Method and system for addressing anomalies in industrial equipment

Abstract

A method (100) and a system (200) for addressing anomalies in an industrial plant are disclosed. The method (100) includes receiving information about anomalies and matching it to historical anomalies represented in the knowledge graph (400). The graph includes nodes representing historical anomalies in the sensor data and historical work orders related to the industrial equipment. The method then retrieves historical work orders associated with the matched anomalies from the knowledge graph, parses the historical work orders to extract information about encountered questions and remedial measures employed, and derives potential remedial measures based on the extracted information. The method further generates an output to provide the derived potential remedial action to an operator of the industrial plant.

Description

Technical Field

[0001] This invention generally relates to the field of industrial equipment maintenance and anomaly solutions. More specifically, this invention relates to computer-implemented methods and systems that utilize knowledge graphs and machine learning techniques to detect anomalies in industrial equipment, link these anomalies to past anomalies and associated work orders, and generate potential remedial measures based on historical data. Background Technology

[0002] In various industries, anomalies in industrial equipment frequently lead to factory malfunctions or complete shutdowns. These anomalies can be characterized by sudden changes in sensor behavior, categorized as step changes or ramp changes. Following anomalies, investigations are typically conducted to identify the root cause of the problem, which usually traces back to these anomalous behaviors. Addressing such anomalies is a pressing need, and there is a growing demand for proactive and intelligent methods for anomaly detection and resolution, as these methods can help factory operators anticipate and resolve potential issues before they escalate into major problems.

[0003] Traditionally, rule-based and machine learning (ML) techniques have been used for anomaly detection in industrial environments. Some methods use time-series-based algorithms to predict ramp-up anomalies in advance. Additionally, solution recommendation systems exist that build upon analysis of all historical work orders. Using fuzzy search and text matching principles, these systems can locate the most relevant work orders and their solutions. However, these methods often focus on actual downtime and disruption, rather than necessarily on slowly emerging issues that may or may not lead to downtime.

[0004] A major challenge in the current state of anomaly detection and resolution is the lack of integrated data sources. Data silos are rampant, with each source having its own set of standards and symbols. Existing solutions also fail to adequately establish a link between detected anomalies and work orders, and therefore cannot provide a comprehensive understanding of the problem. Consequently, these solutions are reactive and often require significant manual intervention, which can lead to delays and increased costs in problem resolution.

[0005] More sophisticated methods for anomaly detection have been developed, including various machine learning models. However, these models typically require large amounts of labeled data for training and are sensitive to the quality and representativeness of the training data. Furthermore, such models often operate as black boxes, providing little insight into the underlying causes of detected anomalies or their predicted causes, making it difficult for operators to take informed action.

[0006] Knowledge graphs (KGs) are rarely used in this context, and the literature indicates a lack of such technology. Knowledge graphs can map different sets of information and derive relationships between anomalies and work orders, thereby capturing individual representations of anomalies to provide an integrated and comprehensive source of information.

[0007] In addition, Natural Language Processing (NLP) techniques have been used to extract information from work orders. However, this approach often lacks the ability to accurately segment problems, causes, and remedies, especially when the information is embedded in unstructured natural language text.

[0008] This invention seeks to overcome these challenges by providing a computer-implemented method and system for resolving anomalies in industrial equipment, which addresses these limitations. The invention provides a comprehensive representation of anomalies and associated work orders, facilitating proactive anomaly detection and resolution, and offering interpretability and insight into the potential causes of anomalies. Furthermore, the invention effectively extracts and segments problem, cause, and remedial information from work orders, even when this information is embedded in unstructured natural language text. Summary of the Invention

[0009] The objective of this invention is achieved through a computer-implemented method for resolving anomalies in industrial equipment. The method includes: receiving information about anomalies in the industrial equipment. The method further includes: matching the anomaly with at least one of historical anomalies represented in a knowledge graph, wherein the knowledge graph includes nodes representing historical anomalies from sensor data of the industrial equipment and historical work orders related to the industrial equipment, wherein each historical anomaly in the sensor data is associated with one or more corresponding historical work orders related to the industrial equipment. The method further includes: retrieving from the knowledge graph one or more corresponding historical work orders related to the industrial equipment that are associated with at least one historical anomaly matching the anomaly. The method further includes: parsing the retrieved one or more historical work orders related to the industrial equipment to extract information related to problems encountered by the industrial equipment and corresponding remedial measures adopted. The method further includes: deriving potential remedial measures for the anomalies in the industrial equipment based on the extracted information. The method further includes: generating an output providing the derived potential remedial measures for the operator of the industrial equipment.

[0010] In one or more embodiments, the knowledge graph includes historical anomalies in the sensor data and historical work orders related to the industrial equipment, represented as nodes in a Resource Description Framework (RDF) graph constructed based on an ontology schema defined within the domain of the industrial equipment.

[0011] In one or more embodiments, creating a node for each historical anomaly in the sensor data includes associating anomaly-specific information with each node, the anomaly-specific information including one or more of the following: the time of occurrence of the historical anomaly, the sensor reading at the time of the historical anomaly, the operating parameters of the industrial equipment at the time of the historical anomaly, and the severity of the historical anomaly.

[0012] In one or more embodiments, creating a node for each historical work order related to the industrial equipment includes associating work order-specific information with each node, the work order-specific information including one or more of the following: the work order ID of the historical work order, the generation time of the historical work order, a problem description in the historical work order, a remedial action description in the historical work order, and operator information in the historical work order.

[0013] In one or more embodiments, anomalies related to the industrial equipment are detected based on monitoring sensor data from the industrial equipment via an autoencoder model trained under normal operating conditions of the industrial equipment, wherein deviations in the reconstruction of sensor data from the industrial equipment via the autoencoder model are detected as anomalies.

[0014] In one or more embodiments, matching the anomaly with at least one of the historical anomalies represented in the knowledge graph includes: using a similarity score derived from a subgraph matching process, the subgraph matching process comparing subgraphs associated with anomalies related to the industrial equipment with subgraphs associated with historical anomalies in the knowledge graph, wherein the similarity score indicates the degree of feature similarity between the compared subgraphs.

[0015] In one or more embodiments, parsing one or more retrieved historical work orders related to the industrial equipment includes: applying a natural language processing (NLP) pipeline to the text of the retrieved one or more historical work orders to segment the text for identifying problems encountered by the industrial equipment and corresponding remedies adopted.

[0016] In one or more embodiments, the NLP pipeline is configured to perform one or more of the following stages to derive structured information from one or more texts retrieved from historical work orders for text segmentation: text preprocessing, named entity recognition, semantic role labelling, and relation extraction.

[0017] In one or more embodiments, the method further includes: applying an algorithm based on Point Mutual Information (PMI) scores to the segmented text as output of the NLP pipeline to classify problems and remedies among one or more texts retrieved from historical work orders; and applying the SHAP (SHapley Additive exPlanations) algorithm to interpret the contribution of each word in the classified text to the identification of the problem encountered by the industrial equipment and the corresponding remedy adopted.

