AIR CONDITIONER DIAGNOSIS SYSTEM AND LEARNING DEVICE

Info

Publication number: 20230243535
Type: Application
Filed: Jun 5, 2020
Publication Date: Aug 3, 2023
Inventor: Hitoshi NAKAMURA (Tokyo)
Application Number: 17/908,459

Abstract

An object of the present disclosure is to provide an air conditioner diagnosis system and a learning device that can reduce the frequency of dispatching personnel to the sites. The air conditioner diagnosis system according to the present disclosure is an air conditioner diagnosis system that remotely diagnoses a plurality of air conditioners. The diagnosis system includes an inference device to infer a trouble cause of an air conditioner from input data including model data of the air conditioner and operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and a display device to display the inferred trouble cause.

Description

Description

TECHNICAL FIELD

The present disclosure relates to an air conditioner diagnosis system and a learning device.

BACKGROUND ART

Air conditioner diagnosis devices are conventionally known. For example, an air conditioner controller self-diagnosis device described in PTL 1 determines malfunction of a display, a remote control, and the like with program instructions when a self-diagnosis switch is turned on. The self-diagnosis switch can be turned on only by the personnel of a service company on site.

CITATION LIST Patent Literature

PTL 1: Japanese Patent Laying-Open No. S58-178136

SUMMARY OF INVENTION Technical Problem

Unfortunately, in PTL 1, trouble causes, parts to be replaced, or appropriate workers for repair are unable to be identified unless the personnel of a service company visits sites.

An object of the present disclosure is therefore to provide an air conditioner diagnosis system and a learning device that can reduce the frequency of dispatching personnel to sites.

Solution to Problem

An air conditioner diagnosis system according to the present disclosure is an air conditioner diagnosis system that remotely diagnoses a plurality of air conditioners. The diagnosis system includes: an inference device to infer a trouble cause of an air conditioner, from input data including model data of the air conditioner and operation data of the air conditioner at least at a time when a trouble of the air conditioner occurs; and a display device to display the inferred trouble cause.

A learning device according to the present disclosure includes: a data acquisition unit to acquire learning data including input data including model data of an air conditioner and operation data of the air conditioner data at least at a time when a trouble of the air conditioner occurs, and training data representing a trouble cause of the air conditioner; and a model generation unit to use the learning data to generate a learned model for inferring a trouble cause of the air conditioner, from the model data of the air conditioner and the operation data of the air conditioner at least at a time when a trouble of the air conditioner occurs.

Advantageous Effects of Invention

According to the present disclosure, the frequency of dispatching personnel to sites can be reduced.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 is a diagram illustrating a flow of conventional maintenance service of air conditioners.

FIG. 2 is a diagram illustrating a configuration of air conditioners 21 and a diagnosis system 1.

FIG. 3 is a diagram illustrating a configuration of a learning device 7.

FIG. 4 is a diagram illustrating an example of model data K1 to K2.

FIG. 5 is a diagram illustrating an example of operation data A1 to An.

FIG. 6 is a diagram illustrating an example of trouble causes B1 to Bm of the air conditioner.

FIG. 7 is a diagram illustrating a configuration of a neural network in a first embodiment.

FIG. 8 is a flowchart illustrating the procedure of learning by learning device 7.

FIG. 9 is a diagram illustrating a configuration of an inference device 10.

FIG. 10 is a flowchart illustrating the procedure of inference by inference device 10.

FIG. 11 is a diagram illustrating an example of replacement parts.

FIG. 12 is a diagram illustrating a configuration of a neural network in a second embodiment.

FIG. 13 is a diagram illustrating a configuration of a neural network in a third embodiment.

FIG. 14 is a diagram illustrating an example of the worker's characteristics.

FIG. 15 is a diagram illustrating a configuration of a neural network in a fourth embodiment.

FIG. 16 is a diagram illustrating a configuration of inference device 10 in the fourth embodiment.

FIG. 17 is a diagram illustrating worker data.

FIG. 18 is a diagram illustrating a configuration of a neural network in a fifth embodiment.

FIG. 19 is a diagram illustrating a configuration of a neural network in a sixth embodiment.

FIG. 20 is a diagram illustrating a configuration of a neural network in a seventh embodiment.

FIG. 21 is a diagram illustrating an example of installation environment data E1 to E3.

FIG. 22 is a diagram illustrating another example of the worker's characteristics.

FIG. 23 is a diagram illustrating a hardware configuration of learning device 7 or inference device 10.

DESCRIPTION OF EMBODIMENTS

Embodiments will be described below with reference to the drawings.

First Embodiment

FIG. 1 is a diagram illustrating a flow of conventional maintenance service of air conditioners.

An owner possess an air conditioner. A service company performs operations such as test run, inspection, and repair. The number of service parts in possession is estimated from trouble history in the past or the like, and service parts are kept in stock.

(1) The owner becomes aware that the air conditioner develops a trouble. (2) The owner requests repair from an operator at the service company. (3) The owner and the service company arrange a schedule for dispatching a worker at the service company. (4) The operator at the service company instructs a worker to work on site. (5) The worker at the service company checks the site. (6) The worker at the service company diagnoses a trouble cause of the air conditioner. (7) The worker at the service company makes an arrangement for a replacement part, if necessary. (8) The operator at the service company adjusts a schedule of the worker's next visit for repair, depending on the situation of part arrangement. (9) The part is put into storage at the service company. (10) The worker at the service company visits the site again based on the arranged schedule and makes repairs.

As described above, the conventional maintenance service takes time and cost because the personnel of the service company has to visit the site at least twice before completion of repair. The initial response (first visit) is less likely to be considered as repair, and the visit expense is often not paid in full by the owner. Since a quick initial response is demanded, the service company needs to secure personnel in preparation for busy seasons. Moreover, the work load on workers is heavy in busy seasons. Workers have different strengths and weaknesses in repair works and have different levels in skill so that operation times and repair levels vary. Since there is no clear indicator as to how many repair parts should be in stock, the parts need to be stored equally, leading to waste in inventory control.

FIG. 2 is a diagram illustrating a configuration of air conditioners 21 and an air conditioner diagnosis system 1 according to an embodiment.

A plurality of air conditioners 21 are connected to diagnosis system 1 via a communication line 31. Diagnosis system 1 may be installed, for example, in a service company.

Each air conditioner 21 includes an indoor unit 22, an outdoor unit 23, a remote control 24, and a control device 25. Control device 25 includes an indoor unit controller 26, an outdoor unit controller 27, and a system controller 28.

