Method and device for product fault root cause analysis based on knowledge graph

Abstract

The invention discloses a method for performing product fault root cause analysis based on a knowledge graph, which comprises the following steps: obtaining design data associated with a product, the design data comprising semi-structured data and / or non-structured data; production data associated with the product is obtained, and the production data comprises structured data; performing information extraction on the design data to obtain structured design data; performing knowledge fusion on knowledge generated based on the production data and the structured design data to construct a target knowledge graph; and executing a product fault root cause analysis task based on the target knowledge graph.

Description

Technical Field

[0001] In general, the present invention relates to the field of knowledge graphs, and more specifically, to a method and device for performing root cause analysis of product failures based on knowledge graphs. Background Art

[0002] In the field of manufacturing, it is necessary to control the production process to improve product quality and reduce product failure rate while ensuring production capacity. Therefore, when a product failure occurs on the production line, it is necessary to trace the source as soon as possible to find the problem node to ensure subsequent production efficiency. In the traditional manufacturing era, the root cause analysis of product failures usually relies on expert experience and is performed manually. However, with the continuous development of manufacturing systems and processes, the possible failure nodes and generation paths involved in the product have become more and more complex, and it is difficult to find the cause of the failure in a short time by manual labor alone.

[0003] Since entering the era of intelligent manufacturing, attempts have been made to solve the problem of root cause analysis of product failures through intelligent methods, among which the use of knowledge graphs has attracted much attention. Knowledge graphs are structured, machine-understandable semantic knowledge bases that help achieve knowledge modeling, data integration, and data standardization, and provide extended functions including data access and query. Currently, knowledge graphs are usually built based on data from production systems, such as data from manufacturing execution systems (MES) and SAP systems, so the root cause analysis of product failures is only based on relevant knowledge on the production side.

[0004] However, in actual manufacturing, the cause of product failure not only comes from the production side, but may also occur on the design side. The data from the production side is usually structured data, stored in relational databases such as Oracle and Impala, but the original data from the design side is usually semi-structured or unstructured data. There are barriers between multi-source heterogeneous data, which is not easy for computers to directly process and analyze. Therefore, an improved solution for building a knowledge graph based on multi-source heterogeneous data is needed to achieve a deeper and more comprehensive root cause analysis of product failures. Summary of the invention

[0005] In manufacturing, when a product failure occurs on the production line, it is necessary to perform a root cause analysis of the product failure. Considering that the field of intelligent manufacturing is usually knowledge-intensive, it is advantageous to introduce knowledge graph technology in the field of intelligent manufacturing to perform root cause analysis of product failures. Knowledge graph is a structured semantic network that provides a powerful ability to organize, manage and understand massive amounts of information. Specifically, a network is constructed using entities extracted from existing data / raw data, relationships between entities, and entity attributes to construct a knowledge graph, and the semantics contained in the knowledge graph are used for reasoning to obtain possible failure paths, and further quantify the probability of the root cause of the failure.

[0006] It is desirable to provide a method and device for product failure root cause analysis based on knowledge graph, which can obtain entities and relationships between entities, entity attributes based on multi-source heterogeneous data from various links from design to production, and use the obtained entities and relationships between entities, entity attributes to build a knowledge graph, thereby avoiding the problem that when product failure root cause analysis is performed based on knowledge graph, only the production side is involved and the cause of the failure cannot be fully identified. Therefore, instead of only based on data from the production system, the knowledge graph can be further constructed based on data from the design side, expert experience, etc., and the reasoning ability of the knowledge graph can be advantageously used to infer a complete and deep possible failure path when a product failure occurs.

[0007] According to one aspect of the present invention, there is provided a method for performing root cause analysis of product failures based on a knowledge graph, comprising: obtaining design data associated with a product, wherein the design data includes semi-structured data and / or unstructured data; obtaining production data associated with the product, wherein the production data includes structured data; performing information extraction on the design data to obtain structured design data; performing knowledge fusion on knowledge generated respectively based on the production data and the structured design data to construct a target knowledge graph; and performing a product failure root cause analysis task based on the target knowledge graph.

[0008] According to another aspect of the present invention, a system for performing root cause analysis of product failures based on a knowledge graph is provided, comprising: a memory; a processor coupled to the memory, the processor being configured to enable one or more units to execute a method according to any one of the various embodiments of the present invention.