[0018] In one or more embodiments, deriving potential remedies for the anomaly includes applying a machine learning algorithm to the extracted information, the machine learning algorithm being trained on a dataset including the historical work orders and anomalies associated with the historical work orders, and the machine learning algorithm being configured to predict potential remedies for the anomaly based on the historical data.

[0019] In one or more embodiments, generating output for the operator includes presenting the operator with a list of derived potential remedies, the derived potential remedies being sorted based on one or more of the following: the historical effectiveness of each of the derived potential remedies, the time spent implementing each of the derived potential remedies, the cost involved in implementing each of the derived potential remedies, and operator feedback for each of the derived potential remedies.

[0020] In one or more embodiments, the method further includes dynamically updating the knowledge graph based on one or more of new sensor data from the industrial equipment, detected anomalies, processed work orders, and operator feedback.

[0021] In one or more embodiments, the method further includes: utilizing a time-series prediction model trained on historical sensor data, the time-series prediction model being configured to identify degradation patterns over time in the sensor data to detect potential future anomalies in the industrial equipment.

[0022] The object of the present invention is also achieved by a system comprising one or more processing units and a memory unit configured to perform the steps of the computer-implemented method according to any of the foregoing method steps.

[0023] The object of the present invention is further achieved by a computer program product having machine-readable instructions stored therein, which, when executed by one or more processing units, cause the processing units to perform the aforementioned method steps.

[0024] Other aspects, features, and advantages of the invention will become apparent from the following detailed description merely by illustrating several specific embodiments and implementations, including those of the best mode contemplated for carrying out the invention. The invention is also capable of other and different embodiments, and several details thereof may be modified in various obvious ways without departing from the scope of the invention. Therefore, the drawings and description are to be considered substantially illustrative rather than restrictive. Attached Figure Description

[0025] A more complete understanding of the invention and its many accompanying aspects will become readily apparent and better understood when considered in conjunction with the accompanying drawings and by referring to the following description: Figure 1 This is a flowchart representation of a computer-implemented method for resolving anomalies in industrial equipment according to one or more embodiments of the present invention; Figure 2 This is a block diagram representation of a system for resolving anomalies in industrial equipment according to one or more embodiments of the present invention; Figure 3 This is a diagram of an exemplary work order according to one or more embodiments of the present invention; Figure 4 This is an exemplary knowledge graph according to one or more embodiments of the present invention; and Figure 5 It is a flowchart of a workflow performed according to one or more embodiments of the present invention.

[0026] Various embodiments are described with reference to the accompanying drawings, wherein the same reference numerals are used throughout to refer to the same elements in the drawings. In the following description, numerous specific details are set forth for purposes of explanation in order to provide a thorough understanding of one or more embodiments. It will be apparent that such embodiments can be practiced without these specific details. Detailed Implementation

[0027] This document discloses examples of methods, systems, and computer program products for resolving anomalies in industrial equipment. In the following description, numerous specific details are set forth for purposes of explanation in order to provide a thorough understanding of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details or with equivalent arrangements. In other instances, well-known structures and apparatuses are illustrated in block diagram form to avoid unnecessarily obscuring embodiments of the invention.

[0028] Now for reference Figure 1 The illustration shows a flowchart of a method for resolving anomalies in industrial equipment (as indicated by reference numeral 100) according to an embodiment of the present invention. As used herein, the term "industrial equipment" refers to various types of machines and devices used in industrial environments, including but not limited to manufacturing plants, power generation facilities, pharmaceutical production units, and other industrial contexts. Industrial equipment can encompass a wide variety of devices, systems, and machines, such as turbines, ovens, pharmaceutical manufacturing units, etc., each of which can be equipped with various sensors to monitor their operating status and performance. These sensors generate data that can be used to detect anomalies, predict potential functional failures, and guide maintenance activities. Furthermore, the term "anomaly" refers to abnormal or undesirable behavior in the operation of industrial equipment, typically detected through analysis of sensor data. These anomalies may be caused by sudden changes in sensor behavior, which can be categorized as step changes or ramp changes. In some cases, anomalies can indicate potential functional failures or shutdowns of the unit or the entire plant, thus requiring immediate attention and resolution.

[0029] refer to Figure 2The illustration is a block diagram of a system 200 for resolving anomalies in industrial equipment according to one or more embodiments of the present invention. It will be understood that the system 200 described herein can be implemented in various forms of hardware, software, firmware, dedicated processors, or combinations thereof. One or more of the embodiments herein can take the form of a computer program product comprising program modules accessible from a computer-usable or computer-readable medium storing program code for use by or in connection with one or more computers, processors, or instruction execution systems. For the purposes of this description, the computer-usable or computer-readable medium can be any means capable of containing, storing, transmitting, propagating, or transporting a program for use by or in connection with an instruction execution system, apparatus, or device. The medium can be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or device or apparatus), or a transmission medium serving as a signal carrier, and is not, by itself, included in the definition of a physical computer-readable medium, which includes semiconductor or solid-state memory, magnetic tape, removable computer floppy disks, random access memory (RAM), read-only memory (ROM), hard disks, and optical disks (such as high-density disk read-only memory (CD-ROM), high-density disk read / write, and digital versatile disks (DVDs)). As is known to those skilled in the art, both the processor and program code used to implement each aspect of this technology can be centralized or distributed (or a combination thereof).

[0030] In one example, system 200 may be embodied as a computer program product programmed to implement the stated purpose. System 200 may be incorporated into one or more physical packages (e.g., chips). For example, a physical package includes an arrangement of one or more materials, components, and / or wires on a structural assembly (e.g., a substrate) to provide one or more properties, such as physical strength, dimensional conservation, and / or limitations on electrical interactions. Contemplate that, in some embodiments, the computing device may be implemented in a single chip. As illustrated, system 200 includes communication mechanisms, such as bus 202, for transferring information between components of system 200. System 200 includes one or more processing units 204 and one or more memory units 206. Hereinafter, memory units 206 are communicatively coupled to processing units 204. In one example, memory unit 206 may be embodied as a computer-readable medium storing program code segments of a computer program that can be loaded into and / or executed within the system to cause system 200 to perform steps for implementing the stated purpose.

[0031] Generally, as used herein, the term "processing unit" refers to a computing element operable to respond to and process instructions of the drive system 200. Optionally, a processing unit includes, but is not limited to, a microprocessor, microcontroller, Complex Instruction Set Computing (CISC) microprocessor, Reduced Instruction Set Computing (RISC) microprocessor, Very Long Instruction Word (VLIW) microprocessor, or any other type of processing circuitry. Furthermore, the term "processing unit" can refer to one or more individual processors, processing devices, and various elements associated with processing devices that can be shared by other processing devices. Additionally, one or more individual processors, processing devices, and elements are arranged in various architectures for responding to and processing instructions of the drive system 200.

