, namely a low speed region, a medium speed region, and a high speed region. For example, the low speed region may be defined as 0-99 rotations per minute (RPM), the medium speed region may be defined as 100-200 RPM, and the high speed region may be defined as 200+ RPM. Based on an identified rotational speed, the filter time constant can be switched respective of the observed current. In particular, the filter time constant may be the largest in the low speed region and shortest in the high speed region. In this way, as rotational speed increases, the filter time constant can decrease to obtain more granular resolution of current.

With regard to reverse current conversion, a dq inverse transformation can be performed by compressor vibration controller 1310 based on the following dq inverse transformation equation:

$[\begin{matrix} I_{α c} \\ I_{β c} \end{matrix}] = [\begin{matrix} \cos (θ_{d c}) & - \sin (θ_{d c}) \\ \sin (θ_{d c}) & \cos (θ_{d c}) \end{matrix}] [\begin{matrix} I_{d c} \\ I_{q c} \end{matrix}]$

In this way, values of I_αcand I_βccan be obtained. Compressor vibration controller 1310 can further perform an αβ inverse transformation to obtain values of I_ucand I_wcusing the following equation:

$[\begin{matrix} {\hat{I}}_{u c} \\ {\hat{I}}_{w c} \end{matrix}] = [\begin{matrix} 1 & 0 \\ - \frac{1}{2} & - \frac{\sqrt{3}}{2} \end{matrix}] [\begin{matrix} I_{α c} \\ I_{β c} \end{matrix}]$

Referring now to FIG. 15, a graph 1500 a relationship between crank angle and torque for different types of compressors is shown, according to some embodiments. Specifically, graph 1500 illustrates a relationship between a crank angle and a gas compression torque. As described in detail above with reference to FIGS. 13A and 13B, a torque applied by a motor (e.g., motor 1308) can affect vibrations of a compressor. Accordingly, a torque applied by the motor should be monitored and maintained as to ensure excessive vibrations of the compressor are avoided.

Graph 1500 is shown to include a series 1502, a series 1504, and a series 1506. Series 1502 is associated with torque measurements of a single-rotary compressor, series 1504 is associated with torque measurements of a twin-rotary compressor, and series 1506 is associated with torque measurements of a scroll compressor all with respect to a crank angle of the motor. As should be appreciated based on graph 1500, each type of compressor may have a different relationship between crank angle and torque (and thereby vibrations). For example, the single-rotary compressor represented by series 1502 is shown to have a high peak torque with respect to the crank angle whereas the scroll compressor represented by series 1506 has a fairly constant relationship between the crank angle and torque. During a training process for an AI model, compressor vibration controller 1310 can leverage these relationships and generate different AI models for various types of compressors.

Referring now to FIG. 16, a block diagram of compressor vibration controller 1310 in greater detail is shown, according to some embodiments. Compressor vibration controller 1310 can be configured to utilize an AI model to generate predictions for values of current that can be used to operate compressor 1302 (and/or some other compressor) and reduce degradation of compressor 1302 due to vibrations. Compressor vibration controller 1310 can be implemented in a variety of ways. In some embodiments, compressor vibration controller 1310 is a local controller of a building system. For example, compressor vibration controller 1310 may be implemented on a desktop computer, a mobile device, a thermostat, and/or some other computing device/system local to a building. In some embodiments, compressor vibration controller 1310 may be implemented as a component of a VRF system. For example, compressor vibration controller 1310 may be implemented as a component of inverter 1306 as described with reference to FIG. 13A. In this case, inverter 1306 itself may be able to determine a frequency of a current to provide to compressor 1302. In some embodiments, compressor vibration controller 1310 is implemented via some other computing device/system such as by a cloud computing system.

Compressor vibration controller 1310 is shown to include a communications interface 1608 and a processing circuit 1602. Communications interface 1608 may include wired or wireless interfaces (e.g., jacks, antennas, transmitters, receivers, transceivers, wire terminals, etc.) for conducting data communications with various systems, devices, or networks. For example, communications interface 1608 may include an Ethernet card and port for sending and receiving data via an Ethernet-based communications network and/or a Wi-Fi transceiver for communicating via a wireless communications network. Communications interface 1608 may be configured to communicate via local area networks or wide area networks (e.g., the Internet, a building WAN, etc.) and may use a variety of communications protocols (e.g., BACnet, IP, LON, etc.).

Communications interface 1608 may be a network interface configured to facilitate electronic data communications between compressor vibration controller 1310 and various external systems or devices (e.g., inverter 1306, sensors 1620, a user device 1622, etc.). For example, compressor vibration controller 1310 may provide AC amplitude setpoints to inverter 1306 via communications interface 1608.

Processing circuit 1602 is shown to include a processor 1604 and memory 1606. Processor 1604 may be a general purpose or specific purpose processor, an application specific integrated circuit (ASIC), one or more field programmable gate arrays (FPGAs), a group of processing components, or other suitable processing components. Processor 1604 may be configured to execute computer code or instructions stored in memory 1606 or received from other computer readable media (e.g., CDROM, network storage, a remote server, etc.).

Memory 1606 may include one or more devices (e.g., memory units, memory devices, storage devices, etc.) for storing data and/or computer code for completing and/or facilitating the various processes described in the present disclosure. Memory 1606 may include random access memory (RAM), read-only memory (ROM), hard drive storage, temporary storage, non-volatile memory, flash memory, optical memory, or any other suitable memory for storing software objects and/or computer instructions. Memory 1606 may include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described in the present disclosure. Memory 1606 may be communicably connected to processor 1604 via processing circuit 1602 and may include computer code for executing (e.g., by processor 1604) one or more processes described herein. In some embodiments, one or more components of memory 1606 are part of a singular component. However, each component of memory 1606 is shown independently for ease of explanation.

Memory 1606 is shown to include a training data collector 1610. Training data collector 1610 can collect training data used to train an artificial intelligence model from one or more training data sources 1618. Specifically, training data collector 1610 can obtain training data associated with vibrations of compressor 1302.

The training data collected by training data collector 1610 can include any relevant data that can be used to train an AI model to learn associations between certain inputs and vibrations of compressor 1302. Alternatively or additionally, the training data may include information indicative of a relationship between certain inputs and noise generated by compressor 1302 if noise is used as a proxy for vibrations. The training data may include values for inputs including, for example, an axial error between an A-axis and a d-axis of motor 1308, a detected q-axis current, a frequency of the AC signal provided by inverter 1306, measured noised near compressor 1302, etc. The training data can also include values of current characteristics (e.g., amplitude) such that the AI model can learn relationships between the inputs and outputs.

To gather the training data, training data collector 1610 may transmit queries to training data sources 1618 to obtain the training data. In some embodiments, training data collector 1610 may passively receive training data from training data sources 1618 without needing to actively request the training data. Training data sources 1618 can include any source of data that can store and/or provide training data to training data collector 1610. For example, training data sources 1618 may be or include a user device (e.g., a laptop, a desktop computer, a mobile device, a tablet, etc.) that can provide a stored training data set to training data collector 1610. As another example, training data sources 1618 may be or include a database (e.g., a cloud database) that stores data collected during operation of an actual VRF system. In this way, the AI model can be trained based on data collected directly actual VRF devices in operation.

In some embodiments, training data collector 1610 utilizes one or more simulation models to generate some or all of the training data used by model generator 1612 to generate the AI model. A simulation model (also referred to as a “simulation” or “simulation framework” herein) can simulate how an actual VRF system may operate under various conditions and limitations (e.g., weather conditions, heating/cooling loads, intrinsic device limitations, etc.). The simulation model can account for relationships between components of the VRF system and how the components may react to changing conditions. For example, the simulation model can model how compressor 1302 may operate to achieve certain required heating/cooling loads and how compressor 1302 vibrates as a result.

By utilizing the simulation model, training data collector 1610 may not need to retrieve training data from training data sources 1618 and instead can generate the training data within compressor vibration controller 1310. In some embodiments, the simulation model is hosted by a third party controller/device/system (e.g., a cloud computing system) which can provide the training data generated as a result of running the simulation model to compressor vibration controller 1310. In any case, the simulation model can be used/executed to generate a variety of training data representing various operating conditions that can be used to train an AI model.

Advantageously, the simulation model can generate a large amount of data in a shorter amount of time as compared to gathering training data from an actual VRF system in operation. In this way, compressor 1302 can be operated based on decisions of an AI model in a shorter at an earlier time of time, which may increase the longevity of compressor 1302 as vibrations are managed sooner. Moreover, the simulation model can be used to generate training data for scenarios that may not frequently occur during actual VRF system operation. For example, the simulation model may be able to generate training data representative of operating compressor 1302 under intense heating/cooling loads that may not occur frequently during standard operation of an actual VRF system. In this way, a large amount of training data representative of a variety of cases can be generated and made available to model generator 1612 to generate an accurate and representative model.