[0009] According to yet another aspect of the present invention, a computer readable medium is provided, which stores a computer program including instructions, which, when executed by a processor, cause one or more units to perform a method according to any one of the various embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Various embodiments of the claimed subject matter will now be described by way of example with reference to the accompanying drawings. In the different drawings, the same reference numerals are used to designate the same or similar components.

[0011] Figure 1 A module and a data flow diagram for performing root cause analysis of product failures based on a knowledge graph according to an embodiment of the present invention are shown.

[0012] Figure 2 An example product assembly design diagram is shown according to an embodiment of the present invention.

[0013] Figure 3A causal chain diagram for performing root cause analysis of product failures using the constructed knowledge graph according to an embodiment of the present invention is shown.

[0014] Figure 4 A flowchart of performing product failure root cause analysis based on a knowledge graph according to an embodiment of the present invention is shown.

[0015] Figure 5 A block diagram of a device that can perform root cause analysis of product failures based on a knowledge graph according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0016] In the following description, many specific details are set forth to provide a thorough understanding of the embodiments of the present invention. However, those skilled in the relevant art will recognize that the present invention can be practiced without one or more of the specific details, or alternative methods, components, etc. can be used to practice the present invention. In some instances, well-known structures and operations are not shown or described in detail to avoid unnecessarily obscuring the present invention.

[0017] In the field of manufacturing, when a product failure occurs on the production line, it is crucial to trace the source as quickly as possible to find the problem node and ensure subsequent production efficiency. With the development of various intelligent manufacturing technologies, we have gradually moved away from relying solely on technicians to analyze and identify the root causes of product failures, and instead use computers to extract information from massive data to assist in decision-making. Knowledge graphs can organize data from different sources into a network in the form of entities, relationships between entities, and entity attributes. It is a human-recognizable and machine-friendly knowledge representation, and has been widely used in product failure root cause analysis tasks due to its knowledge integration and reasoning capabilities.

[0018] In some cases, the possible source of product failure problems is on the production side, for example, product failure caused by failure of the equipment used to produce the product. Therefore, at present, the conventional practice is to create knowledge graphs based on data from the production side, such as data from the Manufacturing Execution System (MES) and SAP systems. The data from these systems are usually structured, and they are stored in various relational databases such as Oracle and Impala, which are easy to extract information by computers and map to ontology to create knowledge graphs.

[0019] However, in actual production, product failure problems potentially also come from the design side, for example, product failures caused by incorrect component dimensions or material labeling errors in the design. Moreover, product failures caused by the design side actually have a more profound impact on the entire production process, so it is necessary to create a knowledge graph based on design-related data. However, data related to product design is usually semi-structured or unstructured, and it is impossible for a computer to directly obtain effective information like structured data. In some cases, design-related data can be stored in the form of semi-structured data, for example, in the Extensible Markup Language (XML) format, JSON format, or Hypertext Markup Language (HTML) format. In other cases, design-related data can be stored in the form of unstructured data, such as CAD format, JPG format, any applicable image format, or PDF format.

[0020] To this end, the present invention proposes a method for product failure root cause analysis based on knowledge graph, which does not rely solely on data from the production side, but can further advantageously use multi-source heterogeneous data from the design side, expert experience, etc. to create a knowledge graph, so as to infer a more complete and in-depth possible failure path when a product failure occurs, and provide more comprehensive decision support for product failure root cause analysis. The proposed method for product failure root cause analysis based on knowledge graph will be specifically described below. Those skilled in the art should understand that the present invention can be applied to any production line or production process without being limited to a specific field.

[0021] Figure 1 The modules and data flow diagram for product failure root cause analysis based on knowledge graph according to an embodiment of the present invention are shown, wherein the dashed box represents an optional module. Each module provided in this embodiment can be implemented by software and / or hardware, and integrated into a computing device capable of executing this embodiment, or can be distributed in multiple computing devices capable of executing this embodiment.

[0022] See also Figure 1 , obtaining design data D101 associated with the product, the design data D101 is input and stored in a design data storage unit M101 in an applicable format, wherein the design data D101 may be stored as semi-structured data and / or unstructured data.