[0032] In this document, memory unit 206 may be volatile memory and / or non-volatile memory. Memory unit 206 may be coupled for communication with processing unit 204. Processing unit 204 may execute instructions and / or code stored in memory unit 206. Various computer-readable storage media may be stored in and accessed in memory unit 206. Memory unit 206 may include any suitable elements for storing data and machine-readable instructions, such as read-only memory, random access memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, hard disk drive, removable media drive for handling high-density disks, digital video optical discs, floppy disks, tape cartridges, memory cards, etc.

[0033] Specifically, processing unit 204 has a connection to bus 202 to execute instructions and process information stored in memory unit 206. Processing unit 204 may include one or more processing cores, each configured to execute independently. Multi-core processors enable multiple processing within a single physical package. Examples of multi-core processors include two, four, eight, or more processing cores. Alternatively or additionally, processing unit 204 may include one or more microprocessors configured in series via bus 202 to enable independent execution of instructions, pipelining, and multithreading. Processing unit 204 may also be accompanied by one or more dedicated components to perform certain processing functions and tasks, such as one or more digital signal processors (DSPs) and / or one or more application-specific integrated circuits (ASICs). Other dedicated components that help perform the functions of the invention described herein include one or more field-programmable gate arrays (FPGAs) (not shown), one or more controllers (not shown), or one or more other dedicated computer chips.

[0034] System 200 may further include an interface 208, such as a communication interface (the terms are used interchangeably), which enables system 200 to communicate with other systems to receive and transmit information. Communication interface 208 may include a medium (e.g., a communication channel) through which system 200 communicates with other systems. Examples of communication interface 208 may include, but are not limited to, communication channels in a computer cluster, local area communication channels (LANs), cellular communication channels, wireless sensor communication channels (WSNs), cloud communication channels, metropolitan area communication channels (MANs), and / or the Internet. Optionally, communication interface 208 may include one or more of wired connections, wireless networks, cellular networks such as 2G, 3G, 4G, and 5G mobile networks, and Zigbee connections.

[0035] System 200 also includes database 210. As used herein, database 210 is an organized collection of structured data, typically stored in a computer system, and designed to be easily accessed, managed, and updated. Database 210 may take the form of a central information store that can be queried, analyzed, and processed to support various applications and business processes. In system 200, database 210 provides mechanisms for storing, retrieving, updating, and deleting data, and typically includes functions such as data validation, security, backup and recovery, and data modeling.

[0036] System 200 further includes an input device 212 and an output device 214. Depending on the specific application of system 200, the input device 212 can take various forms. In one example, the input device 212 may include one or more of a keyboard, mouse, touchscreen display, microphone, camera, or any other hardware component that enables a user to interact with system 200. Furthermore, the output device 214 may take the form of a monitor, printer, etc. It should be understood that when the term "display" is used in this invention, it generally refers to either the display screen itself or the screen and associated housing, driving circuitry, and possibly the physical support structure, all or part of which are provided for displaying information.

[0037] In this system 200, the processing unit 204 and accompanying components have a connection to the memory unit 206 via a bus 202. The memory unit 206 includes both dynamic memory (e.g., RAM, disk, writable optical disc, etc.) and static memory (e.g., ROM, CD-ROM, etc.) for storing executable instructions that, when executed, implement the method steps described herein for resolving anomalies in industrial equipment. Specifically, the memory unit 206 includes a module set 216 for performing steps for resolving anomalies in industrial equipment. Furthermore, in system 200, the memory unit 206 can be configured to store data associated with or generated by the execution of the steps of the invention.

[0038] Joint Reference Figure 1 and Figure 2 The various steps of method 100, as described below, can be executed in system 200, or specifically in processing unit 204 of system 200, for resolving anomalies in industrial equipment. For the purposes of this invention, method 100 is embodied as an anomaly resolution algorithm for resolving anomalies in industrial equipment. It will be understood that although method 100 is illustrated and described as a sequence of steps, it is contemplated that various embodiments of method 100 can be performed in any order or combination, and it is not necessary to include all the illustrated steps.

[0039] In an embodiment of the invention, at step 110, method 100 includes receiving information about anomalies in industrial equipment (hereinafter sometimes simply referred to as "equipment"). Step 110 involves collecting and processing data indicating abnormal or undesirable behavior during the operation of the industrial equipment. Such information can come from various sources, such as sensors installed on the equipment, and can include data such as temperature readings, pressure levels, operating parameters, and other relevant data points. Hereinafter, processing unit 204 may receive information about anomalies in the industrial equipment via a direct connection to sensors of the equipment, via a network connection, or by accessing a database or other storage system where information is stored. In this embodiment, method 100 may include detecting anomalies in the industrial equipment. That is, method 100 itself may involve steps to detect anomalies in the industrial equipment, rather than receiving information about anomalies from some other source.

[0040] refer to Figure 3The illustration depicts an exemplary work order (as indicated by reference numeral 300) according to one or more embodiments of the present invention. As depicted, work order 300 represents a comprehensive record of a functional failure occurring in a particular industrial machine. Work order 300 details the specific problem, the cause of the problem, and the solution implemented to resolve the issue. All these details are interconnected in a diagram, where various aspects of work order 300 are represented as nodes. It will be understood that the work order itself is represented as a node and includes the text of the work order, thereby providing comprehensive information about the issue, its cause, and the implemented solution. The node also provides time information, such as when the work order was created and when the functional failure began.

[0041] Furthermore, for example, a node representing a faulty machine contains key information such as the machine's name and additional details about it, such as its function or location. This node is connected to another node representing the specific component of the machine that failed, which includes its name and the method used to detect the failure, such as a dependency parser used in natural language processing. Additionally, the problem identified in work instruction 300 is represented as a separate node containing a detailed description of the problem. The cause of the problem also has its own node, containing a description of the root cause of the issue. Finally, the solutions implemented to solve the problem are represented as nodes containing a detailed description of the actions taken.

[0042] It is understood that, in this paper, the schema of the metadata for work order 300 can include the problem, remedy, cause, fault, solution, faulty device, delay, etc., along with the text of the work order (which can be a single sentence). The sentence representation can include linguistic aspects of the text, along with one or more of the following: token containers, NER containers, SRL containers, relation containers, clauses, and dependency parser containers. The integrated representation source on the ontology schema inherits the inference and reasoning of the graph structure and schema. This graph, in itself, has the ability to derive insights from inferences made based on an ontology structure with node generalization and representation. The integrated representation source on the ontology schema inherits the inference and reasoning of the graph structure and schema. Such a graph, in itself, has the ability to derive insights from inferences made based on an ontology structure with node generalization and representation.

[0043] Therefore, the graph structure utilized in this invention provides a comprehensive and structured representation of the work order 300. This graph effectively links anomalies (failures of machines and their specific components) to the problems, causes, and solutions described in detail in the work order. This structure allows for the retrieval of valuable historical information when new anomalies are detected, facilitating the identification of potential problems and remedial measures based on past incidents.