Based on the obtained training data, training data collector 1610 can combine the collected training data into a training data set and provide the training data set to a model generator 1612. Based on the training data set, model generator 1612 can generate an AI model that models a relationship between a set of inputs describing operating conditions of compressor 1302 and corresponding outputs describing amplitudes of AC signals that can be provided to compressor 1302 to manage vibrations of compressor 1302. In some embodiments, model generator 1612 trains the AI model to predict AC signal amplitudes that can be provided to motor 1308 via inverter 1306 as to reduce noise generated by compressor 1302 while still satisfying required heating/cooling loads.

The AI model generated by model generator 1612 can be of any various AI model architectures. For example, the AI model may be an RNN such as an LSTM network.

Inputs to the AI model generated by model generator 1612 can include a variety of inputs associated with operation of compressor 1302 and/or motor 1308. For example, inputs to the AI model may include an axial error (also referred to as an axial difference) between the d-axis and A-axis of motor 1308, a q-axis current detection value, an inverter frequency indicative of a rotational speed of motor 1308, and a real and/or required noise level. In some embodiments, the rotational speed of motor 1308 (e.g., in rotations per minute (RPMs)) is used as an input instead of or in addition to the inverter frequency. Outputs of the AI model can input target amplitude values of the AC signal (e.g., in volts, in amps, etc.) to provide to compressor 1302 in order to manage a torque applied by motor 1308. Example illustrations of the AI models that can be generated by model generator 1612 are described in greater detail below with reference to FIGS. 17A and 17B.

Model generator 1612 may utilize a variety of training techniques to generate the AI model. For example, model generator 1612 may utilize a stochastic gradient descent with momentum approach, an adaptive moment estimation approach, a root mean square propagation approach, etc. With specific regard to the root mean square propagation approach, model generator 1612 may utilize a root mean squared error (RMSE) to measure how accurate model predictions are relative to the training data provided by training data collector 1610. To monitor the RMSE over time, model generator 1612 may utilize the following equation:

RMSE=√{square root over ((Y_pred,t−Y_test,t)²)}

where Y_pred,tis a previous prediction of the AI model for a variable Y at a time step t, and Y_test,tis an actual value of the variable Y as indicated by the training data at time step t. The calculation of (Y_pred,t-Y_test,t)²can be performed for each time step t=1. . . n where n is a total number of predictions. Each difference can then be averaged together. During the training process, model generator 1612 can refine the AI model to reduce the RMSE.

Model generator 1612 can provide the generated AI model to a prediction generator 1614. Prediction generator 1614 can use the AI model to generate predictions for values of AC signal amplitude that reduce an amount of vibrations and/or noise generated by compressor 1302. In order to generate said predictions, prediction generator 1614 can operate to obtain values of inputs required by the AI model from a variety of sources. For example, prediction generator 1614 may obtain measured variables from sensors 1620, values of variables manually observed by users via user device 1622, and/or any other appropriate source of input values. With regard to specific inputs to the AI model, a rotational speed of motor 1308 may be observed by a user, directed calculated by prediction generator 1614 based on a known frequency being provided to motor 1308, etc. In some embodiments, the axial difference between the A-axis and d-axis of motor 1308 is estimated by an observer (e.g., a user) by providing a small current to motor 1308 and analyzing feedback from motor 1308 to estimate the axial difference. In some embodiments, the axial difference is measured by a sensor of sensors 1620. However, manual estimation may be preferred due to a price associated with sensors configured to measure the axial difference. In some embodiments, the q-axis feedback current associated with compressor 1302 is directly measured via a current sensor of sensors 1620. With respect to noise, a real noise level may be measured via audio sensors of sensors 1620 and/or estimated by a user. If the AI model utilizes a required noise level as input, the required noise level may be provided by a user via user device 1622, estimated/determined by prediction generator 1614 (or another component of compressor vibration controller 1310), etc. In this case, the required noise level describes a predetermined noise value (e.g., in dB) that should be achieved by compressor 1302.

As a result of passing received values of inputs through the AI model generated by model generator 1612, prediction generator 1614 can determine an amplitude setpoint for the AC signal provided to compressor 1302 by inverter 1306. In some embodiments, the AC amplitude setpoint includes distinct amplitudes for the current provided for the d-axis and the current provided for the q-axis. In some embodiments, a single AC amplitude setpoint is determined as a result of passing the input values through the AI model. In this case, the single AC amplitude setpoint can be applied to both the d-axis current and the q-axis current.

In some embodiments, prediction generator 1614 directly provides the AC amplitude setpoint to inverter 1306 (e.g., via communications interface 1608). In some embodiments, prediction generator 1614 provides the AC amplitude setpoint to a corrective action generator 1616 of memory 1606. Corrective action generator 1616 can be configured to generate and initiate one or more corrective actions based on the AC amplitude setpoint. In some embodiments, one corrective action is to provide the AC amplitude setpoint to inverter 1306. In this case, the corrective action generated and initiated by corrective action generator 1616 may be to operate inverter 1306 based on a control signal indicative of the AC amplitude setpoint. In this way, a torque applied by motor 1308 can be adjusted. In some embodiments, another corrective action that may be generated and initiated by corrective action generator 1616 is to provide a notification to user device 1622 as to notify a user about the AC amplitude setpoint. Said notification may allow the user to appreciate how differing AC amplitudes affect vibrations and may allow the user to take preemptive measures to ensure building equipment (e.g., compressor 1302) avoids rapid degradation due to vibrations. In some embodiments, corrective action generator 1616 generates and initiates other types of corrective actions such as, for example, scheduling maintenance on compressor 1302 in response to determining that the AC amplitude setpoint needed to avoid rapid degradation is below a threshold value, thereby jeopardizing fulfilment of heating/cooling loads of the building.

Referring now to FIG. 17A, an illustration of a neural network (NN) 1700 for predicting values of a current to provide to a compressor motor is shown, according to some embodiments. NN 1700 can illustrate an example structure of an AI model that can be generated by model generator 1612 as described with reference to FIG. 16. NN 1700 is shown to include input nodes in an input layer that correspond to a set of inputs. Specifically, NN 1700 is shown to include input nodes for an axial error between the A-axis and d-axis of motor 1308, a q-axis current detection value, an inverter frequency describing the frequency of the AC signal outputted by inverter 1306, and a real noise level. Of course, it should be appreciated that these inputs are provided purely for sake of example. Model generator 1612 can generate AI models that utilize different inputs that can be used to estimate vibrations/noise of a compressor. In terms of outputs, NN 1700 is shown to include a separate output for a d-axis current amplitude and a q-axis current amplitude. Separate outputs for the d-axis and q-axis may be useful in more precisely operating compressor 1302. In some embodiments, NN 1700 outputs a single current amplitude value that can be applied to both the d-axis current and the q-axis current. Outputting a single value may reduce complexity of NN 1700, thereby improving processing efficiency when generating predictions.

As described above with reference to FIG. 16, the axial error may be measured by a user and/or a sensor, the q-axis current detection value may be measured by a current sensor, and the inverter frequency may be measured, a known value, etc. As for the real noise level, the real noise level may be measured by an audio sensor with a predefined physical proximity of compressor 1302 (e.g., within 10 feet, within 1 foot, physically attached to compressor 1302, etc.). Based on the real noise level, NN 1700 intrinsically predict vibrations of compressor 1302 and can output an amplitude setpoint respective of the predicted vibrations. In this case, NN 1700 is being provided with feedback regarding operation of compressor 1302. Specifically, as noise can be used as a proxy for vibrations, NN 1700 is effectively obtaining knowledge indicative of vibrations of compressor 1302. As such, NN 1700 can leverage this information to predict AC signal amplitudes that avoid dangerous operating conditions. For example, if the noise level of relatively low (e.g., <50 dB, <20 dB, etc.), NN 1700 may output larger amplitudes (e.g., in volts or amps) as compared to if the noise level is relatively high (e.g., >50 dB, >100 dB, >200 dB, etc.). In this way, NN 1700 can change values of the amplitudes over time to achieve appropriate operating conditions for compressor 1302 that avoid dangerous operating conditions and vibrations.

Referring now to FIG. 17B, an illustration of a neural network (NN) 1750 for predicting values of a current to provide to a compressor motor is shown, according to some embodiments. In some embodiments, NN 1750 is similar to and/or the same as NN 1700 as described above with reference to FIG. 17A. Accordingly, NN 1750 may be another example of an AI model that can be generated by model generator 1612 as described with reference to FIG. 16. As shown in FIG. 17B, NN 1750 is shown to receive a required noise level as input as compared to the real noise level of NN 1700. The required noise level may be a target value for noise generated by compressor 1302. In some embodiments, the required noise level functions as a threshold value such that the noise generated by compressor 1302 is below the threshold. In particular, the required noise level may be a noise level (e.g., in dB) that is known to be associated with vibrations that are determined (e.g., by a user) not to place compressor 1302 in dangerous operating conditions. For example, the required noise level may be 50 dB which is expected to maintain vibrations of the compressor 1302 under a rate of 30 Hz.