[0023] In one embodiment, the design data D101 associated with the product may include product requirement documents, product assembly drawings, component design documents, component design drawings, circuit design files, circuit graphics, design logs, etc. Among them, the circuit design files and design logs can be stored in the extended markup language (XML) format or the hypertext markup language (HTTP) format, and they have a semi-structured data structure. Among them, the product requirement documents and component design documents can be stored in any applicable office document or text format, and the product assembly drawings, component design drawings, and circuit graphics can be stored in CAD format, JPG format or other picture formats or PDF format, and they have an unstructured data structure. The above-mentioned design data sources are only used as examples and do not limit the scope of the present invention in any way.

[0024] Figure 2 An example product assembly design diagram according to an embodiment of the present invention is shown. Figure 2 , the example product component design drawing has the design drawing number ID marked in the upper left corner, the shape and size of the component are shown in the middle, the box 201 in the lower left corner usually contains relevant text annotations for the design, and the middle boxes 202 and 203 usually mark the component sizes. For example, box 202 can mark the axis size of the elliptical component with numbers, and box 203 can mark the sphere diameter of the spherical component with numbers. As shown, the design drawing usually includes both graphic information and digital information, as well as text information, and the positions of the graphics, numbers, and text vary from design drawing to design drawing. Constructing a knowledge graph requires obtaining valid information that can be read by a computer from similar massive component design drawings, so additional processing is required.

[0025] Continue to see Figure 1 , the semi-structured and / or unstructured design data D102 is input into the design data pre-processing unit M102 for pre-processing to be converted into structured data D103.

[0026] In one embodiment, when the design data D102 is semi-structured data, the design data pre-processing unit M102 performs an information extraction operation on the design data D102.

[0027] In one embodiment, the information extraction operation includes one or more of the following: entity extraction, relationship extraction, and attribute extraction. Entities are identified from the design data D102 through the entity extraction operation. For example, the entity extraction operation can identify the entities "component 1", "component 2", and "component 3" based on the description of "the shape of component 1 is an ellipse, and component 1 includes component 2 and component 3". Relationships between entities are identified from the design data D102 through the relationship extraction operation. For example, the relationship extraction operation can identify that "component 1" and "component 2" and "component 3" have an "include" relationship based on the description of "the shape of component 1 is an ellipse, and component 1 includes component 2 and component 3". Attribute information of entities is identified from the design data D102 through the attribute extraction operation. For example, the entity extraction operation can identify that "component 1" has a "shape" attribute based on the description of "the shape of component 1 is an ellipse, and component 1 includes component 2 and component 3", and its value is "ellipse". In some embodiments, the attribute extraction operation can also be implemented as an instance of the relationship extraction operation.

[0028] In one embodiment, entity extraction can be implemented using one or any combination of the following: rule-based methods, such as regular matching, etc.; statistical model-based methods, such as hidden Markov models, conditional Markov models, maximum entropy models, and conditional random field models, etc.; and deep learning-based methods, such as methods based on convolutional neural networks, recurrent neural networks, and neural networks that introduce attention mechanisms, such as LSTM-CRF, LSTM-CNN-CRF, Transformer, etc.

[0029] In one embodiment, relationship extraction can be implemented using one or any combination of the following: a template-based relationship extraction method; a supervised learning-based relationship extraction method, such as converting relationship extraction into a classification problem and training models such as support vector machines and naive Bayes; a weakly supervised learning-based relationship extraction method, such as the Bootstrapping method.

[0030] In one embodiment, entity extraction and relation extraction are performed separately in sequence. In another embodiment, entity extraction and relation extraction are performed jointly, for example, using methods such as BERT, LSTM-RNN, DGCNN, etc.

[0031] In one embodiment, attribute extraction can be implemented using one or any combination of the following: unsupervised methods, such as rule-based slot filling methods, clustering-based methods, etc.; dependency-based semi-supervised methods, such as using the PageRank algorithm, etc.; deep learning-based sequence labeling methods, such as CRF models, BI-GRU+CRF models, etc.; meta-pattern-based attribute extraction methods, etc.

[0032] In one embodiment, the information extraction operation may include performing one or more of the following tasks: word segmentation, vocabulary query, syntactic analysis, corpus processing, named entity recognition, term extraction, part-of-speech tagging, word meaning role tagging and entity relationship recognition, sequence tagging.