[0044] In one embodiment, anomalies related to industrial equipment are detected based on monitoring sensor data from the equipment. Specifically, an autoencoder is trained based on sensor data collected under normal operating conditions. This means the autoencoder learns to accurately encode and decode data patterns representing the normal, expected behavior of the equipment. Once trained, the autoencoder can be used to monitor new sensor data from the industrial equipment. The autoencoder attempts to recreate the incoming sensor data based on its learned model. If the equipment is operating under normal conditions, the autoencoder can be able to accurately reconstruct the sensor data. However, if anomalies exist, i.e., deviations from normal operation, this can lead to higher reconstruction errors. This deviation in the reconstruction of the sensor data is detected as an anomaly. In other words, if the autoencoder fails to accurately reconstruct the sensor data, this indicates that the current sensor readings represent an abnormal condition or anomaly in the operation of the equipment. This approach is advantageous because it does not rely on predefined thresholds or rules, but instead learns "what is considered normal operation" from the data itself, thus enabling it to detect a wide range of potential anomalies.

[0045] As used in this paper, autoencoders are a type of artificial neural network used to learn efficient data representations, often for dimensionality reduction purposes. These networks are unsupervised learning models, meaning they do not require labeled data for training. Instead, these networks learn to encode input data into a compressed representation and then decode that representation back into the original data format. The goal is to minimize the difference between the original input and the reconstructed output—known as the reconstruction error. An autoencoder consists of two main parts: an encoder and a decoder. The encoder transforms the input data into different, typically lower-dimensional, representations. The decoder then attempts to reconstruct the original input data from this encoded representation. The network is trained to minimize the difference between the original input and the reconstructed output, effectively forcing the network to learn a compact representation of the input data.

[0046] This method 100 utilizes a knowledge graph, which allows for modeling complex relationships between different types of entities. A knowledge graph is typically a network where nodes represent entities, and edges represent relationships or associations between those entities. In this invention, entities are historical anomalies from sensor data of industrial equipment and historical work orders related to the industrial equipment. Each node in the knowledge graph represents an anomaly or work order. Anomaly nodes include data about past anomalies detected in the equipment, such as the nature of the anomaly, the time of occurrence, the sensor readings at that time, and other relevant operating parameters. Work order nodes include information generated in response to these anomalies from work orders, such as a description of the problem, proposed remedies, and the results of the implemented remedies. Furthermore, the edges of the knowledge graph represent the relationships between these anomalies and work orders. Specifically, an edge from an anomaly node to a work order node indicates that the anomaly is associated with a corresponding work order. This relationship means that the anomaly was identified as a problem in the work order, and a remedy was adopted to resolve the anomaly.

[0047] Therefore, in this paper, knowledge graphs allow for the mapping of complex relationships between anomalies and work orders, thereby capturing a comprehensive representation of the individual characteristics of anomalies and their associated work orders. Specifically, representations in knowledge graphs allow for understanding the context of anomalies, retrieving relevant work orders, and generating potential remedial measures based on historical data. Knowledge graph-based approaches enhance the ability to derive meaningful insights from data and make informed decisions about anomaly solutions. Using knowledge graphs in this way allows for a better understanding of anomalies at hand by considering not only the anomaly itself but also the historical context in which it occurred and the multiple actions taken in response. This enables more effective and informed decision-making when addressing anomaly issues.

[0048] refer to Figure 4The illustration depicts an exemplary knowledge graph (as indicated by reference numeral 400) according to one or more embodiments of the present invention. As depicted, knowledge graph 400 includes multiple nodes and edges. Each node represents an entity, either an anomaly from sensor data of an industrial device or a historical work order related to the industrial device. For example, knowledge graph 400 is shown as including an anomaly node related to "Anomaly 10" and a work order node related to "Work Order 4". In knowledge graph 400, these nodes have been visually represented as having an elliptical shape. Each such node can be connected to other nodes that provide additional details, such as to provide more context about the node. For example, an anomaly node can be connected to other nodes that provide details including the nature of the anomaly, the time when the anomaly occurred, the sensor readings at that time, etc., such as the input pressure at the time of the anomaly with the corresponding reading, the motor frequency at the time of the anomaly with the corresponding reading, the date of the anomaly with the corresponding reading, etc. (as shown). Similarly, for example, a work order node can be connected to other nodes that provide additional details, such as details about the associated work order, including a description of the problem and the remedies taken. Such nodes can be represented in different shapes, such as circles or rectangles as shown in Knowledge Graph 400. Furthermore, as shown, exception nodes and work order nodes can be connected to other exception nodes and work order nodes, respectively. Edges in Knowledge Graph 400 represent the relationships between these entities. Specifically, an edge from an exception node to a work order node indicates that an exception is associated with a corresponding work order. This is visually represented by arrows and lines connecting the relevant nodes.

[0049] In one embodiment, the knowledge graph 400 includes historical anomalies in sensor data and historical work orders related to industrial equipment, represented as nodes in a Resource Description Framework (RDF) graph. That is, the knowledge graph 400 is constructed as an RDF graph—a widely used standard for representing data in networks. RDF graphs allow information to be modeled in the form of subject-verb-object triples, which is particularly suitable for representing complex, interconnected information about historical anomalies and associated work orders. Within the RDF graph, individual historical anomalies in sensor data and individual historical work orders related to industrial equipment are represented as nodes. These nodes include key information about the anomalies and work orders, such as the nature of the anomaly, the time of the anomaly, the sensor readings at the time, the problem description in the work order, the remedial action description, and other relevant details. The relationships between these anomalies and work orders are represented as edges in the RDF graph. For example, an edge can connect anomaly nodes to work order nodes, indicating that the anomaly is associated with a corresponding work order. These edges effectively capture the association between anomalies and work orders, thus providing a comprehensive representation of the historical context of the anomaly's occurrence and the actions taken in response.

[0050] RDF graphs are constructed based on ontology architecture patterns defined within the domain of industrial equipment. An ontology is a formal representation of knowledge within a specific domain, typically including a vocabulary of terms and detailed descriptions of the meanings of those terms. This includes definitions of classes, properties, and their relationships. In this context, ontology architecture patterns can define the types of nodes (such as exceptions and work orders), the types of relationships (such as the association between exceptions and work orders), and other constraints or rules regarding the structure of the RDF graph. By fully leveraging RDF graphs and ontology architecture patterns, Knowledge Graph 400 enables the capture, representation, and reasoning of complex relationships in data for exception resolution.

[0051] It is understandable that there may be similar nodes / entities from different data sources, namely work orders and anomalies associated with the primary contributor's device tag. The idea behind this graph is to generalize nodes and relationships as much as possible to reduce the feature space dimensionality of the graph's vectorized representation and to leverage ontological architecture patterns to obtain better and more concrete representations. This method 100 involves vectorizing a pre-trained model to generalize relationships and nodes (or entities) along with a domain taxonomy for entities, if available. This allows processes to have relationships of similarity for faults using domain taxonomy (such as semantic generality). These generalized representations can create associations on the graph with work order and anomaly representations.