In order to use required noise level as an input, NN 1750 may be trained based on a known correlation between noise and vibrations. In this way, NN 1750 can adjust the outputted current amplitudes based on other inputs (e.g., the inverter frequency which is known to be associated with an amount of fluid passing through compressor 1302) relative to the known correlation. In some embodiments, model generator 1612 generates an AI model that utilizes both the real noise level and the required noise level as inputs.

Referring now to FIG. 18, a flow diagram of a process 1800 for predicting an AC signal amplitude to provide to a compressor using an AI model is shown, according to some embodiments. The AC signal amplitude may be an amplitude value (e.g., in volts, in amps, etc.) that is predicted to result in decreased vibrations of a compressor while still fulfilling needed heating/cooling loads and/or other requirements of a building. In some embodiments, the amplitude of the AC signal affects a torque applied by a motor of the compressor. Specifically, increasing amplitudes may relate to increasing torque applied by the motor. In some embodiments, some and/or all of the steps of process 1800 are performed by compressor vibration controller 1310 as described with reference to FIGS. 13 and 16.

Process 1800 is shown to include obtaining training data describing a relationship between conditions affecting operation of a compressor and noise generated by the compressor (step 1802). In this case, noise can be used as a proxy for vibrations of the compressor. The training data may include values of, for example, an axial error between an A-axis and a d-axis of a motor of the compressor, a q-axis current detection value, a frequency of an AC signal provided by an inverter (which may represent an amount of fluid passing through the compressor and thereby a required heating/cooling load), a real and/or required noise level, noise measurements, etc. The training data may be obtained from one or more training data sources such as, for example, a cloud database, a database local to a building, a user, sensors in the building, etc. In some embodiments, step 1802 is performed by training data collector 1610.

Process 1800 is shown to include training an AI model for predicting amplitudes of an AC signal that affects a torque applied by a motor of the compressor based on the training data (step 1804). In some embodiments, larger AC signal amplitudes correspond with increased torque applied by the motor. However, as the torque applied by the motor increases, vibrations of the compressor may increase which can be measured based on an amount of noise outputted by the compressor. The AI model trained in step 1804 can learn what AC signal amplitudes correspond with certain noise levels dependent on a set of input values. In some embodiments, a goal of the AI model is to learn what AC signal amplitudes reduce and/or minimize noise outputted by the compressor while still fulfilling necessary heating/cooling loads. Of course, some noise/vibrations may be inevitable due to operation, however the AI model can be trained to identify amplitudes that avoid excessive degradation of the equipment while still fulfilling needs of a building. In some embodiments, step 1804 is performed by model generator 1612.

Process 1800 is shown to include obtaining data associated with operation of the compressor (step 1806). In this case, the data may be associated with inputs to the AI model trained in step 1804. For example, the data obtain in step 1806 may include frequencies of the AC signal provided and/or to be provided to the compressor, an ambient temperature, a real noise level near the compressor, etc. The data can be obtained from a variety of sources such as, for example, a user, sensors, as feedback from equipment, etc. In some embodiments, step 1806 is performed by prediction generator 1614.

Process 1800 is shown to include using the AI model to generate an AC amplitude setpoint based on the obtained data (step 1808). The AI model can generate the AC amplitude setpoint such that a required heating/cooling load (and/or some other requirement/need of a building) is satisfied and that corresponding noise produced by the compressor is reduced. In this way, the motor is operated to compress a certain amount of fluid while avoiding unnecessary degradation. Advantageously, using the AI model to generate the AC amplitude setpoint can reduce a frequency of situations where the compressor experiences unnecessary degradation as a result of the motor applying unnecessary amounts of torque. In some embodiments, the AC amplitude setpoint includes separate amplitude setpoints for a d-axis and a q-axis of the motor. In some embodiments, step 1808 is performed by prediction generator 1614.

Process 1800 is shown to include providing the AC amplitude setpoint to an inverter in order to operate the motor (step 1810). In some embodiments, the inverter generates the actual AC signal provided to the compressor. In this way, the inverter can operate such that the AC signal outputted by the inverter has an amplitude equal to (or approximately equal to) the AC amplitude setpoint. As a result, the motor can operate to apply torque respective of the amplitude of the AC signal. In some embodiments, step 1810 is performed by prediction generator 1614 and/or corrective action generator 1616.

VRF System Fault Condition Prediction

Referring generally to FIGS. 19-22, system and methods for predicting fault conditions of a VRF system are shown and described, according to some embodiments. Fault conditions can include any sort of fault that can raise costs and/or result in some other undesirable trait of the VRF system. For example, fault conditions can include refrigerant leakage, outdoor unit frost, clogging of an indoor fan, a dirty indoor filter, a dirty heat exchanger, a dirty outdoor fan, motor demagnetization, compressor oil leakage, etc. Fault conditions may result in imperfect efficiency of compressor that can affect costs (e.g., operational costs) associated with the VRF system.

As described in greater detail throughout FIGS. 19-22, an AI model can be used to predict the fault conditions of components in VRF system. In particular, the AI model can be used to predict fault conditions associated with a compressor of the VRF system. For each different fault situation, a percentage can be used to represent an amount the failure influences the VRF system. Accordingly, there may be two models (e.g., two RNNs) for detailed fault classification. The first model can be structured to output a fault classification describing what fault conditions, if any, exist in the VRF system. Effectively, the first model can be structured to determine/predict what fault (if any) the VRF system and/or components therein are experiencing. The second model can be structured to determine/predict a severity of each fault condition using a representative percentage and/or some other metric. The severity of a fault can effectively indicate a degree of influence that the fault has on the VRF system. For example, an output of a refrigerant leakage index of 0.5 (i.e., 50%) by the second model may indicate that 50% of refrigerant is being lost through leakage. As another example, an output of an indoor fan clogging index of 0.9 by the second model may indicate that an indoor fan is 90% clogged. In some embodiments, the first model and the second model are combined into a single model. In this case, the single model may output a fault classification indicating fault conditions within the VRF system and a corresponding metric(s) representing a severity associated with each condition. In some embodiments, the second model may not be utilized. In this case, a fault classification generated by the first model is utilized in determining a corrective action to initiate. In other words, if the second model is not utilized, a corrective action may be initiated based on the fault classification and irrespective of an amount the fault impacts the VRF system.

As should be appreciated, the systems and methods shown and described below throughout FIGS. 19-22 can be applied to a variety of building systems and may not necessarily be limited to VRF systems. For example, the systems and methods can be similarly applied to other HVAC systems of a building.

Referring now to FIG. 19, a block diagram of a VRF system 1900 is shown, according to some embodiments. VRF system 1900 can be illustrative of an example VRF system to which an AI model can be applied to generate predictions of fault classifications. In particular, VRF system 1900 can illustrate sources of input values that may be used in generating fault classifications via the AI model. In some embodiments, VRF system 1900 is similar to and/or the same as VRF system 600 as described with reference to FIGS. 6A and 6B.

VRF system 1900 is shown to include an outdoor unit (ODU) 1902 and an indoor unit (IDU) 1904. ODU 1902 is shown to include a compressor 1906 and a heat exchanger 1908. Compressor 1906 can receive electric current I from an inverter via a compressor power line 1914. I_invertermay be any type of electric power signal such as, for example, AC. I_invertercan power compressor 1906. In some embodiments, compressor 1906 operates a motor respective of a frequency and an amplitude of I_inverteras described in greater detail above with reference to FIGS. 13A-18.

Compressor 1906 can also receive a fluid (e.g., a refrigerant) from heat exchanger 1908. To obtain the fluid, compressor 1906 can suction the fluid from heat exchanger 1908 via a suction line 1922. The suction of the fluid can be associated with a suction pressure P_swhich may be measured in a suitable pressure metric such as pascals. In some embodiments, P_sis measured by a pressure sensor associated with suction line 1922.

As a result of suctioning the fluid via suction line 1922, compressor 1906 can compress the fluid and output the compressed fluid to IDU 1904 via discharge line 1916. The output of the fluid via discharge line 1916 can be associated with a discharge pressure P_dand a discharge temperature T_d. In some embodiments, P_dand T_dare measured by a pressure sensor and a temperature sensor associated with discharge line 1916, respectively.