Indoor unit 22 is installed indoors. Outdoor unit 23 is installed outdoors. Remote control 24 accepts the user's operation input. Indoor unit controller 26 controls indoor unit 22. Outdoor unit controller 27 controls outdoor unit 23. System controller 28 controls the entire air conditioner 21.

Diagnosis system 1 includes an input device 2, a communication circuit 3, a display device 4, a data storage device 5, a learned model storage device 6, a learning device 7, and an inference device 10.

Input device 2 accepts, for example, an operation input by the operator at the service company.

Communication circuit 3 exchanges data with a plurality of air conditioners 21 through communication line 31.

Data storage device 5 stores operation data transmitted from a plurality of air conditioners 21 through communication line 31, each with a timestamp and a device ID. This device ID is the ID that identifies air conditioner 21 that has transmitted operation data. When a trouble occurs in air conditioner 21, data storage device 5 stores a trouble cause diagnosed by a worker at the service company or the like, with a timestamp at the time when the trouble occurs and a device ID. This device ID is the ID that identifies the diagnosed air conditioner.

Learning device 7 generates a learned model that infers a trouble cause of the air conditioner, from model data of the air conditioner and operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Learned model storage device 6 stores the learned model generated by learning device 7.

Inference device 10 infers a trouble cause of the air conditioner, from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Display device 4 displays an inference result by inference device 10.

FIG. 3 is a diagram illustrating a configuration of learning device 7.

Learning device 7 includes a data acquisition unit 71 and a model generation unit 72.

Data acquisition unit 71 reads a combination of the timestamped trouble cause and the device ID stored in data storage device 5. Data acquisition unit 71 further reads the timestamped operation data stored in data storage device 5 in association with the read device ID. Data acquisition unit 71 acquires learning data including input data including the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data representing a trouble cause of the air conditioner, from these pieces of read data. At least the time when a trouble of the air conditioner occurs represents time t1 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t1, where Δt≥0. The model data can be identified from the read device ID.

FIG. 4 is a diagram illustrating an example of model data K1 to K2. The model data includes a model number and an item name.

FIG. 5 is a diagram illustrating an example of operation data A1 to An. The operation data is all sorts of data that can represent the operating state of air conditioner 21. Examples of the operation data include the operating ratio of compressor, setting temperature, outside temperature, operation mode, ejection temperature of compressor, intake temperature of compressor, degree of supercooling, frequency of compressor, opening degree of expansion valve, inlet temperature of heat exchanger of indoor unit, outlet temperature of heat exchanger of indoor unit, inlet temperature of heat exchanger of outdoor unit, outlet temperature of heat exchanger of outdoor unit, inlet pressure of heat exchanger of indoor unit, outlet pressure of heat exchanger of indoor unit, inlet pressure of heat exchanger of outdoor unit, outlet pressure of heat exchanger of outdoor unit, ejection pressure of compressor, intake pressure of compressor, and degree of heating.

FIG. 6 is a diagram illustrating an example of trouble causes B1 to Bm of the air conditioner. These are trouble causes that can be determined by operation data of the air conditioner. Examples of the trouble cause include abnormal display of remote control, communication abnormality of indoor unit, ejection temperature abnormality, heater operation malfunction, shutdown of indoor unit, low pressure abnormality, high pressure abnormality, drain pump abnormality, dust box abnormality, power supply signal synchronization abnormality, leakage abnormality, address error, no response error, drain pump abnormality, water stoppage abnormality, outside temperature abnormality, remote board abnormality, gas leakage, drain sensor submersion, and filter operation malfunction.

When the trouble cause of the air conditioner is Bi, data acquisition unit 71 acquires learning data including input data including model data K1 to K2 of the air conditioner and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data Z1 to Zm representing trouble cause Bi of the air conditioner, where Zi=1 and Zj=0 (j≠i).

Model generation unit 72 uses the learning data to generate a learned model for inferring a trouble cause of the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Model generation unit 72 generates a learned model, for example, by supervised learning in accordance with a neural network model. Supervised learning is a method of inferring the result from inputs by giving data sets of cause and result (label) to model generation unit 72 to learn the characteristics in these learning data.

The neural network consists of an input layer including a plurality of neurons, a hidden layer including a plurality of neurons, and an output layer including a plurality of neurons. One or more hidden layers may be provided. The activation function of the output unit of the output layer can be the softmax function.

FIG. 7 is a diagram illustrating a configuration of a neural network in a first embodiment. FIG. 7 illustrates a neural network having a plurality of layers.

When a plurality of inputs (causes) are input to the input layer, the value of each input is multiplied by a weight in the input layer, and the multiplication result is passed to the next layer. In the next layer, the input multiplication result is multiplied by a weight, and the multiplication result is further passed to the next layer. Finally, output data (result) is output from the output layer. The output result varies depending on the weights.

Data input to the input unit of the neural network is model data K1 to K2 and operation data A1 to An of the air conditioner, and data output from the output unit of the output layer is trouble causes B1 to Bm.

The neural network learns trouble causes by supervised learning in accordance with learning data created based on a combination of model data K1 to K2 and operation data A1 to An of the air conditioner, and Z1 to Zm (training data) acquired by data acquisition unit 71. In other words, the neural network learns by inputting model data K1 to K2 and operation data A1 to An of the air conditioner to the input layer and adjusting the weights such that trouble causes B1 to Bm output from the output layer approach training data Z1 to Zm.

Model generation unit 72 generates a learned model by performing the learning as described above and outputs the learned model to learned model storage device 6.

Learned model storage device 6 stores the learned model output from model generation unit 72.

FIG. 8 is a flowchart illustrating the procedure of learning by learning device 7. At step S101, data acquisition unit 71 acquires learning data including model data K1 to K2, operation data A1 to An of the air conditioner, and training data Z1 to Zm. Although it is assumed that model data K1 to K2, operation data A1 to An of the air conditioner, and training data Z1 to Zm (ground truth) are simultaneously acquired, model data K1 to K2, operation data A1 to An of the air conditioner, and training data Z1 to Zm may be acquired at different timings as long as model data K1 to K2, operation data A1 to An of the air conditioner, and training data Z1 to Zm are input in association with each other.

At step S102, model generation unit 72 generates a learned model by supervised learning, in accordance with the learning data created based on a combination of model data K1 to K2, operation data A1 to An of the air conditioner, and training data Z1 to Zm (ground truth) acquired by data acquisition unit 71.

At step S103, learned model storage device 6 stores the learned model generated by model generation unit 72.

FIG. 9 is a diagram illustrating a configuration of inference device 10.

Inference device 10 includes a data acquisition unit 73 and an inference unit 74.