[0052] In one embodiment, creating a node for each historical anomaly in the knowledge graph 400 involves associating anomaly-specific information with each node (i.e., each node has specific information related to the anomaly). This anomaly-specific information includes key details that help create a comprehensive representation of each anomaly, identify patterns, and derive meaningful insights for future anomaly resolution. The information associated with each anomaly node includes the specific date and time the anomaly was detected, which can be useful for understanding any temporal patterns or trends during the anomaly's occurrence. Additionally, it includes sensor readings from various sensors installed on the industrial equipment at the time of the anomaly. These readings represent a wide variety of operating parameters, particularly such as temperature, pressure, and vibration levels. The specific operating parameters of the industrial equipment at the time of the anomaly (such as operating speed, load, or the specific task being performed) are also part of the information associated with the anomaly node. Finally, it includes an assessment of the anomaly's severity, which can be determined based on the level of disturbance to machine operation, potential or actual damage to the machine, or the risk to safety. By combining all this information, the knowledge graph 400 provides a repository of historical data on anomalies, enabling new anomalies to be matched more accurately with historical anomalies and facilitating more intelligence-based decision-making regarding potential remedies.

[0053] In one embodiment, creating a node for each historical work order related to industrial equipment includes integrating work order-specific information with each node (i.e., each node has specific information related to the work order). This process helps compile a detailed account for each work order, which facilitates the extraction of valuable insights for handling future anomalies. The work order-specific information associated with each node incorporates several important aspects. The work order ID of a historical work order provides a unique identifier for each work order, facilitating easy referencing and retrieval. The generation time of the historical work order provides the temporal context of "when the anomaly was recorded and resolved," which can be useful in observing trends or patterns over time. Furthermore, the problem description in the historical work order provides a detailed explanation of the problem encountered by the industrial equipment, thus providing crucial insights into the nature of the anomaly. On the other hand, the remedial action description in the historical work order outlines the steps taken to mitigate the problem, which can be invaluable when searching for solutions to similar problems in the future. In addition, operator information in the historical work order can include details such as who reported the anomaly, who handled the anomaly, and any comments or observations they made. By associating each work order node with these specific details, Knowledge Graph 400 becomes a comprehensive repository of historical data on anomaly handling. This helps to more accurately match new anomalies with historical work orders, understand the specific circumstances and context of past anomalies, and make more intelligence-based decisions regarding potential remedial measures.

[0054] At step 120, method 100 includes matching the anomaly with at least one of historical anomalies represented in a knowledge graph (such as knowledge graph 300). That is, comparing or matching the detected anomaly in the industrial equipment with historical anomalies represented in the knowledge graph. This matching process aims to find similar patterns or conditions from the past, which can help understand and resolve the current anomaly. As discussed, the knowledge graph includes nodes representing historical anomalies and their associated work orders. Each node encapsulates various details about the anomaly or work order, such as the nature of the anomaly, the time of occurrence, the sensor readings at the time, the problem description in the work order, the remedial action description, and other operating parameters. The matching process can involve various techniques, such as similarity scoring, subgraph matching, or other machine learning algorithms. The goal is to find historical anomalies in the knowledge graph that share similar features or characteristics with the current anomaly. For example, method 100 may involve comparing sensor readings, operating parameters, or other features associated with the current anomaly with those of historical anomalies. If a historical anomaly has similar features to the current anomaly, a match can be considered achieved.

[0055] In one embodiment, matching an anomaly with at least one of historical anomalies represented in a knowledge graph includes: utilizing a similarity score derived from a subgraph matching process that compares subgraphs associated with anomalies related to industrial equipment and subgraphs associated with historical anomalies in the knowledge graph. Here, the similarity score indicates the procedure for assessing the feature similarity between the compared subgraphs. Specifically, the similarity score is derived from a subgraph matching process that compares subgraphs associated with the current anomaly and subgraphs associated with historical anomalies represented in the knowledge graph. It is understood that a subgraph is a subset of the nodes and edges of a graph. In this example, a subgraph may consist of nodes representing anomalies (current or historical), nodes representing associated work orders, and edges connecting these nodes. Therefore, each subgraph includes details of a specific anomaly and its associated work orders.

[0056] The subgraph matching process involves comparing subgraphs associated with the current anomaly and subgraphs associated with historical anomalies. This comparison can involve various aspects of the subgraph, such as the number and type of nodes, the number and type of edges, node attributes (such as sensor readings at the time of the anomaly), and the subgraph's topology (how nodes are connected by edges). The result of the subgraph matching process is a similarity score for each historical anomaly. The similarity score quantifies the degree of similarity between the subgraph of the current anomaly and the subgraphs of each historical anomaly. A high similarity score indicates a strong similarity between the features of the current anomaly and the features of historical anomalies, suggesting that they are closely related.

[0057] By considering not only the anomaly itself, but also related work orders and the relationships between them, the historical anomalies most relevant to the current anomaly are identified. Typically, step 120 fully utilizes historical data to gain insights into the potential causes and solutions for the current anomaly. By finding historical anomalies similar to the current anomaly, the system can retrieve related work order documents that can provide valuable information about potential problems and remedies related to the anomaly.

[0058] At step 130, method 100 includes retrieving from the knowledge graph one or more corresponding historical work orders related to the industrial equipment that are associated with at least one historical anomaly matching the anomaly in the historical anomalies. Here, the retrieval of these historical work orders is facilitated by a structured representation of the data within the knowledge graph (such as knowledge graph 400). Specifically, after an anomaly in the industrial equipment has been detected and matched with at least one historical anomaly in the knowledge graph, method 100 then proceeds to retrieve the corresponding historical work orders. These work orders are directly linked to the matched historical anomaly in the knowledge graph, thus representing past instances of "when similar anomalies occurred and were recorded."

[0059] The retrieval process involves traversing the knowledge graph from nodes representing matched historical anomalies to nodes representing associated work orders. This can be accomplished using graph traversal algorithms such as depth-first search or breadth-first search to efficiently navigate the network of nodes and edges in the knowledge graph. The retrieved historical work orders provide valuable information about past instances when similar anomalies occurred. These instances may include details about the problems encountered, the steps taken to resolve them, the effectiveness of those steps, and other operational parameters at the time of the anomaly. This information can then be used to understand the current anomaly and derive potential remedial measures.

[0060] At step 140, method 100 includes parsing one or more retrieved historical work orders related to industrial equipment to extract information related to problems encountered by the industrial equipment and corresponding remedial actions taken. This parsing process allows for the extraction of necessary details from the work orders, particularly information related to the problems encountered by the industrial equipment and the remedial actions taken to resolve those problems. Specifically, in this context, parsing involves processing the text of historical work orders to identify and extract specific pieces of information of interest. The goal of the parsing process is to extract two main pieces of information from each historical work order: the problem and the remedial action. The problem refers to the issue encountered by the industrial equipment, as described in the work order. This may include details about the nature of the anomaly, its impact on machine operation, and any other relevant information. The remedial action refers to the actions taken to resolve the problem, as recorded in the work order. This may involve a description of the steps taken, the tools or components used, the personnel involved, and the results of the remedial action.