A fan 1912 of IDU 1904 can receive the compressed fluid (e.g., a compressed refrigerant) via discharge line 1916 as well as an indoor fan current I_fan,ivia an indoor fan power line 1918. Using the compressed fluid, fan 1912 can operate to provide heating/cooling to a space. Specifically, heat emitted by the compressed fluid can be pushed using fan 1912 into the space. In some embodiments, IDU 1904 includes multiple fans 1912 that provide heating/cooling to various locations within the space.

IDU 1904 can recirculate the fluid back to ODU 1902 via a return line 1920. Specifically, the fluid can be provided back to heat exchanger 1908. A fan 1910, powered by outdoor fan current I_fan,ovia outdoor fan power line 1924, can provide air to heat exchanger 1908. In this way, some heat may be dissipated from the fluid to an external space. Finally, the fluid can be provided back to compressor 1906 via suction line 1922.

The variables shown in VRF system 1900 and as described above (e.g., I_fan,i, P_s, P_d, T_d, etc.) can be measured by respective sensors attached, inside, or otherwise associated with lines 1914-1924. Measurements of said variables can be provided back to a controller that utilizes an AI model to identify faults associated with VRF system 1900. For example, the variables may be provided to VRF fault controller 2000 as described in detail below with reference to FIG. 20. As should be appreciated, the variables and components shown in FIG. 19 are provided purely for sake of example. VRF system 1900 may include different components than as shown in FIG. 19. Likewise, variables other than or in addition to those shown in FIG. 19 may be provided to the controller.

Referring now to FIG. 20, a block diagram of a VRF fault controller 2000 for predicting faults in a VRF system is shown, according to some embodiments. VRF fault controller 2000 can be used to generate predictions regarding fault associated with a VRF system (e.g., VRF system 600 as described with reference to FIG. 16, VRF system 1900 as described with reference to FIG. 19, etc.). VRF fault controller 2000 can utilize one or more AI models to generate a fault classification and a corresponding severity of fault conditions identified by the fault classification. In some embodiments, VRF fault controller 2000 is integrated with other controllers described herein as a single controller. For example, functionality of VRF fault controller 2000 may be integrated with oil management controller 800 as described with reference to FIG. 8 and/or compressor vibration controller 1310 as described with reference to FIG. 16.

VRF fault controller 2000 is shown to include a communications interface 2008 and a processing circuit 2002. Communications interface 2008 may include wired or wireless interfaces (e.g., jacks, antennas, transmitters, receivers, transceivers, wire terminals, etc.) for conducting data communications with various systems, devices, or networks. For example, communications interface 2008 may include an Ethernet card and port for sending and receiving data via an Ethernet-based communications network and/or a Wi-Fi transceiver for communicating via a wireless communications network. Communications interface 2008 may be configured to communicate via local area networks or wide area networks (e.g., the Internet, a building WAN, etc.) and may use a variety of communications protocols (e.g., BACnet, IP, LON, etc.).

Communications interface 2008 may be a network interface configured to facilitate electronic data communications between VRF fault controller 2000 and various external systems or devices (e.g., training data sources 2018, sensors 2020, a user device 2024, etc.). For example, VRF fault controller 2000 may provide notifications describing a fault classification of equipment 2022 to user device 2024 via communications interface 2008.

Processing circuit 2002 is shown to include a processor 2004 and memory 2006. Processor 2004 may be a general purpose or specific purpose processor, an application specific integrated circuit (ASIC), one or more field programmable gate arrays (FPGAs), a group of processing components, or other suitable processing components. Processor 2004 may be configured to execute computer code or instructions stored in memory 2006 or received from other computer readable media (e.g., CDROM, network storage, a remote server, etc.).

Memory 2006 may include one or more devices (e.g., memory units, memory devices, storage devices, etc.) for storing data and/or computer code for completing and/or facilitating the various processes described in the present disclosure. Memory 2006 may include random access memory (RAM), read-only memory (ROM), hard drive storage, temporary storage, non-volatile memory, flash memory, optical memory, or any other suitable memory for storing software objects and/or computer instructions. Memory 2006 may include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described in the present disclosure. Memory 2006 may be communicably connected to processor 2004 via processing circuit 2002 and may include computer code for executing (e.g., by processor 2004) one or more processes described herein. In some embodiments, one or more components of memory 2006 are part of a singular component. However, each component of memory 2006 is shown independently for ease of explanation.

Memory 2006 is shown to include training data collector 2010. In some embodiments, training data collector 2010 is similar to and/or the same as training data collector 1610 as described with reference to FIG. 16. Training data collector 2010 can be configured to obtain training data for use in training an AI model from training data sources 2018. Training data sources 2018 can include a variety of sources that can provide training data to VRF fault controller 2000. For example, training data sources 2018 may include a cloud database, a database on-site of a building, a user device, building equipment, etc. In some embodiments, training data sources 2018 are similar to and/or the same as training data sources 1618.

The training data collected by training data collector 2010 can include any information relevant for training an AI model to predict what faults may exist in a VRF system and, in some cases, a severity of said faults (e.g., a percentage value of the fault with 0% representing no issue exists and 100% representing a catastrophic failure). For example, the training data may include values of compressor speed, ambient temperature, discharge temperature, suction pressure, discharge pressure, indoor fan mode and VRF mode (e.g., heating or cooling), outdoor fan step, etc. The training data may also include fault classifications associated with the values. In this case, the fault classifications may be entered into the training data by users (e.g., maintenance personnel, a building owner, etc.) respective of the values. Alternatively or additionally, the fault classifications may be automatically entered into the training data based on equipment feedback indicating when certain devices have failed and/or are otherwise in fault states. In some embodiments, the fault classifications are entered into the training data via some other source of fault classifications.

In some embodiments, training data collector 2010 utilizes a simulation framework to generate some and/or all of the training data necessary to train the one or more AI models. In this case, the simulation framework can execute simulations of test cases representing operation of a VRF system by using a closed loop functional mock-up unit (FMU) model of the VRF system. The simulation framework can model how devices degrade over time and what faults may arise as a result. The simulation framework can be executed for a variety of different loads, over varying durations, with different equipment, etc. In this way, a comprehensive collection of training data can be obtained that can be used in training the AI model to learn how to identity faults within the VRF system. Advantageously, the simulation framework can ensure that the AI model has enough training data to accurately model dynamics within the VRF system and how building devices are affected as a result. Simulation training data can be used instead of or in addition to training data gathered from training data sources 2018.

Training data collector 2010 can provide the collected training data as a training data set to a model generator 2012. In some embodiments, model generator 2012 is similar to and/or the same as model generator 1612 as described with reference to FIG. 16. Model generator 2012 can utilize the training data obtained from training data collector 2010 to train one or more AI models for predicting characteristics of fault conditions. Specifically, model generator 2012 may train a first AI model to output a fault classification respective of a set of inputs describing operation of the VRF system and a second AI model to output a severity (e.g., a percentage value) of faults identified by the first AI model have on the VRF system. In some embodiments, model generator 2012 generates a single AI model that outputs a fault classification and corresponding severities of each fault condition identified in the fault classification. In some embodiments, model generator 2012 generates a single AI model that only outputs a fault classification. An example of an AI model that outputs a fault classification is described in greater detail below with reference to FIG. 21.

A fault classification outputted by the one or more AI models can describe faults in the VRF system. For example, the fault classifications may include textual strings such as “refrigerant leakage,” “indoor fan clogged,” “outdoor unit frost,” “dirty indoor filter,” “dirty heat exchanger,” “dirty outdoor fan,” “motor demagnetization,” “compressor oil leakage,” “imperfect efficiency of compressor,” etc. As another example, the fault classification may indicate a presence of certain fault conditions using binary variables (e.g., 0 or 1) representing whether a particular fault condition exists. The fault classification can be predicted by the one or more AI models based on inputs describing the VRF system. Specifically, the one or more AI models may learn correlations between values of inputs variables such as compressor speed, ambient temperature, discharge temperature, suction pressure, discharge pressure, indoor fan mode, outdoor fan step, etc. and corresponding a fault classification. As a particular example, an AI model may learn, based on the training data, that if the suction pressure is above a particular threshold value (e.g., 20 pounds per square inch (PSI), 30 PSI, etc.), there may be refrigerant leakage within the VRF system. The AI model may be further configured to predict a percentage value of how much refrigerant is leaking from the VRF system. In the example, the AI model may predict 50% of refrigerant is being lost due to leakage based on inputs such as compressor speed, discharge temperature, etc.

The one or more AI models generated by model generator 2012 can be any type of AI model. For example, the one or more AI models may be RNNs (e.g., LSTM models), convolutional neural networks, multi-layer perceptrons, feed forward neural networks, etc. In particular, the one or more AI models may be RNNs due to a high speed and reliability of RNNs. More particularly, LSTMs may be utilized due to an ability of LSTMs to process entire sequences of time-series data and make predictions, even with lags of unknown duration between important events in the time-series. As a specific example structure, one of the AI models generated by model generator 2012 may have one sequence input layer, one bidirectional LSTM layer, one fully connected layer, one softmax layer, and one classification layer. In this way, the AI model can be used to generate a fault classification and can learn bidirectional long-term dependencies between time steps of time-series or sequence data. These dependencies can be useful for the network to learn from the complete time-series at each time step.