Data acquisition unit 73 reads the timestamped operation data stored in data storage device 5 in association with the device ID of the air condition that is the target to be inferred. Data acquisition unit 73 acquires input data including model data K1 to K2 of the air conditioner and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, from the read data. Here, at least the time when a trouble of the air conditioner occurs represents time t2 when the trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t2, where Δt≥0. Model data K1 to K2 can be identified from the device ID of the air conditioner that is the target to be inferred.

Inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a trouble cause of the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, stored in learned model storage device 6, and outputs data representing a trouble cause of the air conditioner.

Inference unit 74 uses a learned neural network.

Data input to the input unit of the input layer of the neural network is model data K1 to K2 and operation data A1 to An of the air conditioner, and data output from the output unit of the output layer is trouble causes B1 to Bm. Inference unit 74 determines maximum value Bk among the output trouble causes B1 to Bm as the trouble cause of the air conditioner this time.

FIG. 10 is a flowchart illustrating the procedure of inference by inference device 10.

At step S201, data acquisition unit 73 acquires input data including model data K1 to K2 and operation data A1 to An of the air conditioner.

At step S202, inference unit 74 inputs the input data including model data K1 to K2 and operation data A1 to An of the air conditioner to the learned model stored in learned model storage device 6.

At step S203, inference unit 74 obtains outputs B1 to Bm of the learned model. When the maximum value among trouble causes B1 to Bm is Bk, inference unit 74 determines trouble cause Bk as the trouble cause of the air conditioner this time.

As described above, according to the present embodiment, when a trouble occurs in an air conditioner, the trouble cause can be diagnosed remotely, based on the model data and the operation data of the air conditioner. This configuration can reduce the frequency of visits by personnel of the service company to the sites. This configuration can be expected to suppress cost burden and improve cost resistance.

Second Embodiment

Data storage device 5 stores operation data transmitted from a plurality of air conditioners 21 through communication line 31, each with a timestamp and a device ID. This device ID is the ID that identifies air conditioner 21 that has transmitted operation data. When a trouble occurs in air conditioner 21, data storage device 5 stores a replacement part among a plurality of parts that constitute the air conditioner, with a timestamp at the time when the trouble occurs and the device ID at the time of repair by the worker or the like of the service company. This device ID is the ID that identifies the repaired air conditioner.

FIG. 11 is a diagram illustrating an example of replacement parts. These are replacement parts that can be determined by the model data and the operation data of the air conditioner or replacement parts that can be determined by a trouble cause of the air conditioner. Examples of the replacement part include control board, solenoid valve, thermistor, outdoor unit fan, indoor unit fan, low pressure sensor, heat exchanger of outdoor unit, heat exchanger of indoor unit, drain sensor, drain pipe, float switch, water feed tank, humidification solenoid valve, fuse, motor, compressor, shunt controller, Lossnay (energy recovery ventilator), remote control, and system controller.

Data acquisition unit 71 reads a combination of the timestamped replacement part among a plurality of parts that constitute the air conditioner and the device ID stored in data storage device 5. Data acquisition unit 71 further reads the timestamped operation data stored in data storage device 5 in association with the read device ID. Data acquisition unit 71 acquires learning data including input data including the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data representing a replacement part among a plurality of parts that constitute the air conditioner, from these pieces of read data. Here, at least the time when a trouble of the air conditioner occurs represents time t1 when the trouble occurs in the air conditioner and the time within a predetermined time Δt prior to time t1, where Δt≥0. The model data can be identified from the read device ID.

When the replacement part among a plurality of parts that constitute the air conditioner is Ci, data acquisition unit 71 acquires learning data including input data including model data K1 to K2 of the air conditioner and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data Z1 to Zp representing component Ci of the air conditioner, where Zi=1 and Zj=0 (j≠i).

Model generation unit 72 uses the learning data to generate a learned model for inferring a replacement part of the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs. Model generation unit 72 generates a learned model, for example, by supervised learning in accordance with a neural network model.

FIG. 12 is a diagram illustrating a configuration of a neural network in a second embodiment.

Data input to the input unit of the input layer of the neural network is model data K1 to K2 and operation data A1 to An of the air conditioner, and data output from the output unit of the output layer is replacement parts C1 to Cp. The activation function of the output unit of the output layer can be the softmax function.

The neural network learns replacement parts by supervised learning in accordance with learning data created based on a combination of model data K1 to K2 and operation data A1 to An of the air conditioner and Zi to Zp (training data) acquired by data acquisition unit 71. In other words, the neural network learns by inputting model data K1 to K2 and operation data A1 to An of the air conditioner to the input layer and adjusting the weights such that replacement parts C1 to Cp output from the output layer approach training data Z1 to Zp.

Model generation unit 72 generates a learned model by performing the learning as described above and outputs the learned model to learned model storage device 6.

Inference device 10 infers a replacement part among a plurality of parts that constitute the air conditioner, from input data including the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Data acquisition unit 73 reads the timestamped operation data stored in data storage device 5 in association with the device ID of the air conditioner that is the target to be inferred. Data acquisition unit 73 acquires input data including model data K1 to K2 of the air conditioner and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, from the read data. Here, at least the time when a trouble of the air conditioner occurs represents time t2 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t2, where Δt≥0. The model data can be identified from the device ID of the air conditioner that is the target to be inferred.

Inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner.

Inference Unit 74 Uses a Learned Neural Network.

Data input to the input unit of the input layer of the neural network is model data K1 to K2 and operation data A1 to An of the air conditioner, and data output from the output unit of the output layer is replacement parts C1 to Cp. Inference unit 74 determines maximum value Ck among the output replacement parts C1 to Cp as the replacement part of the air conditioner this time.

As described above, according to the present embodiment, when a trouble occurs in an air conditioner, the replacement part can be identified remotely, based on the model data and the operation data of the air conditioner. This configuration can reduce the frequency of visits by personnel of the service company to the sites. This configuration can be expected to suppress cost burden and improve cost resistance.

Third Embodiment

When a trouble occurs in air conditioner 21, data storage device 5 stores a trouble cause of the air conditioner diagnosed by the worker or the like of the service company and a replacement part among a plurality of parts that constitute the air conditioner in association with each other.

Data acquisition unit 71 acquires learning data including input data including a trouble cause of the air conditioner and training data representing a replacement part among a plurality of parts that constitute the air conditioner, from a trouble cause of the air conditioner and a replacement part among a plurality of parts that constitute the air conditioner, stored in association with each other in data storage device 5.

