Abstract: Existing systems for fault detection and classification have the disadvantage that they have limited or no capability for fault localization and root cause identification, probably due to the challenges associated with modeling the nonlinear interactions among process variables and capturing the nonstationary behavior that is typical of most industrial processes. The disclosure herein generally relates to industrial manufacturing systems, and, more particularly, to method and system for localization of faults in an industrial manufacturing plant. The system uses a perturbation based approach for fault localization, in which the system determines variables having dominant effect on identified faults, in terms of a perturbation score calculated for each of the variables.

Abstract: It is important to know the flow rates of oil and gas from individual wells in connected oil and gas wells. The existing methods for multiphase flow measurement are prohibitively expensive and used infrequently. The system is configured to ingest real-time and non-real-time data from a plurality of well data sources. Utilizing this data, a plurality of physics-guided data-driven well surveillance models run in real-time for forecasting a plurality of parameters including the flow rates of oil, gas and brine from individual wells, computing the health of well assets and performing fault detection and localization in well assets. The system is also configured to automatically compose a well performance optimization problem based on the current performance of the wells and health of well assets and solve the problem to identify optimal process settings for improving the operation of connected oil and gas wells.

Abstract: The disclosure relates to anomaly detection in an industrial environment including multiple industrial units and systems, generating huge volume of data. The conventional methods rely only on sensor data alone. The techniques of handling missing data plays a crucial role in determining the performance of industrial anomaly detection system. Further, imputation of missing data could cause error in computation, thus affecting the accuracy of the industrial anomaly detection system. The present disclosure addresses the problems associated with missing data by utilizing a masking technique. Further, the present disclosure utilizes quantitative and qualitative metadata associated with industrial system along with the sensor data to improve anomaly detection performance. Furthermore, the present disclosure includes a model recommendation system which provides transfer learning based utilization of existing models for similar industrial systems.

Abstract: Fault diagnosis in industries typically involves identification of key variables/sensors bearing fault signature, classification of detected fault into known fault classes and detecting root causes/sources of the fault. This disclosure relates to a method and system for a deep learning based causal inference in a multivariate time series data of abnormal events and failures in industrial manufacturing processes and equipment. The system generates causal networks for non-linear and non-stationary multivariate time series data. The causal network learns for a dynamic non-stationary and nonlinear complex process or system fault using observed data without any prior process knowledge. The causal networks of faults are identified in real-time using a deep learning-based causal network learning technique.

Abstract: Industrial processes and equipment are prone to operational changes and faulty operation of such processes and equipment can adversely affect output of the overall setup. Existing systems for monitoring and fault detection consider individual instances of data for fault detection, which may not be suitable for industrial processes. Disclosed herein is a system and a method for anomaly detection in an industrial enterprise. The system collects data from a plurality of sensors as input. The system processes the collected data along temporal dimension, during which the data is split to multiple segments of fixed window size. Data in each segment is processed to identify anomalous data, and data in segments identified as containing the anomalous data is further processed to identify one or more sensors that are faulty and are contributing to the anomalous data.

Abstract: Data is prone to various attacks such as cyber-security attacks, in any industry. State of the art systems in the domain of data security fail to identify adversarial attacks in real-time, and this leads to security issues, as well as results in the process/system providing unintended results. The disclosure herein generally relates to data security analysis, and, more particularly, to a method and system for assessing impact of adversarial attacks on time series data and providing defenses against such attacks. The system performs adversarial attacks on a selected data-driven model to determine impact of the adversarial attacks on the selected data model, and if the impact is such that performance of the selected data model is less than a threshold, then the selected data model is retrained.

Abstract: Gas turbines are one of leading sources for power generation with lower greenhouse gas emissions. However, due to environmental concerns, gas turbines are moving towards adopting greener fuels. The shift towards greener fuels comes with own set of challenges as performance of gas turbine at different operating points needs to be accurately predicted as experiments are very costly to perform. Existing arts perform their analysis at operating line and performance estimation at other operating points is not specified. Present application provides systems and methods for estimating performance of gas turbine accurately in wide operating region. The system first accurately estimates outlet conditions for each stage of compressor. The system then utilizes estimated outlet conditions to determine outlet conditions associated with other component of gas turbine.