[0033] In another embodiment, when the design data D102 is unstructured data, the design data preprocessing unit M102 applies an optical character recognition operation to the design data D102 to convert the text in the design data D102 into a computer-readable text format before performing the information extraction operation on the design data D102. After the unstructured design data D102 is converted into a computer-readable text format, the information extraction operation can be continued with reference to the above description.

[0034] In one embodiment, the structured data D103 may be stored in a relational database. In a further embodiment, the structured data D103 is organized in the form of design metadata with a corresponding design drawing number, wherein the design drawing number may correspond to an index of a data table, and the design metadata may correspond to the headers of multiple data tables, wherein each data table is regarded as a class, each row in the table represents an instance or object of the class, and each column in the table represents the attributes possessed by the class. And wherein the design metadata may be extracted from the design data D102 via an information extraction operation by the design data preprocessing unit M102, and the design metadata may correspond to the extracted entities, relationships and / or attributes. For example, the design metadata may include one or any multiple of the following: product name, product number, component quantity, component name, component number, component shape, component material, component size, assembly sequence, etc.

[0035] Continue to see Figure 1 , the structured data D103 is input to the design knowledge generating unit M103 to obtain the design knowledge D104.

[0036] In one embodiment, the design knowledge D104 may be obtained by converting the structured data D103 into a Resource Description Framework (RDF) format or an RDFS format.

[0037] Specifically, the structured data D103 may be converted into a Resource Description Framework (RDF) or RDFS triple format. For example, RDF or RDFS triples may organize mapped data in the format of {subject-predicate-object}. For another example, RDF or RDFS triples may organize mapped data in the format of {subject-attribute-attribute value}. Furthermore, since the assembly order of product components has an important influence on the root cause analysis of faults, time sequence information may be added to the RDF or RDFS triples to obtain RDF or RDFS quads.

[0038] In a further embodiment, the design knowledge D104 may be obtained by mapping the structured data D103 to an ontology in a created ontology library. In this case, the design knowledge D104 may be a knowledge graph constructed based on the design data D101.

[0039] The process of building a knowledge graph may include first developing an ontology based on the design data associated with the product. An ontology can generally be understood as a conceptual description of a specific field (e.g., automotive parts production), and can characterize knowledge data that can be understood by technicians in this field in terms of classes and relationships. Design-related ontologies can be developed in a bottom-up manner based on the design data associated with the product. Various ontology development methods can be used to develop the above-mentioned ontologies. For example, methods such as Common-KADS, TOVE, IDEF5, METHONTOLOGY, etc. can be used to develop ontologies. The development of an ontology can be achieved using Web Ontology Language (OWL). In addition, in the process of developing an ontology, expert experience can also be incorporated to further modify or improve the ontology in a top-down manner based on expert knowledge.

[0040] As mentioned above, the structured data D103 can be converted into RDF triple or RDF quad format, and then the RDF triple or RDF quad data can be merged with the created ontology OWL to obtain the design knowledge D104. Specifically, R2RML (RDB to RDF Mapping Language) mapping can be used to map the data fields in the RDF triple or RDF quad to the created ontology. In addition, other mapping methods can be used, such as direct mapping, D2RQ, etc.

[0041] Continue to see Figure 1, obtaining production data D105 associated with the product, the production data D105 is input and stored in a production data storage unit M105 in an applicable format, wherein the production data D105 can be stored as structured data.

[0042] In one embodiment, the production data D105 associated with the product may include any production-related records, for example, data from a factory, data from a production line, or data from a production device, including data from a manufacturing execution system (MES), data from a SAP system, production statistics (such as factory data, plant data, production line data, equipment data, etc.), etc. The production data D105 is usually structured data, having a format that is easy to be processed by a computer, for example, data from an MES system usually has a unified format: machine = {machine type, machine status ...}. The production data D105 can be stored in a relational database, such as Oracle, DB2, MySQL, Impala, etc. The above-mentioned production data sources are used only as examples and do not limit the scope of the present invention in any way.

[0043] Continue to see Figure 1 , the structured data D106 is input to the production knowledge generating unit M106 to obtain production knowledge D107.

[0044] In one embodiment, similar to the design knowledge D104, the production knowledge D107 may be obtained by converting the structured data D106 into a Resource Description Framework (RDF) format or an RDFS format.