[0061] In one embodiment, parsing one or more retrieved historical work orders related to industrial equipment includes applying a Natural Language Processing (NLP) pipeline to the text of the retrieved one or more historical work orders to segment the text for identifying problems encountered by the industrial equipment and corresponding remedial actions taken. "Applying the NLP pipeline to the text of historical work orders" is used to segment the text into different parts, each part corresponding to a different aspect of the information contained in the work order. Specifically, the pipeline helps identify text segments describing problems encountered by the industrial equipment and text segments describing the corresponding remedial actions taken. This may involve various NLP techniques, particularly such as tokenization, part-of-speech tagging, named entity recognition, and relation extraction. These techniques help break down text into manageable units, identify the roles of different words or phrases, and establish relationships between them.

[0062] Furthermore, the NLP pipeline is configured to perform one or more of the following stages to derive structured information from one or more texts retrieved from historical work orders for text segmentation: text preprocessing, named entity recognition, semantic role labeling, and relation extraction. In other words, the NLP pipeline can include several stages. Initial stages may include text preprocessing tasks such as word segmentation, where text is broken down into individual words or 'lexicons'; and normalization, where text is transformed into a standard form (e.g., by converting all characters to lowercase or by removing punctuation). Subsequent stages may involve more complex tasks such as part-of-speech tagging, where each word is labeled with its grammatical role in a sentence; named entity recognition, where entities (such as the names of machine parts or specific anomalies) are identified; and dependency parsing, where grammatical relations between words are established. The final stage of the pipeline may involve text segmentation and information extraction. In these stages, the preprocessed and analyzed text is divided into segments, each segment corresponding to a specific piece of information. For example, one section might contain a description of a problem encountered by industrial equipment, while another section might contain a description of the remedial measures taken. The information within these sections is then extracted and stored in a structured format, ready for use in subsequent steps of the troubleshooting process.

[0063] In some embodiments, method 100 further includes applying advanced algorithms to segmented text output from an NLP pipeline, namely, algorithms based on Point Mutual Information (PMI) scores and the SHAP (Shapley Additive Interpretation) algorithm. First, the PMI-based algorithm is applied. PMI is a statistical measure in information theory and statistics used to quantify the degree of association between two events. In this invention, the PMI-based algorithm is used to classify text segments into two categories: problems and remedies. This classification is based on PMI scores of words or phrases in the text segments that reflect their association with the concepts of 'problem' and 'remedy'. Segments most strongly associated with the concept of 'problem' (based on the PMI score of said segment) are classified as describing problems encountered by industrial equipment, while segments most strongly associated with 'remedies' are classified as describing the corresponding remedies adopted. After classifying the text segments, the SHAP algorithm is applied. SHAP is a game-theoretic method for interpreting the output of any machine learning model. It uses classic Shapley values ​​from cooperative game theory to connect optimal credit allocation with local interpretation. In this invention, the SHAP algorithm is used to interpret the contribution of each word in the classified text to the identification of the problem and remedy. This involves calculating the SHAP value of a word or phrase in the text, which represents the average contribution of each word to the classification decision across all possible word combinations.

[0064] By applying PMI score-based algorithms and the SHAP algorithm to segmented text, Method 100 effectively classifies text into problem segments and remedial action segments, and understands the contribution of each word to that classification. This allows for more accurate and interpretable information extraction from historical work orders, thus aiding in problem identification and corresponding remedial actions. This, in turn, supports more informed and effective methods for addressing current and future anomalies in industrial equipment.

[0065] At step 150, method 100 includes deriving potential remedies for anomalies in industrial equipment based on the extracted information. This step 150 leverages insights gathered from the past to inform solutions to the current anomaly. The derivation of potential remedies involves analyzing problem and remedy information extracted from historical work orders, particularly those identified as being associated with historical anomalies similar to the current anomaly. This analysis can provide valuable insights into “what kinds of issues tend to arise under similar conditions” and “which solutions have effectively addressed these issues in the past.” Such a process may involve various techniques, depending on the complexity of the problem and the available data. For example, it may involve a simple lookup of the most common remedies associated with similar problems. Alternatively, it may involve more sophisticated techniques, such as machine learning algorithms trained on historical data, to predict potential remedies for a given problem.

[0066] In one embodiment, deriving potential remedies for anomalies involves applying a machine learning algorithm to the extracted information. This machine learning algorithm is trained on a dataset including historical work orders and their associated anomalies, and is configured to predict potential remedies for anomalies based on historical data. In this document, the machine learning algorithm is configured to predict potential remedies for anomalies based on historical data. This prediction is based on patterns and relationships learned by the algorithm during its training process. For example, if the algorithm has learned that a certain type of anomaly is frequently resolved successfully with a particular remedy, it can predict that this remedy is likely a suitable solution for current anomalies of the same type. In this way, the machine learning algorithm contributes to anomaly resolution by proposing potential remedies based on past experience. This method fully leverages information contained in historical work orders to inform decision-making, potentially leading to more effective and efficient anomaly solutions.

[0067] refer to Figure 5The illustration shows a flowchart of a workflow (as indicated by reference numeral 500) performed by the system 200 of the present invention. As depicted, workflow 500 can run simultaneously on two parallel pipelines: a health data workflow and a work instruction data workflow. Each pipeline includes a series of steps, which are represented as boxes in the workflow, with arrows indicating the movement of data between steps.

[0068] In this paper, at box 502, the health data workflow begins with the step of acquiring health domain data. This is the initial stage, where system 200 receives information about the operating status of industrial equipment, including various sensor readings and other relevant measurements. This data provides the system with a real-time snapshot of the equipment's condition, enabling the system to detect any anomalies. Then, at box 504, system 200 proceeds to the task of separating problem and remedial sentences. Using NLP techniques, system 200 segments the incoming health domain data into discrete sentences. Each sentence is then categorized as either describing a problem associated with the equipment's operating status or describing a remedial action associated with the equipment's operating status. Furthermore, in the health data workflow, at box 506, system 200 processes the separated sentences. The purpose of this step is to simplify and standardize the text, thereby ensuring a consistent format for further analysis. Techniques employed in this step may include, in particular, tokenization, stemming, and stop word removal. The result is a more concise and simplified version of the original text, which retains key information while excluding irrelevant details. After the sentence has been processed, at box 508, system 200 calculates point mutual information (PMI) weights. PMI is a statistical measure that quantifies the associations between different words in text. By calculating PMI weights, system 200 can identify and highlight words that are particularly important or informative in the context of the problem and remedial sentences. Figure 5 As depicted in the diagram, box 508 involves performing a PMI calculation using formula (i) (as shown in box 508A) which uses PMI weights and lexical units associated with remedies (as shown in box 508B) and PMI weights and lexical units associated with problems (as shown in box 508C). Furthermore, in the health data workflow, at box 510, system 200 calculates the average weighted cosine similarity using formula (ii) (as shown in box 508C). In this document, system 200 can use PMI weights and lexical units derived from previous steps to calculate the cosine similarity between different words or sentences. The similarity scores are then weighted with PMI scores to produce an average weighted cosine similarity that reflects the overall similarity between the problem sentence and the remedy sentence, taking into account both the lexical similarity of the problem sentence and the remedy sentence and the importance of their constituent words.