When a trouble cause of the air conditioner is Bk and a replacement part among a plurality of parts that constitute the air conditioner is Ci, data acquisition unit 71 generates input data including data X1 to Xm representing trouble cause Bk of the air conditioner, and training data Z1 to Zp representing replacement part Ci of the air conditioner. Here, Xk=1 and Xj=0 (j≠k). Zi=1 and Zj=0 M.

Model generation unit 72 uses the learning data to generate a learned model for inferring a replacement part of the air conditioner, from the input data including a trouble cause of the air conditioner.

Model generation unit 72 generates a learned model, for example, by supervised learning in accordance with a neural network model.

FIG. 13 is a diagram illustrating a configuration of a neural network in a third embodiment.

Data input to the input unit of the input layer of the neural network is data X1 to Xm representing trouble cause Bk of the air conditioner, and data output from the output unit of the output layer is replacement parts C1 to Cp. The activation function of the output unit of the output layer can be the softmax function.

The neural network learns replacement parts by supervised learning in accordance with learning data created based on a combination of data X1 to Xm representing a trouble cause of the air conditioner and Zi to Zp (training data) acquired by data acquisition unit 71. In other words, the neural network learns by inputting data X1 to Xm representing a trouble cause of the air conditioner to the input layer and adjusting the weights such that replacement parts C1 to Cp output from the output layer approach training data Z1 to Zp.

Model generation unit 72 generates a learned model by performing the learning as described above and outputs the learned model to learned model storage device 6.

When a trouble occurs in air conditioner 21, air conditioner 21 transmits data representing a trouble cause of the air conditioner to diagnosis system 1. For example, it is assumed that air conditioner 21 can identify a trouble cause of the air conditioner from the operating state of air conditioner 21, in the inside of the air conditioner 21 itself.

When a trouble occurs in the air conditioner, inference device 10 of diagnosis system 1 infers a replacement part among a plurality of parts that constitute air conditioner 21, from the input data including a trouble cause of the air conditioner acquired from air conditioner 21.

When a trouble occurs in air conditioner 21, data acquisition unit 73 acquires input data including a trouble cause of the air conditioner from air conditioner 21.

When a trouble cause of the air conditioner is Bk, data acquisition unit 73 generates input data including data X1 to Xm representing trouble cause Bk of the air conditioner. Here, Xk=1 and Xj=0 (j≠k).

Inference unit 74 inputs the input data from air conditioner 21 acquired by data acquisition unit 73 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from the input data including a trouble cause of the air conditioner stored in learned model storage device 6, and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner.

Inference unit 74 uses a learned neural network. Data input to the input unit of the input layer of the neural network is data X1 to Xm representing a trouble cause of the air conditioner, and data output from the output unit of the output layer is replacement parts C1 to Cp. Here, Xk=1, and Xj=0 (j≠k).

Inference unit 74 determines maximum value Ck among replacement parts C1 to Cp as the replacement part of the air conditioner this time.

As described above, according to the present embodiment, when a trouble occurs in an air conditioner, the replacement part can be identified remotely, based on a trouble cause of the air conditioner. This configuration can reduce the frequency of visits by personnel of the service company to the sites. This configuration can be expected to suppress cost burden and improve cost resistance.

First Modification to Third Embodiment

In the third embodiment, data acquisition unit 73 acquires input data including a trouble cause of the air conditioner from air conditioner 21 when a trouble occurs in air conditioner 21. Inference unit 74 inputs the input data from air conditioner 21 acquired by data acquisition unit 73 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from the input data including a trouble cause of the air conditioner, and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner.

In the present modification, inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a trouble cause of the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and outputs data representing a trouble cause of the air conditioner, in the same manner as in the first embodiment.

Inference unit 74 inputs, as the input data, data including the trouble cause inferred by inference unit 74 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from the input data including a trouble cause of the air conditioner, and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner.

Fourth Embodiment

Data storage device 5 stores operation data transmitted from a plurality of air conditioners 21 through communication line 31, each with a timestamp and a device ID. This device ID is the ID that identifies air conditioner 21 that has transmitted operation data. When a trouble occurs in air conditioner 21, data storage device 5 stores the characteristics of the worker repairing the air conditioner with a timestamp at the time when the trouble occurs and a device ID. This device ID is the ID that identifies the repaired air conditioner.

FIG. 14 is a diagram illustrating an example of the worker's characteristics. These are the characteristic of the worker repairing the air conditioner that can be determined by the model data and the operation data of the air conditioner, the characteristic of the worker repairing the air conditioner that can be determined by a trouble cause of the air conditioner, the characteristic of the worker repairing the air conditioner that can be determined by a replacement part of the air conditioner, or the characteristic of the worker repairing the air conditioner that can be determined by a trouble cause of the air conditioner and a replacement part of the air conditioner. For example, the workers' characteristics D1, D2, D3, D4, D5, D6, and D7 are years of experience, skill level pertaining to outdoor units, skill level pertaining to indoor units, skill level pertaining to electronic control, skill level pertaining to mechanics, skill level pertaining to communication, and skill level pertaining to display, respectively. The values of D1 to Ds are 0 or greater and 1 or smaller.

Data acquisition unit 71 reads a combination of the timestamped characteristics of the worker repairing the air conditioner and the device ID stored in data storage device 5. Data acquisition unit 71 further reads the timestamped operation data stored in data storage device 5 in association with the read device ID.

Data acquisition unit 71 acquires learning data including input data including model data K1 to K2 of the air conditioner and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data Z1 to Zs representing characteristics D1 to Ds of the worker repairing the air conditioner. Here, at least the time when a trouble of the air conditioner occurs represents time t1 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t1, where Δt≥0. The model data can be identified from the read device ID.

Model generation unit 72 uses the learning data to generate a learned model for inferring the characteristics of the worker repairing the air conditioner, from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs. Model generation unit 72 generates a learned model, for example, by supervised learning in accordance with a neural network model.

FIG. 15 is a diagram illustrating a configuration of a neural network in a fourth embodiment.

Data input to the input unit of the input layer of the neural network is model data K1 to K2 and operation data A1 to An of the air conditioner, and data output from the output unit of the output layer is characteristics D1 to Ds of the worker repairing the air conditioner.

The neural network learns the characteristics of the worker repairing the air conditioner by supervised learning in accordance with learning data created based on a combination of model data K1 to K2 and operation data A1 to An of the air conditioner, and Zi to Zs (training data) acquired by data acquisition unit 71. In other words, the neural network learns by inputting model data K1 to K2 and operation data A1 to An of the air conditioner to the input layer and adjusting the weights such that characteristics D1 to Ds of the worker output from the output layer approach training data Z1 to Zs.