[0045] In a further embodiment, similar to the design knowledge D104, the production knowledge D107 may be obtained by mapping the structured data D106 to an ontology in the created ontology library. In this case, the production knowledge D107 may be a knowledge graph created based on the production data D105.

[0046] The process of building a knowledge graph may include first developing an ontology based on production data associated with a product. An ontology can generally be understood as a conceptual description of a specific field (e.g., automotive parts production), and can characterize knowledge data that can be understood by a person skilled in the art in terms of classes and relationships. Production-related ontology can be developed in a top-down manner based on production data associated with a product. Various ontology development methods can be used to develop the above-mentioned ontology. For example, methods such as Common-KADS, TOVE, IDEF5, METHONTOLOGY, etc. can be used to develop the ontology. The development of the ontology can be achieved using Web Ontology Language (OWL). In addition, in the process of developing the ontology, expert experience can also be incorporated to further modify or improve the ontology in a top-down manner based on expert knowledge.

[0047] The process of obtaining the production knowledge D107 based on the structured data D106 by using the production knowledge generation module M106 can refer to the description of the process of obtaining the design knowledge D104 based on the structured data D103, which will not be repeated here.

[0048] In the actual product failure root cause analysis process, design data and production data reveal more possible failure link paths between products, components and production equipment. In comparison, expert experience can provide real failure link paths about past product failure root cause analysis. Therefore, introducing expert experience is also beneficial for product failure root cause analysis.

[0049] Continue to see Figure 1 Optionally, expert data D108 associated with the expert experience is obtained, and the expert data D108 is input and stored in an expert data storage unit M107 in a suitable format.

[0050] In some cases, the expert data D108 may include: fault detection records, production equipment manuals, maintenance history documents, etc., wherein the expert data D108 may be stored as structured data, semi-structured data and / or unstructured data. The above expert data sources are only used as examples and do not limit the scope of the present invention in any way.

[0051] In some cases, the expert knowledge D111 may be RDF triple or RDF quad data. In some cases, the expert knowledge D111 may be a knowledge graph created based on the expert data D108. The process of obtaining the expert knowledge D111 based on the structured data D109 using the expert knowledge generation unit M109 may refer to the description of the process of obtaining the design knowledge D104 based on the structured data D103, which will not be repeated here. The process of obtaining the expert knowledge D111 based on the semi-structured and / or unstructured data D109 using the expert data preprocessing unit M108 and the expert knowledge generation unit M109 may refer to the description of the process of obtaining the design knowledge D104 based on the semi-structured and / or unstructured data D102, which will not be repeated here.

[0052] Continue to see Figure 1 , design knowledge D104 and production knowledge D107, as well as optional expert knowledge D111 are input into the knowledge fusion unit M104 for knowledge fusion to generate fused knowledge D112.

[0053] The goal of knowledge fusion is to fuse knowledge at all levels (such as concept level, data level). In one embodiment, knowledge fusion may include one or more of the following processes: data preprocessing, segmentation, load balancing, record linking, result evaluation and / or result output.

[0054] In the case where the design knowledge D104 and the production knowledge D107, and optionally the expert knowledge D111, are respectively RDF triples or RDF quads, knowledge fusion aims to find equivalent instances between data from different sources. In one embodiment, knowledge fusion includes one or more of the following tasks: data cleaning, entity alignment, entity matching, entity disambiguation, relationship deduplication, etc. In this case, the fused knowledge D112 is RDF triples or RDF quads.

[0055] In the case where the design knowledge D104 and the production knowledge D107, and optionally the expert knowledge D111 are knowledge graphs created based on the design data D101, the production data D105 and the expert data D108, respectively, the knowledge fusion aims to discover equivalent instances, equivalent or inclusive concepts or attributes between multiple knowledge graphs. In one embodiment, the knowledge fusion includes one or more of the following tasks: data cleaning, ontology alignment, ontology matching, entity alignment, entity matching, entity disambiguation, relationship deduplication, etc. In this case, the fused knowledge D112 is a complete knowledge graph that integrates the design knowledge D104, the production knowledge D107 and the expert knowledge D111 (optional).