[0069] .

[0070] Simultaneously, the work order data workflow begins in workflow 500. At box 512, this pipeline starts with the input of work order data. This data includes historical work orders related to industrial equipment, providing system 200 with a source of information about past anomalies and how they were resolved. Then, at box 514, system 200 processes the data into an analyzable format, similar to the processing steps in the health data workflow. By simplifying and standardizing the text, the processing of the work order data prepares it for further analysis. Once the work order data has been processed, at box 516, system 200 applies a part-of-speech (POS) tagger and a lexicon filter to the data. This step involves tagging each word with its grammatical role and filtering out irrelevant words based on the lexicon. At box 518, the result is a set of sentence lexical units, which are prepared for comparison with lexical units derived from the health data workflow. Thus, similar to the work order data workflow, the health data workflow provides an average weighted cosine similarity calculation (same as box 510). This step provides a measure of the overall similarity between work order data and health domain data, taking into account PMI weights. In this invention, node embedding and subgraph embedding are used to perform similarity checks on anomalies and work orders.

[0071] After completing both the health data workflow and the work instruction data workflow, at box 520, system 200 derives the sum of the average similarity scores for each term in the sentence (both remedial action type terms (box 522) and problem type terms (box 524)). This step yields two sums: one representing the overall association between the current operational status and past remedial actions, and the other representing its association with past problems. Finally, at box 526, system 200 calculates a final term vote based on the problem score and remedial action score using a predefined formula (e.g., final term vote = remedial action score - problem score). This final term vote provides a measure of whether the current operational status of the device is more closely associated with past problems or past remedial actions, thereby guiding system 200 to derive potential remedial actions for the current anomaly.

[0072] Now, at step 160, method 100 further includes generating an output for the operator of the industrial equipment providing the derived potential remedies. The purpose of the output is to provide the operator with actionable information that can help them resolve the anomaly. The output may also include additional information such as the severity of the anomaly, the time of its occurrence, and a history of similar anomalies and their remedies. The derived potential remedies provided in the output are based on the analysis of historical work orders and sensor data, and the application of machine learning algorithms. The derived potential remedies can be used to provide a starting point for resolving the current anomaly. It represents a solution based on historical data that is likely to resolve the problem. However, ultimately, it may be the operator of the industrial equipment who is responsible for evaluating the proposed remedies, considering other relevant factors, and determining the best course of action.

[0073] In one embodiment, generating output for the operator includes presenting the operator with a list of derived potential remedies, which are ranked based on one or more of the following: the historical effectiveness of each potential remedy, the time spent implementing each potential remedy, the cost involved in implementing each potential remedy, and operator feedback regarding each potential remedy. That is, the output in list form is not merely a collection of possible solutions, but a ranked list in which remedies are ordered based on various factors reflecting their expected effectiveness and feasibility. Factors used to rank potential remedies include the historical effectiveness of each remedy (i.e., its track record in resolving similar anomalies in the past), the time spent implementing the remedy (especially in an industrial context where downtime can be costly), the cost involved in implementing the remedy (such as the cost of replacement parts or additional labor), and operator feedback associated with the remedy (i.e., the operator's experience in implementing the remedy, its ease of use, its impact on the equipment, and any other observations). Each of these factors provides a different perspective on the potential effectiveness and suitability of the remedy. By prioritizing potential remedies based on these factors, Method 100 provides operators with a comprehensive view of available options. This enables operators to make informed decisions that consider not only the potential effectiveness of each remedy but also its real-world impact in terms of time, cost, and user experience.

[0074] Those skilled in the art will appreciate that, depending on the nature of the information and the operator's preferences, the output can be presented in various formats. For example, the output can be displayed using a display as the output device 214 of system 200, such as on a dashboard, sent as a notification or alarm, or even delivered as a detailed report.

[0075] In some embodiments of this disclosure, method 100 further includes dynamically updating a knowledge graph (e.g., knowledge graph 400) based on one or more of new sensor data from industrial equipment, detected anomalies, processed work orders, and operator feedback. That is, the knowledge graph is not a static entity but is designed to be dynamically updated. This dynamic updating capability allows new information and insights to be incorporated into the knowledge graph as they become available. Sources of this new information may include the latest sensor data from the industrial equipment, newly detected anomalies, recently processed work orders, and feedback from equipment operators. By dynamically updating the knowledge graph based on these new information sources, method 100 can continuously learn and improve over time. This allows for adaptation to changing conditions, learning new experiences, and providing increasingly accurate and effective remedies for anomalies detected in industrial equipment.

[0076] Furthermore, in some embodiments, method 100 further includes: utilizing a time-series prediction model trained on historical sensor data, configured to identify degradation patterns in the sensor data over time, to detect potential future anomalies in industrial equipment. That is, the time-series prediction model is employed to anticipate potential future anomalies in industrial equipment. The model is trained on historical sensor data from the equipment, enabling it to identify degradation patterns in the sensor data over time. For this purpose, the time-series prediction model is used to analyze sensor data from the industrial equipment, which is collected over time and thus forms a time series. Once the time-series prediction model is trained, it can be used to monitor current sensor data from the equipment and predict future data points. If the model predicts a degradation pattern similar to those associated with past anomalies, the model can generate an alert for potential future anomalies. This not only detects these anomalies when they occur, but also potentially detects them even before they occur.

[0077] In some embodiments, method 100 may further provide provisions for human intervention, thereby allowing for a collaborative approach between automated systems and human expertise. Once a list of potential remedies for anomalies detected in industrial equipment has been generated and presented, the operator has the opportunity to consult experts on the problem at hand. Experts, leveraging their knowledge and experience, can then propose different solutions that they believe are the best approach to resolving the detected anomaly. Following expert consultation, the operator can input the expert-recommended solution into system 200 (e.g., using input device 212), thereby integrating human expertise into the machine learning process. Notably, this solution can be labeled as an expert-assisted solution, distinguishing it from the original solution derived by system 200. The next time the same anomaly occurs, system 200 retrieves the expert-assisted solution and treats it as a potential remedy.

[0078] The proposed Method 100 has been applied in many industries where predictive and prescriptive maintenance is crucial. For example, in the pharmaceutical industry, Method 100 can monitor historical anomalies in strictly regulated drug manufacturing processes, preventing quality and legal issues. Method 100 achieves this by creating anomaly detection models using carefully selected process parameters and extracting relevant fields from service orders. Similarly, in the coffee roasting industry, Method 100 can be used to monitor key parameters for optimal coffee production. Method 100 can identify past anomalies and model them along with work orders, tracking coffee characteristics and linking them to known issues and solutions. Furthermore, in turbine plants, Method 100 can analyze internal bore surface inspection and service reports, correlating them with anomaly data and other data related to the Bill of Materials (BOM). This creates an integrated knowledge representation source that helps resolve unresolved issues from past reports, identify new problems, and improve inventory management.