Model generation unit 72 generates a learned model by performing the learning as described above and outputs the learned model to learned model storage device 6.

FIG. 16 is a diagram illustrating a configuration of inference device 10 in the fourth embodiment.

Inference device 10 infers the characteristics of the worker repairing the air conditioner, from the input data including the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Inference Device 10 Includes a Data Acquisition Unit 73, an Inference Unit 74, and a Worker Selecting Unit 81.

Data acquisition unit 73 reads the timestamped operation data stored in data storage device 5 in association with the device ID of the air conditioner that is the target to be inferred. Data acquisition unit 73 acquires input data including model data K1 to K2 of the air conditioner and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, from the read data. Here, at least the time when a trouble of the air conditioner occurs represents time t2 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t2, where Δt≥0. The model data can be identified from the device ID of the air conditioner that is the target to be inferred.

Inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and outputs data representing the characteristics of the worker repairing the air conditioner.

Inference unit 74 uses the learned neural network stored in learned model storage device 6. Data input to the input unit of the input layer of the neural network is model data K1 to K2 and operation data A1 to An of the air conditioner, and data output from the output unit of the output layer is data D1 to Ds representing the characteristics of the worker repairing the air conditioner.

Worker selecting unit 81 selects a worker to be dispatched for repair, in accordance with the inferred characteristics of the worker, and the characteristics and the schedules of a plurality of workers.

FIG. 17 is a diagram illustrating worker data.

The worker data defines the characteristics and the available schedule for dispatch for each worker.

For example, the characteristics D1, D2, D3, D4, D5, D6, D7, . . . of worker M1 are 0.2, 0.1, 0.4, 0.8, 0.5, 0.9, 0.2, . . . , respectively. The available dates for dispatch of worker M1 is 7/14 to 7/29.

Worker selecting unit 81 calculates the root sum square Su of the difference between the worker's characteristics D1 to Ds inferred by inference unit 74 and characteristics D1 to Ds of each worker Mu. Worker selecting unit 81 identifies the worker with the smallest Su and, if the available dates for dispatch of the worker are included in the on-site acceptable schedule, approves the worker as the worker to be dispatched for repair. If the available dates for dispatch of the worker are not included in the on-site acceptable schedule, worker selecting unit 81 identifies the worker with the second smallest Su and, if the available dates for dispatch of the worker are included in the on-site acceptable schedule, approves the worker as the worker to be dispatched for repair. If the available dates for dispatch of the worker are not included in the on-site acceptable schedule, the process above is repeated until the available dates for dispatch of another worker are included in the on-site acceptable schedule.

As described above, according to the present embodiment, when a trouble occurs in an air conditioner, the appropriate worker to repair the air conditioner can be identified remotely, based on the model data and the operation data of the air conditioner. This configuration can reduce the frequency of visits by personnel of the service company to the sites. Furthermore, the appropriate worker may accompany a trainee so that the trainee can gain experience. In addition, education pertaining to lacking skills may be provided in accordance with technician data, thereby raising the level of technological capability of the service company.

Fifth Embodiment

When a trouble occurs in air conditioner 21, data storage device 5 stores a trouble cause of the air conditioner diagnosed by the worker or the like of the service company and the characteristics of the worker repairing the air conditioner in association with each other.

Data acquisition unit 71 acquires learning data including input data including a trouble cause of the air conditioner and training data representing the characteristics of the worker repairing the air conditioner, from a trouble cause of the air conditioner and the characteristics of the worker repairing the air conditioner stored in association with each other in data storage device 5.

When a trouble cause of the air conditioner is Bk, data acquisition unit 71 acquires learning data including input data including data X1 to Xm representing trouble cause Bk of the air conditioner and training data Z1 to Zs representing the characteristics D1 to Ds of the worker repairing the air conditioner. Here, Xk=1 and Xj=0 (j≠k).

Model generation unit 72 uses the learning data to generate a learned model for inferring the characteristics of the worker repairing the air conditioner, from a trouble cause of the air conditioner. Model generation unit 72 generates a learned model, for example, by supervised learning in accordance with a neural network model.

FIG. 18 is a diagram illustrating a configuration of a neural network in a fifth embodiment.

Data input to the input unit of the input layer of the neural network is data X1 to Xm representing trouble cause Bk of the air conditioner, and data output from the output unit of the output layer is characteristics D1 to Ds of the worker repairing the air conditioner.

The neural network learns the characteristics of the worker repairing the air conditioner by supervised learning in accordance with learning data created based on a combination of data X1 to Xm representing a trouble cause and Zi to Zs (training data) acquired by data acquisition unit 71. In other words, the neural network learns by inputting data X1 to Xm representing a trouble cause of the air conditioner to the input layer and adjusting the weights such that characteristics D1 to Ds of the worker output from the output layer approach training data Z1 to Zs.

Model generation unit 72 generates a learned model by performing the learning as described above and outputs the learned model to learned model storage device 6.

When a trouble occurs in air conditioner 21, air conditioner 21 transmits data representing a trouble cause of the air conditioner to diagnosis system 1. For example, it is assumed that air conditioner 21 can identify a trouble cause of the air conditioner from the operating state of air conditioner 21, in the inside of the air conditioner 21 itself.

When a trouble occurs in the air conditioner, inference device 10 of diagnosis system 1 infers the characteristics of the worker repairing the air conditioner, from the input data including a trouble cause of the air conditioner acquired from air conditioner 21.

When a trouble occurs in air conditioner 21, data acquisition unit 73 acquires input data including a trouble cause of the air conditioner from air conditioner 21.

When a trouble cause of the air conditioner is Bk, data acquisition unit 73 generates input data including data X1 to Xm representing trouble cause Bk of the air conditioner. Here, Xk=1 and Xj=0 (j≠k).

Inference unit 74 inputs the input data from air conditioner 21 acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a trouble cause of the air conditioner stored in learned model storage device 6, and outputs data representing the characteristics of the worker repairing the air conditioner.

Inference unit 74 uses a learned neural network. Data input to the input unit of the input layer of the neural network is data X1 to Xm representing trouble cause Bk of the air conditioner, and data output from the output unit of the output layer is data D1 to Ds representing the characteristics of the worker repairing the air conditioner.

As described above, according to the present embodiment, when a trouble occurs in an air conditioner, the appropriate worker to repair the air conditioner can be identified remotely, based on a trouble cause of the air conditioner.