[0056] In other embodiments, in the process of knowledge fusion, expert experience can be incorporated in other ways to further modify or improve the knowledge fusion process. For example, equivalent ontologies / entities, conflicting ontologies / entities, repeated relationships between ontologies / entities, and / or attributes that can be merged can be identified by expert annotation.

[0057] In another embodiment, the knowledge data input into the knowledge fusion unit may include knowledge data from other sources in addition to the example data described above, including but not limited to: procurement knowledge from procurement data, material knowledge from material data, acceptance knowledge from acceptance data, etc. The data sources are only used as examples and do not limit the scope of the present invention in any way. Depending on whether the data from various sources are structured, semi-structured or unstructured data, similar processing may be performed with reference to the above description of design data, production data and / or expert data to obtain corresponding knowledge data.

[0058] Generally, due to the sensitivity of various data associated with products, encryption is required when obtaining raw data to ensure system robustness, especially raw data obtained in electronic form from the cloud such as Azure Cloud. For example, when obtaining production data from an MES system or SAP system from the cloud, the raw data is usually downloaded in the form of key-value pairs, and at least the sensitive data columns therein are encrypted.

[0059] Continue to see Figure 1 Optionally, the fused knowledge data D112 is input into the knowledge graph generation unit M110 to construct the knowledge graph D113.

[0060] In one embodiment, the fused knowledge data D112 may include design knowledge D104, production knowledge D107, and optional expert knowledge D111. In a further embodiment, the fused knowledge data D112 may include procurement knowledge, material knowledge, acceptance knowledge, etc. In the case where the fused knowledge data D112 is RDF triple or RDF quad data, it is mapped to the ontology in the created ontology library to construct a complete knowledge graph D113. Referring to the above description, the ontology library can be constructed in a top-down and / or bottom-up manner based on various original data associated with the product. In one embodiment, the created ontology library includes one or any multiple of the following ontologies: design ontology, product ontology, equipment ontology, material ontology, physical asset ontology, process segment ontology, expert ontology, procurement ontology, acceptance ontology, etc.

[0061] Through the method described above, the complete knowledge graph can integrate multi-source heterogeneous data from the design side, production side, expert experience, etc., so as to infer a more complete and in-depth possible failure path when a product failure occurs, and provide more comprehensive decision support for product failure root cause analysis.

[0062] In one embodiment, the constructed knowledge graph can be stored in a graph database, and the entities, relationships between entities, and / or entity attributes in the knowledge graph can be queried through a query language such as the SPARQL language, thereby providing support for root cause analysis of product failures. In a further embodiment, the constructed knowledge graph can be visualized, for example, a graph can be generated with entities in the knowledge graph D113 as nodes and relationships between entities as connecting edges, wherein attributes can also be designed as entities, and the attribute relationships between entities and entity attributes can be designed as relationships between entities, and thus can be processed in a similar manner.

[0063] Furthermore, the constructed knowledge graph can be used to provide richer applications. Figure 1 , the constructed knowledge graph D113 is input into the knowledge graph application unit M111 to perform the product failure root cause analysis task.

[0064] In one embodiment, the knowledge graph application unit M111 can provide a product failure causal chain generation service. For example, the product failure causal chain generation service can visualize all possible root cause paths of the failure based on the faulty product, component, or failure phenomenon queried by the user. In a further embodiment, the knowledge graph application unit M111 can visualize all possible root cause paths of the failure based on the knowledge graph stored in the graph database, and calculate the weight for each connection edge, wherein the connection edge with a larger weight indicates a higher probability of failure / number of occurrences of failure. In one embodiment, the weight of each connection edge can be calculated based on one of degree centrality, eigenvector centrality, betweenness centrality, or connection centrality.

[0065] Figure 3 FIG. 1 shows a causal chain diagram for performing root cause analysis of product failures using the constructed knowledge graph according to an embodiment of the present invention. Figure 3 As shown in the figure, each point represents an entity, and each connecting edge represents the relationship between entities. For example, node A represents product A, node MA represents production plant A, node PA represents production plant A, node DA represents design file A, and the reference numerals with suffix B are represented in the same way as the reference numerals with suffix A; Figure 3In the example, the product / component attributes contained in the design file are also designed as entities, where Ma represents material, Si represents size, and Dr represents drawing. The existence of connecting edges between nodes indicates the existence of possible failure paths, and the connecting edges with larger weights indicate higher failure probability / failure occurrence times.