Abstract: The disclosure is a method and a system for root cause identification (RCI) of faults in manufacturing and process industries. With complex interrelated multivariate data in manufacturing and process industries, the process of root RCI of faults is challenging. Further, the existing techniques for RCI have significant dependency on manual inputs and subject matter knowledge/experts. The disclosure is method and a system for root cause identification of a fault based on causal maps. The root cause of fault is identified in several steps including: generation of casual maps using data received from a manufacturing and process industry and root cause identification from the causal maps based on a Fault Traversal and Root Cause Identification (FTRCI) technique. The FTRCI identifies root cause from the causal map by identifying a fault traversal pathway from a leaf node in the causal map, wherein the fault traversal pathway is identified for even cyclic paths.

Abstract: State of the art systems used for industrial plant monitoring have the disadvantage that they fail to correctly assess reason for dip in performance of the plant and in turn trigger appropriate corrective measures. The disclosure herein generally relates to industrial plant monitoring, and, more particularly, to a system and method for development and deployment of self-organizing cyber-physical systems for manufacturing industries. The system monitors and collects data with respect to various parameters, from the industrial plant. If any performance dip is detected, the system determines corresponding cause, and also triggers one or more corrective actions to improve performance of the plant and different plant components to a desired performance level.

Abstract: This disclosure relates generally to system and method for molecular property prediction. The disclosed method includes mapping node embeddings of a molecular graph to a graph-level embedding characterizing the molecular graph. The graph level representation is acquired by pooling characteristics of hidden states of the nodes in the molecular graph by performing an iterative content based attention in a plurality of iterations. The content based attention is performed by considering an edge information fused transformed hidden state vector of the nodes of the molecular graph. The graph level embedding is fed through the linear projection to predict the molecular properties of the molecular graphs.

Abstract: This disclosure relates generally to system and method for molecular property prediction. The method utilizes a set-pooling aggregation operator to derive a graph-level representation of a complete input molecular graphs to assist in inductive learning tasks. The method includes iteratively down-sampling the molecular graph into a coarsened molecular graph, and determining adjacency matrix and feature matrix of the coarsened molecular graph. The method then includes computing an average of the hidden state node attributes of the coarsened graph obtained after preforming the iterations to obtain a graph level representation vector of the molecular graph. Using a linear layer from the graph level representation vector the molecular properties are determined.

Abstract: State-of-the-art systems used for plant monitoring and optimization fail to efficiently monitor and improve the performance of blast furnace ironmaking plants due to complexity of such plants. In addition, they attempt optimization without considering the operational stability of the blast furnace. The disclosure herein generally relates to industrial plant monitoring, and, more particularly, to a method and system for ironmaking plant optimization. The system determines an operational stability of the plant in terms of value of a determined Blast Furnace Stability Index (BFSI). Further, if the BFSI or one or more Key Performance Indicators (KPIs) of the plant deviates from corresponding threshold, then the optimization is done.

Abstract: This disclosure relates generally to Error! Reference source not found.system and method for molecular property prediction. The conventional methods for molecular property prediction suffer from inherent limitation to effectively encapsulate the characteristics of the molecular graph. Moreover, the known methods are computationally intensive, thereby leading to non-performance in real-time scenarios. The disclosed method includes performing self-attention on the nodes of a molecular graph of different sized neighborhood, and further performing a shared attention mechanism across the nodes of the molecular graphs to compute attention coefficients using an Edge-conditioned graph attention neural network (EC-GAT). The EC-GAT effectively utilizes the edge characteristics in the molecular graph for molecular property prediction.

Abstract: This disclosure relates generally to system and method for molecular property prediction. The conventional methods for molecular property prediction suffer from inherent limitation to effectively encapsulate the characteristics of the molecular graph. Moreover, the known methods are computationally intensive, thereby leading to non-performance in real-time scenarios. The disclosed method overcomes the limitations of typical dynamic neighborhood aggregation (DNA) method by fusing the static edge attributes in determining the self-attention coefficients. In an embodiment, the disclosed method transforms the hidden state of a sink node by utilizing a neural-net function, which takes as input an aggregated single-message vector obtained by the self-attention mechanism and the self-attention mechanism transformed hidden state of the node.