In some embodiments, model generator 2012 generates an individual AI model for each fault condition. In this case, each AI model can be trained to predict whether a specific fault condition is present in the VRF system. Generating individual AI model may be increase accuracy for predicting whether certain fault conditions exist. In particular, generating individual AI models can avoid overlapping data for multiple fault conditions which can affect generation of models. In some embodiments, if model generator 2012 generates individual AI models for fault conditions, model generator 2012 may combine each individual model into a composite AI model that accounts for multiple fault conditions.

Model generator 2012 can provided the one or more generated AI models to a prediction generator 2014. In some embodiments, prediction generator 2014 is similar to and/or the same as prediction generator 1614. Using the one or more AI models, prediction generator 2014 can generate predictions of what fault conditions, if any, are affecting the VRF system and, in some embodiments, a severity of the fault conditions. To generate the predictions, prediction generator 2014 can obtain values inputs required by the one or more AI models. Specifically, prediction generator 2014 may obtain measured variables from sensors 2020 and equipment feedback from equipment 2022. Sensors 2020 can include one or more sensors configured to measure certain characteristics of the VRF system. For example, sensors 2020 may include temperature sensors, pressure sensors, etc. that measure values of P_d, P_s, T_d, an ambient temperature, etc. Equipment 2022 can include any equipment of the VRF system. For example, equipment 2022 may include ODUs, IDUs, etc. The equipment feedback may include inputs required by the one or more AI models such as, for example, an indoor fan mode, an outdoor fan step, a compressor speed, a discharge temperature, a suction pressure, etc. In some embodiments, sensors 2020 are components of equipment 2022 and/or are otherwise associated with equipment 2022.

Prediction generator 2014 can pass the received values of inputs through the one or more AI models to generate predictions associated with fault characteristics of the VRF system. Prediction generator 2014 can provide said predictions to a corrective action generator 2016. Corrective action generator 2016 can be configured to initiate a variety of different corrective actions respective of what fault conditions are indicated by a fault classification included in the predictions and a severity of the faults conditions. The corrective actions initiate by corrective action generator 2016 can include any appropriate action to resolve one or more of the fault conditions identified in the predictions. For example, corrective actions may include notifying a user regarding the fault conditions by providing a notification to user device 2024, generating and providing control signals to equipment 2022, temporarily disabling some VRF devices of equipment 2022, scheduling maintenance for equipment 2022, etc.

In some embodiments, the corrective action(s) initiated by corrective action generator 2016 are based on the fault classification included in the predictions and a severity of each fault. Some fault conditions may require corrective actions that are associated with a higher cost (e.g., $), require more time on the part of maintenance technicians, etc. as compared to other corrective actions. For example, scheduling maintenance for a particular building device (e.g., an IDU) may have a higher associated cost as compared to providing a notification to user device 2024. Accordingly, corrective action generator 2016 may determine an impact a particular fault condition may have on the VRF system and initiate a corrective action respective of the determined impact. For example, if a fault classification included in the predictions received from prediction generator 2014 indicates an indoor fan filter is 50% clogged, corrective action generator 2016 may determine an impact on the VRF system to be low and may initiate a corrective action for notifying a user via user device 2024 regarding the clogging. However, if the fault classification indicates that a compressor of the VRF system is 90% inefficient, corrective action generator 2016 may determine the inefficiency of the compressor will have a high impact on operational costs and thereby initiate a corrective action to perform maintenance and/or replace the compressor.

To determine what corrective action to initiate, corrective action generator 2016 may utilize a mapping between fault conditions with corresponding severities and certain corrective actions. In particular, if the severities are given as percentages with 0% indicating no impact to the VRF system and 100% indicating severe impact to the VRF system, corrective action generator 2016 may initiate particular corrective actions based on bounds of the severities. In some embodiments, multiple corrective actions are initiated for single fault conditions. An example of a decision tree that can be utilized to determine what corrective action to initiate is provided below in Table 1.

TABLE 1 Corrective Action Decision Table Fault Severity Condition Percentage Corrective Action Refrigerant <50% Provide notification to user device Leakage ≥50% Schedule maintenance for VRF system Indoor Fan <80% Provide notification to user device Clogged ≥80% Provide control signals to indoor fan and provide notification user device Imperfect <25% No action Efficiency of ≥25% and <75% Schedule maintenance for compressor Compressor ≥75% Schedule replacement of compressor

If a corrective action is initiated, a fault condition predicted to exist can be addressed. By utilizing the one or more AI models to predict a fault classification indicating certain fault conditions and corresponding severities, appropriate corrective actions can be initiated to address certain faults predicted by the one or more AI models. Advantageously, corrective actions initiated to address certain faults can be respective of an estimated severity of the fault. In this way, expensive and/or time-consuming corrective actions can be avoided for faults that have a low impact on the VRF system.

Referring now to FIG. 21, an illustration of a neural network (NN) 2100 for predicting a fault classification of a VRF system is shown, according to some embodiments. NN 2100 can illustrate an example structure of an AI model that can be generated by model generator 2012 as described with reference to FIG. 20. Specifically, NN 2100 can represent a RNN structure for fault classification prediction.

NN 2100 is shown to include input nodes in an input layer that correspond to a set of inputs. NN 1700 is shown to include input nodes of compressor speed, ambient temperature, discharge temperature, suction pressure, discharge pressure, indoor fan mode, and outdoor fan step. As should be appreciated, the inputs shown in NN 2100 are provided purely for sake of example. The inputs to NN 2100 can be modified based on what measurements can be gathered from the VRF system, what measurements are relevant to generating a fault classification, etc.

An output of NN 2100 can include a fault classification that identifies specific faults that may exist within a VRF system. For example, NN 2100 may identify fault conditions such as refrigerant leakage, clogging of an indoor fan, frosting of an outdoor unit, a dirty indoor filter, a dirty heat exchanger, a dirty outdoor fan, motor demagnetization, compressor oil leakage, imperfect efficiency of a compressor, etc. Similar to the inputs, the fault classification that can be outputted by NN 2100 can be modified. The fault classification that can be outputted by NN 2100 can be tailored dependent on user preferences, devices within the system, what inputs are available to NN 2100, etc. Specifically, the fault classification can be tailored to only indicate whether a specific set of fault conditions exist within the VRF system.

Referring now to FIG. 22, a flow diagram of a process 2200 for predicting a fault classification for a VRF system using an AI model is shown, according to some embodiments. In some embodiments, some and/or all steps of process 2200 are performed by VRF fault controller 2000.

Process 2200 is shown to include obtaining training data associated with operational conditions of a VRF system and fault conditions of the VRF system (step 2202). The training data can be obtained from a variety of sources such as, for example, cloud databases, directly from users, equipment feedback, etc. The operational conditions included in the training data may include values of variables such as, for example, T_d, P_d, P_s, an ambient temperature, a compressor speed, an outdoor fan step (i.e., outdoor fan speed), an indoor fan speed, and a VRF mode (e.g., heating or cooling). The fault conditions included in the training data may include any sort of fault condition that can be experienced by the VRF system. For example, the fault conditions may include refrigerant leakage, clogging of an indoor fan, frosting of an outdoor unit, a dirty indoor filter, a dirty heat exchanger, a dirty outdoor fan, motor demagnetization, compressor oil leakage, imperfect efficiency of a compressor, etc. In some embodiments, each fault condition included in the training data also includes an estimated severity (e.g., a percentage value, a value on a 1 to 10 scale, etc.). Estimated severities may be manually estimated by users, automatically based on changing operational costs indicated by the training data, etc. In some embodiments, the training data is time-series data such that relationships between changes in operational conditions and certain fault conditions can be identified. In some embodiments, step 2202 is performed by training data collector 2010.

Process 2200 is shown to include training an AI model for predicting a fault classification for the VRF system based on the training data (step 2204). The AI model trained in step 2204 may be any type of AI model such as an RNN. The fault classification predicted by the AI model can indicate whether one or more fault conditions exist within the VRF system and corresponding severities of any existing faults. In some embodiments, two AI models are trained in step 2204 such that a first AI model predicts a fault classification indicating what, if any, fault conditions exist based on input data and a second AI model predicts the severities of the fault conditions that do exist. An example of the AI model that can be generated in step 2204 is described above with reference to FIG. 21. In some embodiments, step 2204 is performed by model generator 2012.