First Modification to Fifth Embodiment

In the fifth embodiment, data acquisition unit 73 acquires input data including a trouble cause of the air conditioner from air conditioner 21 when a trouble occurs in air conditioner 21. Inference unit 74 inputs the input data from air conditioner 21 acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a trouble cause of the air conditioner, and outputs data representing the characteristics of the worker repairing the air conditioner.

In the present modification, inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a trouble cause of the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and outputs data representing a trouble cause of the air conditioner, in the same manner as in the first embodiment.

Inference unit 74 further inputs, as the input data, data including a trouble cause inferred by inference unit 74 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a trouble cause of the air conditioner, and outputs data representing the characteristics of the worker repairing the air conditioner.

Sixth Embodiment

When a trouble occurs in air conditioner 21, data storage device 5 stores a replacement part among a plurality of parts that constitute the air conditioner and the characteristics of the worker repairing the air conditioner in association with each other.

Data acquisition unit 71 acquires learning data including input data including a replacement part among a plurality of parts that constitute the air conditioner and training data representing the characteristics of the worker repairing the air conditioner, from the replacement part among a plurality of parts that constitute the air conditioner and the characteristics of the worker repairing the air conditioner, stored in association with each other in data storage device 5.

When the replacement part among a plurality of parts that constitute the air conditioner is Ck, data acquisition unit 71 acquires learning data including input data including data Y1 to Yp representing replacement part Ck of the air conditioner and training data Z1 to Zs representing the characteristics D1 to Ds of the worker repairing the air conditioner. Here, Yk=1 and Yj=0 (j≠k).

Model generation unit 72 uses the learning data to generate a learned model for inferring the characteristics of the worker repairing the air conditioner, from the replacement part among a plurality of parts that constitute the air conditioner. Model generation unit 72 generates a learned model, for example, by supervised learning in accordance with a neural network model.

FIG. 19 is a diagram illustrating a configuration of a neural network in a sixth embodiment.

Data input to the input unit of the input layer of the neural network is data Y1 to Yp representing replacement part Ck among a plurality of parts that constitute the air conditioner, and data output from the output unit of the output layer is characteristics D1 to Ds of the worker repairing the air conditioner.

The neural network learns the characteristics of the worker repairing the air conditioner by supervised learning in accordance with learning data created based on a combination of data Y1 to Yp representing the replacement part among a plurality of parts that constitute the air conditioner and Zi to Zs (training data) acquired by data acquisition unit 71. In other words, the neural network learns by inputting data Y1 to Yp representing a replacement part among a plurality of parts that constitute the air conditioner to the input layer and adjusting the weights such that characteristics D1 to Ds of the worker output from the output layer approach training data Z1 to Zs.

Model generation unit 72 generates a learned model by performing the learning as described above and outputs the learned model to learned model storage device 6.

When a trouble occurs in air conditioner 21, air conditioner 21 transmits data representing a replacement part among a plurality of parts that constitute the air conditioner to diagnosis system 1. For example, it is assumed that air conditioner 21 can identify a replacement part among a plurality of parts that constitute the air conditioner, from the operating or the abnormal state of air conditioner 21, in the inside of the air conditioner 21 itself.

When a trouble occurs in the air conditioner, inference device 10 of diagnosis system 1 infers the characteristics of the worker repairing the air conditioner, from the input data including a replacement part among a plurality of parts that constitute the air conditioner acquired from air conditioner 21.

When a trouble occurs in air conditioner 21, data acquisition unit 73 acquires input data including the replacement part of the air conditioner from air conditioner 21.

Data acquisition unit 73 acquires input data including replacement part Ck among a plurality of parts that constitute the air conditioner. When the replacement part of the air conditioner is Ck, data acquisition unit 73 generates input data including data Y1 to Yp representing replacement part Ck of the air conditioner. Here, Yk=1 and Yj=0 (j≠k).

Inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a replacement part of the air conditioner stored in learned model storage device 6, and outputs data representing the characteristics of the worker repairing the air conditioner.

Inference unit 74 uses a learned neural network. Data input to the input unit of the input layer of the neural network is data Y1 to Yp representing replacement part Ck of the air conditioner, and data output from the output unit of the output layer is data D1 to Ds representing the characteristics of the worker repairing the air conditioner.

As described above, according to the present embodiment, when a trouble occurs in an air conditioner, the appropriate worker to repair the air conditioner can be identified remotely, based on the replacement part of the air conditioner.

Modification to Sixth Embodiment

In the sixth embodiment, data acquisition unit 73 acquires input data including the replacement part among a plurality parts that constitute the air conditioner from air conditioner 21 when a trouble occurs in air conditioner 21. Inference unit 74 inputs the input data from air conditioner 21 acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a replacement part among a plurality of parts that constitute the air conditioner, and outputs data representing the characteristics of the worker repairing the air conditioner.

In the present modification, inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner, in the same manner as in the second embodiment. Alternatively, inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from a trouble cause of the air conditioner, and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner, in the same manner as in the third embodiment.

Inference unit 74 further inputs, as the input data, data including the replacement part among a plurality of parts that constitute the air conditioner inferred by the inference unit 74 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a replacement part among a plurality of parts that constitute the air conditioner, and outputs data representing the characteristics of the worker repairing the air conditioner.

Seventh Embodiment

When a trouble occurs in air conditioner 21, data storage device 5 stores a trouble cause of the air conditioner diagnosed by the worker or the like of the service company and a replacement part among a plurality of parts that constitute the air conditioner in association with the characteristics of the worker repairing the air conditioner.

Data acquisition unit 71 acquires learning data including input data including a trouble cause of the air conditioner and a replacement part and training data representing the characteristics of the worker repairing the air conditioner, from a trouble cause of the air conditioner, a replacement part among a plurality of parts that constitute the air conditioner, and the characteristics of the worker repairing the air conditioner stored in association with each other in data storage device 5.

When a trouble cause of the air conditioner is Bk and the replacement part of the air conditioner is Cr, data acquisition unit 71 acquires learning data including input data including data X1 to Xm representing trouble cause Bk of the air conditioner and data Y1 to Yp representing replacement part Cr of the air conditioner, and training data Z1 to Zs representing the characteristics D1 to Ds of the worker repairing the air conditioner. Here, Xk=1 and Xj=0 (j≠k). Yr=1 and Yj=0 (j≠r).

Model generation unit 72 uses the learning data to generate a learned model for inferring the characteristics of the worker repairing the air conditioner, from a trouble cause of the air conditioner and a replacement part of the air conditioner. Model generation unit 72 generates a learned model, for example, by supervised learning in accordance with a neural network model.

FIG. 20 is a diagram illustrating a configuration of a neural network in a seventh embodiment.