[0079] The proposed method 100 provides a unique approach to anomaly detection and resolution in industrial equipment. Method 100 fully leverages a knowledge graph (KG), which has a linguistic representation of knowledge derived from anomalies that is structured within an ontology architecture. This allows for sophisticated inference and reasoning, which is uncommon in traditional methods. A key aspect of Method 100 is the technical approach of representing data with integrated and rich representations on a Resource Description Framework (RDF) graph. This graph combines anomaly data and work order information. The work orders and their natural language text capture linguistic representations of domain-specific components and attributes, thus contributing to a richer understanding of the domain. Through this integrated representation, Method 100 allows for the construction and persistence of domain knowledge, which can be continuously improved with the addition of more data. This leads to optimal and better anomaly resolution mechanisms tailored to the specific domain of interest. Furthermore, the proposed method 100's method of capturing data (including both anomaly data and work order information) with rich representations in the graph opens up new possibilities for enriching the domain with inference and reasoning aspects. The resulting system is not only reactive but also proactive in preventing anomalies, leading to improved efficiency and reliability of industrial equipment.

[0080] While the invention has been described in detail with reference to certain embodiments, it should be understood that the invention is not limited to those embodiments. In view of the invention, many modifications and variations can be made by those skilled in the art without departing from the scope of the various embodiments of the invention as described herein. Therefore, the scope of the invention is indicated by the following claims rather than the foregoing description. All changes, modifications, and variations within the meaning and scope of the equivalents of the claims should be considered within its scope.

Claims

1. A computer-implemented method (100) for resolving anomalies in industrial equipment, the method (100) comprising: Receive information about anomalies in the industrial equipment; The anomaly is matched with at least one of the historical anomalies represented in the knowledge graph (400), wherein the knowledge graph (400) includes nodes representing historical anomalies from sensor data of the industrial equipment and historical work orders related to the industrial equipment, wherein each historical anomaly in the sensor data is associated with one or more corresponding historical work orders related to the industrial equipment. Retrieve from the knowledge graph (400) one or more corresponding historical work orders related to the industrial equipment that are associated with at least one historical anomaly that matches the anomaly in the historical anomalies; One or more retrieved historical work orders related to the industrial equipment are parsed to extract information related to the problems encountered by the industrial equipment and the corresponding remedial measures adopted. Based on the extracted information, potential remedial measures for the anomalies in the industrial equipment are derived; and The system generates outputs that provide potential remedies derived from the proposed solutions for the operators of the industrial equipment.

2. The method (100) according to claim 1, wherein, The knowledge graph (400) includes historical anomalies in the sensor data and historical work orders related to the industrial equipment, represented as nodes in a Resource Description Framework (RDF) graph constructed based on an ontology architecture pattern defined within the domain of the industrial equipment.

3. The method (100) according to claim 2, wherein, Creating a node for each historical anomaly in the sensor data includes associating anomaly-specific information with each node, the anomaly-specific information including one or more of the following: the time of occurrence of the historical anomaly, the sensor reading at the time of the historical anomaly, the operating parameters of the industrial equipment at the time of the historical anomaly, and the severity of the historical anomaly.

4. The method (100) according to claim 2, wherein, Creating a node for each historical work order related to the industrial equipment includes associating work order-specific information with each node, the work order-specific information including one or more of the following: the work order ID of the historical work order, the generation time of the historical work order, the problem description in the historical work order, the remedial action description in the historical work order, and the operator information in the historical work order.

5. The method (100) according to claim 1, wherein, Anomalies related to the industrial equipment are detected by monitoring sensor data from the industrial equipment using an autoencoder model trained under normal operating conditions of the industrial equipment, wherein deviations in the reconstruction of sensor data from the industrial equipment using the autoencoder model are detected as anomalies.

6. The method (100) according to claim 1, wherein, Matching the anomaly with at least one of the historical anomalies represented in the knowledge graph (400) includes: using a similarity score derived from a subgraph matching process, which compares subgraphs associated with anomalies related to the industrial equipment and subgraphs associated with historical anomalies in the knowledge graph (400), wherein the similarity score indicates the degree of feature similarity between the compared subgraphs.

7. The method (100) according to claim 1, wherein, Parsing one or more retrieved historical work orders related to the industrial equipment includes: applying a natural language processing (NLP) pipeline to the text of one or more retrieved historical work orders to segment the text for identifying problems encountered by the industrial equipment and corresponding remedial measures adopted.

8. The method (100) according to claim 7, wherein, The NLP pipeline is configured to perform one or more of the following stages: text preprocessing, named entity recognition, semantic role labeling, and relation extraction, to derive structured information from one or more texts retrieved from historical work orders for text segmentation.

9. The method (100) according to claim 7, further comprising: An algorithm based on Point Mutual Information (PMI) scores is applied to the segmented text as the output of the NLP pipeline to classify issues and remedies among one or more texts retrieved from historical work orders. as well as The SHAP (SHapley Additive Interpretation) algorithm is applied to interpret the contribution of each word in the classified text to the identification of the problems encountered by the industrial equipment and the corresponding remedies adopted.

10. The method (100) according to claim 1, wherein, Derivation of potential remedies for the anomaly includes applying a machine learning algorithm to the extracted information. The machine learning algorithm is trained on a dataset including the historical work orders and anomalies associated with the historical work orders, and is configured to predict potential remedies for the anomaly based on the historical data.

11. The method (100) according to claim 1, wherein, Generating output for the operator includes presenting the operator with a list of derived potential remedies, which are sorted based on one or more of the following: the historical effectiveness of each potential remedy, the time spent implementing each potential remedy, the cost involved in implementing each potential remedy, and operator feedback for each potential remedy.

12. The method (100) according to claim 1, further comprising: The knowledge graph (400) is dynamically updated based on one or more of the following: new sensor data from the industrial equipment, detected anomalies, processed work orders, and operator feedback.

13. The method (100) according to claim 1, further comprising: Using a time-series prediction model trained on historical sensor data, the time-series prediction model is configured to identify degradation patterns in the sensor data over time in order to detect potential future anomalies in the industrial equipment.

14. A system (200) comprising: One or more processing units; as well as A memory unit communicatively coupled to the one or more processing units, wherein the memory unit includes an exception resolution module stored in the form of machine-readable instructions executable by the one or more processing units, wherein the exception resolution module is configured to perform the steps of a method (100) for resolving anomalies in industrial equipment according to any one of claims 1 to 13.

15. A computer program product having machine-readable instructions stored therein, which, when executed by one or more processing units, cause the one or more processing units to perform the method (100) according to any one of claims 1 to 13.

Images