Process 2200 is shown to include obtaining data associated with operation of the VRF system (step 2206). The data obtained in step 2206 can include values of inputs required by the one or more AI models generated in step 2204. For example, the data may include values of T_d, P_d, P_s, an ambient temperature, a compressor speed, an outdoor fan step, an indoor fan speed, and a VRF mode. The data can be obtained from a variety of sources such as sensors, equipment feedback, from users, etc. In some embodiments, step 2206 is performed by prediction generator 2014.

Process 2200 is shown to include using the AI model to predict a fault classification for the VRF system based on the obtained data (step 2208). The fault classification can indicate what fault conditions, if any, are identified to exist within the VRF system. The fault classification may indicate existing fault conditions through any appropriate method such as, for example, by including textual strings of the existing fault conditions, by including an array of binary variables indicating the existence or nonexistence of certain fault conditions, etc. In some embodiments, step 2208 also includes using the AI model (or a second AI model) to predict severities of each existing fault condition based on the obtained data. In some embodiments, step 2208 is performed by prediction generator 2014.

Process 2200 is shown to include determining a corrective action based on the fault classification outputted by the AI model (step 2210). In some embodiments, the corrective action is determined based on what fault conditions, if any, are identified by the fault classification. In some embodiments, the corrective action is further determined based on an indicated severity of each fault condition identified by the fault classification. The corrective action can include any action that addresses the fault conditions indicated by the fault classification. For example, the corrective action may include notifying a user regarding the identified fault conditions and corresponding severities, scheduling maintenance/replacement for VRF equipment, generating and transmitting control signals to the VRF equipment, logging the fault conditions and corresponding severities to a database, etc. In some embodiments, multiple corrective actions are determined in step 2210. In some embodiments, step 2210 is performed by corrective action generator 2016.

Process 2200 is shown to include initiating the corrective action (step 2212). Step 2212 can include performing any steps/processes necessary to ensure the corrective action is successfully executed. For example, if the corrective action is notifying a user, step 2412 may include generating a notification and communicating the notification using a communications interface. As another example, if the corrective action is disabling a VRF device, step 2412 may include generating a control signal that turns off the VRF device and providing the control signal to the VRF device. In some embodiments, step 2212 is performed by corrective action generator 2016.

Motor Efficiency Prediction

Referring generally to FIGS. 23-24, systems and methods for predicting efficiency of a motor in a VRF system using an AI model are shown and described, according to some embodiments. In some embodiments, the systems and methods described below can be integrated with any and/or all of the systems and methods described above throughout FIGS. 8-22.

Efficiency of a motor for a compressor can be defined based on an input power and an output power of the motor. Specifically, motor efficiency η_motorcan be given by the following equation:

$η_{motor} = \frac{φ_{out, motor}}{φ_{i n, motor}}$

where φ_out,motoris an output power of the motor and φ_in,motoris an input power of the motor. In some embodiments, the motor is powered based on a received signal from an inverter. Efficiency of the inverter η_invertercan also be determined and can be defined based on the following equation:

$η_{inverter} = \frac{φ_{out, inverter}}{φ_{i n, inverter}}$

where φ_out,inverteris an output power of the inverter and φ_in,inverteris an input power of the inverter. In this case, the output power of the inverter should be the same (or nearly the same) as the input power of the motor (i.e., φ_out,inverter=φ_in,motor).

Changes in motor efficiency may be an important index to detect faults and inference reasons leading to efficiency reduction. The detection of motor efficiency can allow a system (e.g., a VRF system) to identify abnormal information early (e.g., prior to catastrophic failure). Based on detected changes in motor efficiency, operation of the motor can be adjusted accordingly. Specifically, detected changes in motor efficiency may result in making offset compensations to the motor drive and/or stopping the machine entirely. As such, predictions of motor efficiency can reduce overall hardware costs (e.g., operational costs, maintenance costs, etc.) and can improve reliability of the motor drive.

As described in greater detail below, an AI model can be utilized to predict motor efficiency and, optionally, efficiency of an inverter to detect abnormal conditions of the motor drive system. The AI model used to predict efficiency can be any type of AI model. In particular, the AI model for efficiency prediction may be a linear RNN model due to a case-dependent nature of data describing efficiency. If a linear RNN model is used, the predicted efficiency may be an instantaneous efficiency of the motor.

The predicted motor efficiency may be indicative of a probability that the motor will fail. As the motor becomes more inefficient, the probability that the motor will fail may increase. In this way, the predicted motor efficiency can be used to predict when the motor is expected to fail. In some embodiments, a threshold efficiency value can be used to define failure of the motor. For example, the motor may be considered to have failed if the predicted efficiency is below 50%, 30%, 20%, etc. In this way, a trend in changes to motor efficiency can be identified and projected forward in time to predict a time when the motor efficiency is expected to fall below the threshold efficiency and the motor is considered to have failed. Similar predictions can be made for inverter efficiency and inverter failure.

Based on the predicted efficiency of the motor (and/or predicted times when the motor is expected to fail), a determination can be made regarding what corrective actions, if any, should be initiated to address the efficiencies. For example, if the motor is predicted to be operating at full efficiency (i.e., 100% efficiency), a determined corrective action may be to log the predicted efficiency. As another example, if the motor if predicted to be operating at 50% efficiency, a corrective action that schedules maintenance for the motor may be scheduled in order to increase the efficiency of the motor.

Referring now to FIG. 23, a block diagram of a motor efficiency controller 2300 is shown, according to some embodiments. Motor efficiency controller 2300 can be configured to predict efficiency of a motor used in a building system. Specifically, motor efficiency controller 2300 can be configured to predict efficiency of a motor of a compressor in a VRF system. However, predictions of motor efficiency can be similarly applied to motors in other systems (e.g., other HVAC systems). In some embodiments, functionality of motor efficiency controller 2300 is combined with one or more controllers described herein (e.g., oil management controller 800, compressor vibration controller 1310, and/or VRF fault controller 2000).

Motor efficiency controller 2300 is shown to include a communications interface 2308 and a processing circuit 2302. Communications interface 2308 may include wired or wireless interfaces (e.g., jacks, antennas, transmitters, receivers, transceivers, wire terminals, etc.) for conducting data communications with various systems, devices, or networks. For example, communications interface 2308 may include an Ethernet card and port for sending and receiving data via an Ethernet-based communications network and/or a Wi-Fi transceiver for communicating via a wireless communications network. Communications interface 2308 may be configured to communicate via local area networks or wide area networks (e.g., the Internet, a building WAN, etc.) and may use a variety of communications protocols (e.g., BACnet, IP, LON, etc.).

Communications interface 2308 may be a network interface configured to facilitate electronic data communications between motor efficiency controller 2300 and various external systems or devices (e.g., training data sources 2318, sensors 2320, a motor 2322, a user device 2324, etc.). For example, motor efficiency controller 2300 may provide notifications describing motor efficiency predictions to user device 2324 via communications interface 2308.

Processing circuit 2302 is shown to include a processor 2304 and memory 2306. Processor 2304 may be a general purpose or specific purpose processor, an application specific integrated circuit (ASIC), one or more field programmable gate arrays (FPGAs), a group of processing components, or other suitable processing components. Processor 2304 may be configured to execute computer code or instructions stored in memory 2306 or received from other computer readable media (e.g., CDROM, network storage, a remote server, etc.).

Memory 2306 may include one or more devices (e.g., memory units, memory devices, storage devices, etc.) for storing data and/or computer code for completing and/or facilitating the various processes described in the present disclosure. Memory 2306 may include random access memory (RAM), read-only memory (ROM), hard drive storage, temporary storage, non-volatile memory, flash memory, optical memory, or any other suitable memory for storing software objects and/or computer instructions. Memory 2306 may include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described in the present disclosure. Memory 2306 may be communicably connected to processor 2304 via processing circuit 2302 and may include computer code for executing (e.g., by processor 2304) one or more processes described herein. In some embodiments, one or more components of memory 2306 are part of a singular component. However, each component of memory 2306 is shown independently for ease of explanation.

Memory 2306 is shown to include a training data collector 2310. In some embodiments, training data collector 2310 is similar to and/or the same as training data collector 1610 as described with reference to FIG. 16 and/or training data collector 2010 as described with reference to FIG. 20. Training data collector 2310 can obtain training data associated with predicting motor efficiency from training data sources 2318. Training data sources 2318 can include any appropriate source of training data such as cloud databases, local storage, user inputs, mobile devices, equipment feedback, etc. The training data can include training data that describes operation of motor 2322 under various conditions. For example, the training data may include values of a rotational speed of motor 2322, a power supplied by an inverter 2328 to motor 2322, required heating/cooling loads, etc. Likewise, the training data can include estimated efficiency values of the motor. The efficiency values may be estimated by users, automatically based on detected changes to costs associated with motor 2322 (e.g., changes in operational costs where higher costs indicate lower efficiency), and/or estimated via some other process.