Data input to the input unit of the input layer of the neural network is data X1 to Xm representing trouble cause Bk of the air conditioner and data Y1 to Yp representing replacement part Cr of the air conditioner, and data output from the output unit of the output layer is characteristics D1 to Ds of the worker repairing the air conditioner.

The neural network learns the characteristics of the worker repairing the air conditioner by supervised learning in accordance with learning data created based on a combination of data X1 to Xm representing a trouble cause, data Y1 to Yp representing a replacement part, and Zi to Zs (training data) acquired by data acquisition unit 71. In other words, the neural network learns by inputting data X1 to Xm representing a trouble cause of the air conditioner and data Y1 to Yp representing a replacement part of the air conditioner to the input layer and adjusting the weights such that characteristics D1 to Ds of the worker output from the output layer approach training data Z1 to Zs.

Model generation unit 72 generates a learned model by performing the learning as described above and outputs the learned model to learned model storage device 6.

When a trouble occurs in air conditioner 21, air conditioner 21 transmits data representing a trouble cause of the air conditioner and a replacement part among a plurality of parts that constitute the air conditioner to diagnosis system 1. For example, it is assumed that air conditioner 21 can identify a trouble cause of the air conditioner and a replacement part among a plurality of parts that constitute the air conditioner, from the operating or the abnormal state of air conditioner 21, in the inside of the air conditioner 21 itself.

When a trouble occurs in the air conditioner, inference device 10 of diagnosis system 1 infers the characteristics of the worker repairing the air conditioner, from the input data including a trouble cause of the air conditioner and a replacement part among a plurality of parts that constitute the air conditioner acquired from air conditioner 21.

When a trouble occurs in air conditioner 21, data acquisition unit 73 acquires input data including a trouble cause of the air conditioner and a replacement part of the air conditioner from air conditioner 21.

Data acquisition unit 73 acquires input data including trouble cause Bk of the air conditioner and replacement part Cr of the air conditioner. When the trouble cause of the air conditioner is Bk and the replacement part of the air conditioner is Cr, data acquisition unit 73 generates input data including data X1 to Xm representing trouble cause Bk of the air conditioner and data Y1 to Yp representing replacement part Cr of the air conditioner. Here, Xk=1, Xj=0 (j≠k), Yr=1, and Yj=0 (j≠r).

Inference unit 74 inputs the input data from air conditioner 21 acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a trouble cause of the air conditioner and a replacement part of the air conditioner stored in learned model storage device 6, and outputs data representing the characteristics of the worker repairing the air conditioner.

Inference unit 74 uses a learned neural network. Data input to the input unit of the input layer of the neural network is data X1 to Xm representing trouble cause Bk of the air conditioner and data Y1 to Yp representing replacement part Cr of the air conditioner, and data output from the output unit of the output layer is data D1 to Ds representing the characteristics of the worker repairing the air conditioner.

As described above, according to the present embodiment, when a trouble occurs in an air conditioner, the appropriate worker to repair the air conditioner can be identified remotely, based on a trouble cause of the air conditioner and a replacement part of the air conditioner.

First Modification to Seventh Embodiment

In the seventh embodiment, data acquisition unit 73 acquires input data including a trouble cause of the air conditioner and a replacement part among a plurality parts that constitute the air conditioner from air conditioner 21 when a trouble occurs in air conditioner 21. Inference unit 74 inputs the input data from air conditioner 21 acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a trouble cause of the air conditioner and a replacement part among a plurality of parts that constitute the air conditioner, and outputs data representing the characteristics of the worker repairing the air conditioner.

In the present modification, inference unit 74 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a trouble cause of the air conditioner from the model data of the air conditioner and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and outputs data representing a trouble cause of the air conditioner, in the same manner as in the modification to the third embodiment. Inference unit 74 further inputs a trouble cause of the air conditioner estimated by inference unit 74 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from a trouble cause of the air conditioner, and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner.

Inference unit 74 further inputs, as the input data, data including a trouble cause of the air conditioner and a replacement part among a plurality of parts that constitute the air conditioner inferred by the inference unit 74 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including a trouble cause of the air conditioner and a replacement part among a plurality of parts that constitute the air conditioner, and outputs data representing the characteristics of the worker repairing the air conditioner.

Modifications

(1) Although the learning device and the inference device are provided inside the diagnosis system, the present invention is not limited thereto. The learning device and the inference device may be provided outside the diagnosis system.

(2) Deep learning in which extraction of feature amounts is learned may be used as a learning algorithm used in the model generation unit, and machine learning may be performed in accordance with any other known methods, such as genetic programming, inductive logic programming, and support vector machines.

(3) In some of the foregoing embodiments, the model data of the air conditioner is input to the learning device and the inference device. However, installation environment data may be additionally input.

FIG. 21 is a diagram illustrating an example of installation environment data E1 to E3. As illustrated in FIG. 21, installation environment data E1 to E3 include the kind of the building in which the air conditioner is installed, the region where the air conditioner is installed, and season.

For example, installation environment data E1 to E3 may be added to data input to the input layer of the neural network in FIG. 7, and learning device 7 and inference device 10 may operate as follows.

When the trouble cause of the air conditioner is Bi, data acquisition unit 71 of learning device 7 acquires learning data including input data including model data K1 to K2 of the air conditioner, installation environment data E1 to E3 of the air conditioner, and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data Z1 to Zm representing trouble cause Bi of the air conditioner. Here, Zi=1 and Zj=0 (j≠i).

Model generation unit 72 of learning device 7 uses the learning data to generate a learned model for inferring a trouble cause of the air conditioner from the model data of the air conditioner, the installation environment data of the air conditioner, and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Data acquisition unit 73 of inference device 10 acquires input data including model data K1 to K2 of the air conditioner, installation environment data E1 to E3 of the air conditioner, and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs. Here, at least the time when a trouble of the air conditioner occurs represents time t2 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t2, where Δt≥0.

Inference unit 74 of inference device 10 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a trouble cause of the air conditioner from the model data of the air conditioner, the installation environment data of the air conditioner, and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, stored in learned model storage device 6, and outputs data representing a trouble cause of the air conditioner.

For example, installation environment data E1 to E3 may be added to data input to the input layer of the neural network in FIG. 12, and learning device 7 and inference device 10 may operate as follows.

When the replacement part among a plurality of parts that constitute the air conditioner is Ci, data acquisition unit 71 of learning device 7 acquires learning data including input data including model data K1 to K2 of the air conditioner, installation environment data E1 to E3 of the air conditioner, and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data Z1 to Zp representing component Ci of the air conditioner. Here, Zi=1 and Zj=0 (j≠i).