Abstract: Industrial plants involve a large amount of equipment, which generate a large amount of data. By analyzing this data, the operator can diagnose anomaly in the plant. Analyzing this data is difficult and time taking task. A method and system for diagnosing anomaly in an industrial system in a time efficient and convenient manner has been provided. The system is configured to diagnose the anomaly by finding out one or more sensors responsible for the anomaly. The present disclosure treats the anomaly detection model as a score generating function. Whenever for a particular instance the score given by the anomaly detection model crosses a pre-determined threshold, anomaly is reported and the diagnosis algorithm is triggered. The system is configured to diagnose the anomaly predicted in case of time series as well as non-time series data.

Abstract: This disclosure relates generally to system and method for molecular property prediction. Typically, message-pooling mechanism employed in molecular property prediction using conventional message passing neural networks (MPNN) causes over smoothing of the node embeddings of the molecular graph. The disclosed system utilizes edge conditioned identity mapping convolution neural network for the message passing phase. In message passing phase, the system computes an incoming aggregated message vector for each node of the plurality of nodes of the molecular graph based on encoded message received from neighboring nodes such that encoded message vector is generated by fusing a node information and an connecting edge information of the set of neighboring nodes of the node. The incoming aggregated message vector is utilized for computing updated hidden state vector of each node.

Abstract: This disclosure relates generally to system and method for molecular property prediction using hypergraph message passing neural network (HMPNN). Typical MPNN architectures used for chemical graphs representation learning have limitations, including, inefficiency to learn long-range dependencies for homogeneous graphs, ineffectiveness to model topological properties of graphs taking into consideration the multiscale representations, and lack of anti-smoothing weighting mechanism to address graphs random walk limit distribution. Disclosed method and system HyperGraph attention-driven Hypergraph Convolution. The Hypergraph attention driven convolution, on molecular hypergraph results in learning efficient embeddings on the high-order molecular graph-structured data. By taking into account the transient incidence matrix, the induced inductive bias augments the scope of molecular hypergraph representation learning.

Abstract: This disclosure relates to method and system for training of deep learning based time-series models based on self-supervised learning. The problem of missing data is taken care of by introducing missing-ness masks. The deep learning model for univariate and multivariate time series data is trained with the distorted input data using the self-supervised learning to reconstruct the masked input data. Herein, the one or more distortion techniques include quantization, insertion, deletion, and combination of the one or more such distortion techniques with random subsequence shuffling. Different distortion techniques in the form of reconstruction of masked input data are provided to solve. The deep learning model performs these different distortion techniques, which force the deep learning model to learn better features. It is to be noted that the system uses a lot of unlabeled data available cheaply as compared to the label or annotated data which is very hard to get.

Abstract: In applications such as adaptive learning of physics-based and data-driven models associated with industrial plants, the models are corrected periodically by taking into consideration the dynamic changes occurring in plant conditions and related data. However, accuracy of adaptive learning depends on accuracy of ground truth data being used as reference data. The disclosure herein generally relates to data preprocessing, and, more particularly, to a method and system for ground truth profile correction and instance selection. The system performs a ground truth profile correction for ground truth profiles having a Profile Deviation Index (PDI) value exceeding a threshold of distortion, to reduce the PDI value, and in turn reduce the distortion in the ground truth profiles. Further, the system performs a data instance selection to identify and remove outliers, and the data that remains after the data instance selection may be then used for applications such as model generation or retuning.

Abstract: Industrial plants involve a large amount of equipment, which generate a large amount of data. By analyzing this data, the operator can diagnose anomaly in the plant. Analyzing this data is difficult and time taking task. A method and system for diagnosing anomaly in an industrial system in a time efficient and convenient manner has been provided. The system is configured to diagnose the anomaly by finding out one or more sensors responsible for the anomaly. The present disclosure treats the anomaly detection model as a score generating function. Whenever for a particular instance the score given by the anomaly detection model crosses a pre-determined threshold, anomaly is reported and the diagnosis algorithm is triggered. The system is configured to diagnose the anomaly predicted in case of time series as well as non-time series data.