In some embodiments, training data collector 2310 utilizes a simulation framework to generate some and/or all of the training data. The simulation framework may model how motor 2322 operates based on various operating conditions (e.g., temperatures, heating/cooling duties, etc.). Likewise, the simulation framework can model how efficiency of motor 2322 degrades over time as a result of operation. In this way, training data representative of the operation and degradation of motor 2322 can be generated and utilized for training the AI model.

Training data collector 2310 can provide the collected training data as a training data set to a model generator 2312. In some embodiments, model generator 2312 is similar to and/or the same as model generator 1612 and/or model generator 2012. Using the training data, model generator 2312 can generate an AI model (e.g., a linear RNN) that can be used to predict efficiency values of a motor in a VRF system (e.g., VRF system 600, VRF system 1900, etc.). In particular, the AI model may output a percentage representative of a percentage of maximum efficiency at which motor 2322 is operating. In this case, maximum efficiency can be represented by a situation where an input power to motor 2322 is equal to an output power of motor 2322 (i.e., φ_out,inverter=φ_in,inverter). In some embodiments, model generator 2312 also generates the AI model such that the AI model also outputs a prediction of inverter efficiency. In this case, inverter efficiency can be modeled similarly to motor efficiency and can be based on operating conditions of inverter 2328 (e.g., required heating/cooling duties, ambient temperature, etc.). In some embodiments, model generator 2312 generates separate AI models for predicting motor efficiency and inverter efficiency.

With regard to an AI model structured to predict efficiency of motor 2322, the AI model may include inputs such as a rotational speed of motor 2322, a power provided to motor 2322 (e.g., by inverter 2328), an ambient temperature, required heating/cooling loads, etc. Based on said inputs, the AI model can output a prediction of motor efficiency (e.g., as a percentage of maximum efficiency).

Model generator 2312 can provide the AI model to prediction generator 2314. In some embodiments, prediction generator 2314 is similar to and/or the same as prediction generator 1614 and/or prediction generator 2014. Prediction generator 2314 can utilize the AI model and received values of input variables to generate predictions of motor efficiency. In terms of input variables, prediction generator 2314 may receive values of measured variables from sensors 2320. Sensors 2320 can include any type of sensor (e.g., temperature sensors, pressure sensors, etc.) that can measure values of inputs required by the AI model. For example, sensors 2320 may include temperature sensors that provide values of an ambient temperature to prediction generator 2314. In some embodiments, sensors 2320 includes third party sources that provide needed input values. For example, sensors 2320 may include a weather service that provides ambient temperature values.

Prediction generator 2314 can also obtain values of input variables based on equipment feedback from equipment 2326 including motor 2322 and inverter 2328. The equipment feedback can include values of variables such as, for example, a rotational speed of motor 2322, a power supply current, voltage, and input power provided to motor 2322 by inverter 2328, etc.

Using the obtained values of input variables, prediction generator 2314 can pass the obtained input variables through the AI model to generate predictions of motor efficiency. Said predictions of motor efficiency can be provided to a corrective action generator 2316. In some embodiments, corrective action generator 2316 is similar to and/or the same as corrective action generator 1616 and/or corrective action generator 2016.

Corrective action generator 2316 can be configured to determine what, if any, corrective actions should be initiated based on predictions of motor efficiency. In particular, corrective action generator 2316 can determine whether actions should be taken to address reductions in motor efficiency. Corrective actions that can be initiated by corrective action generator 2316 may include, for example, alerting a user about the predicted motor efficiency via user device 2324, adjusting operation of motor 2322 via control signals, disabling motor 2322, scheduling maintenance/replacement of motor 2322, logging predicted values of motor efficiency to a database, etc.

To determine what corrective action to initiate, corrective action generator 2316 may correspond certain efficiency ranges with certain corrective actions. For example, correction action generator 2316 may associate efficiency values between 75% and 100% with a corrective action that notifies a user of the predicted efficiency, efficiency values between 50% and 75% with a corrective action to adjust operation of motor 2322 via controls signals, and efficiency values between 0% and 50% with a corrective action to schedule maintenance and/or replacement of motor 2322. In some embodiments, corrective action generator 2316 utilizes some other determination for determining what corrective action to initiate based on predicted efficiency values.

As a result of initiating a corrective action, inefficiency of motor 2322 can be addressed. In this way, a frequency of situations where motor 2322 is operated under high inefficiency can be avoided. This can increase reliability of a VRF system, reduce overall costs, and can simplify general upkeep of the VRF system.

Referring now to FIG. 24, a flow diagram of a process 2400 for predicting an efficiency of a motor in a VRF system using an AI model is shown, according to some embodiments. It should be appreciated that while process 2400 is described with reference to a motor of a VRF system, process 2400 can be similarly applied for generating efficiency predictions associated with an inverter of the VRF system. In some embodiments, some and/or all steps of process 2400 are performed by motor efficiency controller 2300.

Process 2400 is shown to include obtaining training data describing operational conditions of a motor of a VRF system and associated efficiency values of the motor (step 2402). The training data can be obtained from a variety of training data sources such as, for example, cloud databases, local storage of a building, based on user input, via feedback from equipment, etc. The training data can include any information relevant for training an AI model to predict efficiency values of the motor. For example, the training data may include values of rotational speed of the motor during operation, ambient temperatures, heating/cooling duties, a current provided to the motor, a voltage provided to the motor, power provided to the motor, estimated efficiency values of the motor, etc. In some embodiments, step 2402 is performed by training data collector 2310.

Process 2400 is shown to include training an artificial intelligence (AI) model for predicting efficiency of the motor based on the training data (step 2404). The AI model can be trained to learn a correlation between a set of variables affecting operation of the motor and corresponding efficiency values of the motor. The AI model trained in step 2404 may be any of a variety of different AI models. For example, the AI model may be a linear RNN for evaluating efficiency of the motor. In some embodiments, step 2404 is performed by model generator 2312.

Process 2400 is shown to include obtaining data associated with operation of the motor (step 2406). The data obtained in step 2406 may be based on the motor during actual operation when predictions of motor efficiency are desired. In particular, the data obtained in step 2406 can include values of input variables to the AI model generated in step 2404. As such, the data can include values of a rotational speed of the motor, an ambient temperature, input power to the motor, etc. In some embodiments, the data may be obtained from some and/or all of the same sources used to obtain the training data in step 2402. In some embodiments, step 2406 is performed by prediction generator 2314.

Process 2400 is shown to include using the AI model to predict an efficiency of the motor based on the obtained data (step 2408). In step 2408, the data obtained in step 2406 can be passed to the AI model as input to generate predictions regarding efficiency of the motor. In some embodiments, the predicted efficiency of the motor is provided as a percentage of maximum efficiency. In some embodiments, the predicted efficiency is provided as another metric (e.g., a rating of “good,” “medium,” or “poor”). In some embodiments, step 2408 is performed by prediction generator 2314.

Process 2400 is shown to include determining a corrective action based on the predicted motor efficiency outputted by the AI model (step 2410). The corrective action can be determined based on an estimated degradation state of the motor. For example, a high efficiency value (e.g., >80% efficiency, >90% efficiency, etc.) may indicate the motor is not significantly degraded. Accordingly, the corrective action determined in step 2410 may be a corrective action associated with a low cost (e.g., monetary cost, time cost, etc.) such as dispatching a notification to a user device or logging the estimated efficiency value to a database. However, if the estimated efficiency is low (e.g., <50% efficiency, <40% efficiency, etc.), the corrective action determined in step 2410 may be a corrective action associated with a higher cost such as scheduling maintenance for the motor, disabling the motor, etc. In this way, the corrective action determined in step 2410 can be dynamically adjusted based on the estimated efficiency of the motor. In some embodiments, step 2410 is performed by corrective action generator 2316.

Process 2400 is shown to include initiating the corrective action (step 2412). Based on the corrective action determined in step 2410, step 2412 can include determining how to initiate the corrective action and executing the steps necessary for the corrective action to be completed. For example, if the corrective action is notifying a user, step 2412 may include generating a notification and communicating the notification using a communications interface. As another example, if the corrective action is disabling the motor, step 2412 may include generating a control signal that turns off the motor and providing the control signal to the motor. In some embodiments, step 2412 is performed by corrective action generator 2316.

Examples of Neural Network Implementations

Referring generally to FIGS. 25A and 25B, illustrations associated with AI model structures that can be for generating predictions are shown, according to some embodiments. The AI model structures described below can be utilized in any of the various controllers and/or methods described herein. For example, the AI model structures provided in FIGS. 25A-25B may be utilized by compressor vibration controller 1310, VRF fault controller 2000, etc. It should be appreciated that the AI model structures provided in FIGS. 25A and 25B are provided purely for sake of example and is not intended to be limiting on AI model structures that can be utilized for generating predictions.