Model generation unit 72 of learning device 7 uses the learning data to generate a learned model for inferring a replacement part of the air conditioner from model data K1 to K2 of the air conditioner, installation environment data E1 to E3 of the air conditioner, and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Data acquisition unit 73 of inference device 10 acquires input data including model data K1 to K2 of the air conditioner, installation environment data E1 to E3 of the air conditioner, and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs. Here, at least the time when a trouble of the air conditioner occurs represents time t2 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t2, where Δt≥0.

Inference unit 74 of inference device 10 inputs the input data acquired by data acquisition unit 73 to the learned model that infers a replacement part among a plurality of parts that constitute the air conditioner from the model data of the air conditioner, the installation environment data of the air conditioner, and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and outputs data representing a replacement part among a plurality of parts that constitute the air conditioner.

For example, installation environment data E1 to E3 may be added to data input to the input layer of the neural network in FIG. 15, and learning device 7 and inference device 10 may operate as follows.

Data acquisition unit 71 of learning device 7 acquires learning data including input data including model data K1 to K2 of the air conditioner, installation environment data E1 to E3 of the air conditioner, and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs, and training data Z1 to Zs representing characteristics D1 to Ds of the worker repairing the air conditioner. Here, at least the time when a trouble of the air conditioner occurs represents time t1 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t1, where Δt≥0.

Model generation unit 72 of learning device 7 uses the learning data to generate a learned model for inferring the characteristics of the worker repairing the air conditioner, from the model data of the air conditioner, the installation environment data of the air conditioner, and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs.

Data acquisition unit 73 of inference device 10 acquires input data including model data K1 to K2 of the air conditioner, installation environment data E1 to E3 of the air conditioner, and operation data A1 to An of the air conditioner at least at the time when a trouble of the air conditioner occurs. Here, at least the time when a trouble of the air conditioner occurs represents time t2 when a trouble of the air conditioner occurs and the time within a predetermined time Δt prior to time t2, where Δt≥0.

Inference unit 74 of inference device 10 inputs the input data acquired by data acquisition unit 73 to the learned model that infers the characteristics of the worker repairing the air conditioner from the input data including the model data of the air conditioner, the installation environment data of the air conditioner, and the operation data of the air conditioner at least at the time when a trouble of the air conditioner occurs, and outputs data representing the characteristics of the worker repairing the air conditioner.

(4) FIG. 22 is a diagram illustrating another example of the worker's characteristics. The characteristics in FIG. 22 include, in addition to the characteristics in FIG. 14, skill levels pertaining to the models of air conditioners, such as skill level D8 pertaining to Model A and skill level D9 pertaining to Model B. The values of D8 and D9 can be set to 0 or greater and 1 or smaller.

In some of the foregoing embodiments, the characteristics levels of the worker in FIG. 22 may be used instead of the characteristics levels of the worker in FIG. 14.

(5) For example, at least one of model data K1 to K2 and installation environment data E1 to E3 may be added to data input to the input layer of the neural networks in FIG. 13, FIG. 18, FIG. 19, and FIG. 20.

(6) The inference device and the learning device described in the foregoing embodiments can be configured as hardware of digital circuitry or software for the corresponding operations. When the functions of the inference device and the learning device are implemented by software, for example, as illustrated in FIG. 23, the inference device and the learning device may include a processor 5002 and a memory 5001 connected via a bus 5003, and a program stored in memory 5001 may be executed by processor 5002.

(7) In the foregoing embodiment, the inference device uses the learned model to infer a trouble cause of the air conditioner, a replacement part of the air conditioner, or the characteristics of the worker repairing the air conditioner, from the input data acquired by the data acquisition unit. However, the present invention is not limited thereto.

For example, the inference device may infer a trouble cause of the air conditioner, a replacement part of the air conditioner, or the characteristics of the worker repairing the air conditioner from the input data acquired by the data acquisition unit, based on rule-based inference or case-based inference.

Embodiments disclosed here should be understood as being illustrative rather than being limitative in all respects. The scope of the present disclosure is shown not in the foregoing description but in the claims, and it is intended that all modifications that come within the meaning and range of equivalence to the claims are embraced here.

REFERENCE SIGNS LIST

- 1 diagnosis system, 2 input device, 3 communication circuit, 4 display device, 5 data storage device, 6 learned model storage device, 7 learning device, 10 inference device, 21 air conditioner, 22 indoor unit, 23 outdoor unit, 24 remote control, 25 control device, 26 indoor unit controller, 27 outdoor unit controller, 28 system controller, 131 communication line, 71, 73 data acquisition unit, 72 model generation unit, 74 inference unit, 81 worker selecting unit, 5001 memory, 5002 processor, 5003 bus.

Claims

1-8. (canceled)

9. An air conditioner diagnosis system that remotely diagnoses a plurality of air conditioners, comprising:

an inference device to infer characteristics of a worker repairing an air conditioner, from input data including model data of the air conditioner and operation data of the air conditioner at least at a time when a trouble of the air conditioner occurs; and

a display device to display the inferred characteristics of the worker.

10. The air conditioner diagnosis system according to claim 9, wherein

the inference device includes a data acquisition circuitry to acquire the input data, and an inference circuitry to input the input data acquired by the data acquisition unit to a learned model that infers characteristics of a worker repairing the air conditioner from model data of the air conditioner and operation data of the air conditioner at least at a time when a trouble of the air conditioner occurs, and infer characteristics of a worker repairing the air conditioner.

11. An air conditioner diagnosis system that remotely diagnoses a plurality of air conditioners, comprising:

an inference device to infer characteristics of a worker repairing an air conditioner, from input data including a trouble cause of the air conditioner; and

a display device to display the inferred characteristics of the worker.

12. The air conditioner diagnosis system according to claim 11, wherein

the inference device includes a data acquisition circuitry to acquire the input data, and an inference circuitry to input the input data acquired by the data acquisition unit to a learned model that infers characteristics of a worker repairing the air conditioner from a trouble cause of the air conditioner, and infer characteristics of a worker repairing the air conditioner.

13-26. (canceled)

27. A learning device comprising:

a data acquisition circuitry to acquire learning data including input data including model data of an air conditioner and operation data of the air conditioner data at least at a time when a trouble of the air conditioner occurs, and training data representing characteristics of a worker repairing the air conditioner; and

a model generation circuitry to use the learning data to generate a learned model for inferring characteristics of a worker repairing the air conditioner, from model data of the air conditioner and operation data of the air conditioner at least at a time when a trouble of the air conditioner occurs.

28-30. (canceled)