Referring now to FIG. 25A, an illustration of an RNN structure 2500 is shown, according to some embodiments. In some embodiments, RNN structure 2500 is similar to and/or the same as RNN structure 900 as described with reference to FIG. 9A. In some embodiments, RNN structure 2500 may illustrate a high-level view of an LSTM model. LSTM models are a specific artificial RNN architecture that can be used in the field of deep learning. LSTM models can classify and process entire sequences of time-series data and make predictions. Advantageously, LSTM models can generate prediction even with lags of unknown duration between important events in the time-series data. An LSTM model can include various layers such as, for example, a sequence input layer, one or more drop out layers, one or more fully connected layers, one or more LSTM layers, an output layer, etc.

As shown in FIG. 25A, RNN structure 2500 can be represented by both a condensed model structure and an “unfolded” model structure. The unfolded model structure can illustrate in greater detail how RNN structure 2500 saves information over time which can affect outputs as information is not completely lost between time steps. In FIG. 25A, x can represent an input to the RNN, U can represent input parameters applied to x, o can represent an output of the RNN, W can represent output parameters applied to an output of h to generate o, h can represent a primary block of the RNN that includes weights and activation functions of the RNN, and v can represent information communicated between time steps.

Referring now to FIG. 25B, an illustration of an LSTM model structure 2550 is shown, according to some embodiments. LSTM model structure 2550 can illustrate how information is saved between time steps in an RNN. LSTM model structure 2550 is shown to include functions f , g, i, and o which can be used to generate outputs to the block shown in FIG. 25B. LSTM model structure 2550 is further shown to include a forget gate, an update gate, and an output gate. The forget gate can be configured to eliminate non-relevant data from being considered and remembered for future time steps in a time sequence. The update gate can apply some operation to combine input information to account for changes in the data. Finally, the output gate can decide what information is passed as output to the next time step. LSTM model structure 2550 can include multiple blocks that pass information associated with a particular time step in a time sequence to the next time step. Advantageously, this structure allows information to be retained and not lost between time steps, thereby increasing an accuracy of prediction for time-series data.

Configuration of Exemplary Embodiments

The construction and arrangement of the systems and methods as shown in the various exemplary embodiments are illustrative only. Although only a few embodiments have been described in detail in this disclosure, many modifications are possible (e.g., variations in sizes, dimensions, structures, shapes and proportions of the various elements, values of parameters, mounting arrangements, use of materials, colors, orientations, etc.). For example, the position of elements can be reversed or otherwise varied and the nature or number of discrete elements or positions can be altered or varied. Accordingly, all such modifications are intended to be included within the scope of the present disclosure. The order or sequence of any process or method steps can be varied or re-sequenced according to alternative embodiments. Other substitutions, modifications, changes, and omissions can be made in the design, operating conditions and arrangement of the exemplary embodiments without departing from the scope of the present disclosure.

The present disclosure contemplates methods, systems and program products on any machine-readable media for accomplishing various operations. The embodiments of the present disclosure can be implemented using existing computer processors, or by a special purpose computer processor for an appropriate system, incorporated for this or another purpose, or by a hardwired system. Embodiments within the scope of the present disclosure include program products comprising machine-readable media for carrying or having machine-executable instructions or data structures stored thereon. Such machine-readable media can be any available media that can be accessed by a general purpose or special purpose computer or other machine with a processor. By way of example, such machine-readable media can comprise RAM, ROM, EPROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to carry or store desired program code in the form of machine-executable instructions or data structures and which can be accessed by a general purpose or special purpose computer or other machine with a processor. Combinations of the above are also included within the scope of machine-readable media. Machine-executable instructions include, for example, instructions and data which cause a general purpose computer, special purpose computer, or special purpose processing machines to perform a certain function or group of functions.

Although the figures show a specific order of method steps, the order of the steps may differ from what is depicted. Also two or more steps can be performed concurrently or with partial concurrence. Such variation will depend on the software and hardware systems chosen and on designer choice. All such variations are within the scope of the disclosure. Likewise, software implementations could be accomplished with standard programming techniques with rule based logic and other logic to accomplish the various connection steps, processing steps, comparison steps and decision steps.

Claims

1-40. (canceled)

41. A controller for predicting faults in a heating, ventilation, or air conditioning (HVAC) system, the controller comprising a processing circuit configured to:

analyze operating data for the HVAC system using a machine learning model to predict a fault classification for the HVAC system, the fault classification identifying a fault condition affecting the HVAC system;

identify a HVAC device of the HVAC system associated with the fault condition; and

automatically initiate a corrective action to address the fault condition responsive to identifying the HVAC device and the fault condition.

42. The controller of claim 41, wherein:

the fault classification includes a severity metric associated with the fault condition, the severity metric indicating a degree of influence that the fault condition has on the HVAC system; and

the corrective action is determined based on both the fault condition and the severity metric associated with the fault condition.

43. The controller of claim 42, wherein the processing circuit is configured to:

automatically initiate a first corrective action in response to a value of the severity metric being below a severity threshold; and

automatically initiate a second corrective action in response to the value of the severity metric being above the severity threshold.

44. The controller of claim 43, wherein:

the first corrective action comprises providing a notification to a user device; and

the second corrective action comprises scheduling maintenance for the HVAC system or replacement of the HVAC device associated with the fault condition.

45. The controller of claim 42, wherein the corrective action comprises taking no action in response to a value of the severity metric being below a severity threshold.

46. The controller of claim 41, wherein

the machine learning model is a recurrent neural network (RNN) model; and

analyzing the operating data comprises providing a time series of values of the operating data as an input to the RNN model and obtaining a prediction of the fault classification as an output of the RNN model.

47. The controller of claim 41, the processing circuit further configured to generate the machine learning model using a set of simulated training data obtained from a simulation model of the HVAC system.

48. The controller of claim 41, wherein the fault classification identifies a plurality of fault conditions affecting the HVAC system, the plurality of fault conditions associated with a plurality of HVAC devices of the HVAC system.

49. The controller of claim 41, wherein the fault condition comprises at least one of:

leakage of a refrigerant;

frosting of an outdoor unit;

clogging of an indoor fan;

clogging of an indoor filter;

clogging of a heat exchanger;

clogging of an outdoor fan;

demagnetization of a motor; or

leakage of oil from a compressor.

50. The controller of claim 41, wherein the machine learning model is a first machine learning model and the processing circuit is configured to:

use the first machine learning model to predict the fault classification for the HVAC system; and

use a second machine learning model to predict a severity of the fault condition identified by the fault classification.

51. A method for predicting faults in a heating, ventilation, or air conditioning (HVAC) system, the method comprising:

analyzing operating data for the HVAC system using a machine learning model to predict a fault classification for the HVAC system, the fault classification identifying a fault condition affecting the HVAC system;

identifying a HVAC device of the HVAC system associated with the fault condition; and

automatically initiating a corrective action to address the fault condition responsive to identifying the HVAC device and the fault condition.

52. The method of claim 51, wherein:

the fault classification includes a severity metric associated with the fault condition, the severity metric indicating a degree of influence that the fault condition has on the HVAC system; and

the corrective action is determined based on both the fault condition and the severity metric associated with the fault condition.

53. The method of claim 52, comprising:

automatically initiating a first corrective action in response to a value of the severity metric being below a severity threshold; and

automatically initiating a second corrective action in response to the value of the severity metric being above the severity threshold.

54. The method of claim 53, wherein:

the first corrective action comprises providing a notification to a user device; and

the second corrective action comprises scheduling maintenance for the HVAC system or replacement of the HVAC device associated with the fault condition.

55. The method of claim 52, wherein the corrective action comprises taking no action in response to a value of the severity metric being below a severity threshold.

56. The method of claim 51, wherein

the machine learning model is a recurrent neural network (RNN) model; and

analyzing the operating data comprises providing a time series of values of the operating data as an input to the RNN model and obtaining a prediction of the fault classification as an output of the RNN model.

57. The method of claim 51, comprising generating the machine learning model using a set of simulated training data obtained from a simulation model of the HVAC system.

58. The method of claim 51, wherein the fault classification identifies a plurality of fault conditions affecting the HVAC system, the plurality of fault conditions associated with a plurality of HVAC devices of the HVAC system.

59. The method of claim 51, wherein the fault condition comprises at least one of:

leakage of a refrigerant;

frosting of an outdoor unit;

clogging of an indoor fan;

clogging of an indoor filter;

clogging of a heat exchanger;

clogging of an outdoor fan;

demagnetization of a motor; or

leakage of oil from a compressor.

60. The method of claim 51, wherein the machine learning model is a first machine learning model and the method comprises:

using the first machine learning model to predict the fault classification for the HVAC system; and

using a second machine learning model to predict a severity of the fault condition identified by the fault classification.