[0066] In one embodiment, a depth-first search (DFS) may be performed in descending order of the weight of the connection edge until the real root cause of the fault is found. For example, when performing a root cause analysis on product B, first follow the connection edge with a weight of 0.6 to reach the manufacturer MB; if the manufacturer MB is not the root cause of the fault, follow the connection edge with a weight of 0.5 to reach the design file DB; in order to determine whether the design file DB is the root cause of the fault, it is necessary to check the dimensions, materials, and drawings in the design file DB in descending order of weight; if the design file DB is not the root cause of the fault, continue to check the design file DA with a weight of 0.3. Follow a similar process until the exact root cause of the fault is found.

[0067] In one embodiment, the knowledge graph application unit M111 can provide a product failure root cause search service. For example, multiple possible failure root cause nodes can be provided based on the failed product, component, or failure phenomenon queried by the user. In one embodiment, the user can use the SPARQL language to query. In one embodiment, the returned failure root cause nodes can be sorted and filtered in the order from large to small according to the connection edge weights described above.

[0068] In one embodiment, the knowledge graph application unit M111 can provide a product failure root cause automatic generation service. For example, when a failure occurs on a production line, the knowledge graph application unit M111 can provide the user with a speculated root cause node about the product failure based on the identified failure. In one embodiment, multiple possible root cause nodes can be provided to the user, and the provided root cause nodes can be sorted and screened in the order from large to small according to the connection edge weights described above.

[0069] The above has described how to construct a knowledge graph using multi-source heterogeneous data associated with a product, and provide a variety of applications for product failure root cause analysis based on the constructed knowledge graph. On this basis, in order to further improve the failure root cause analysis capability of the constructed knowledge graph, the present invention also provides a method for training the created knowledge graph using a machine learning model.

[0070] Continue to see Figure 1 Optionally, the constructed knowledge graph D113 is input into the knowledge graph training unit M112 for training using a machine learning model.

[0071] In one embodiment, the constructed knowledge graph D113 is input into a machine learning model, and supervised training is performed on it based on the most likely fault path or fault node output. For example, the constructed knowledge graph D113 is input into a machine learning model in the form of a graph, wherein the nodes and edges of the knowledge graph can be converted into multi-dimensional vectors respectively; then, the knowledge graph D113 can output the most likely fault path or fault node, which can also be represented as a multi-dimensional vector, for example, the most likely fault path or fault node can be screened in the order from large to small according to the connection edge weights as described above. In one embodiment, the machine learning model can be a graph neural network model, such as one or any multiple of a graph convolutional network, a graph autoencoder, a graph generation network, a graph recurrent network, or a graph attention network.

[0072] During the training process, the machine learning model can use the actual fault path or fault node as a label, and the difference between the output most likely fault path or fault node and the label as the loss function, and train the knowledge graph D113 by minimizing the loss function. For example, the difference between the output most likely fault path or fault node and the label can be measured by cosine similarity, log-likelihood similarity, Euclidean distance, Manhattan distance, etc. Subsequently, the trained knowledge graph D114 is input into the knowledge graph application unit M111 to perform the product failure root cause analysis task.

[0073] Figure 4 A flowchart of performing product failure root cause analysis based on a knowledge graph according to an embodiment of the present invention is shown, wherein the dashed box represents an optional operation. Each step provided in this embodiment can be implemented by one or more devices with computing capabilities through software and / or hardware.

[0074] In step S401, design data associated with a product may be obtained, wherein the design data includes semi-structured data and / or unstructured data. The operation of step S401 may be performed by, for example, referring to Figure 1 The described design is implemented by the data storage unit M101.

[0075] In one embodiment, the design data comes from one or more of the following: product requirement documents, product assembly drawings, component design documents, component design drawings, circuit design files, circuit graphics, and design logs. For example, circuit design files and design logs can be stored in an extended markup language (XML) format or a hypertext markup language (HTTP) format, and they have a semi-structured data structure. For example, product requirement documents and component design documents can be stored in any applicable office document or text format, and product assembly drawings, component design drawings, and circuit graphics can be stored in a CAD format, a JPG format, or other image format or a PDF format, and they have an unstructured data structure.

[0076] In step S402, production data associated with the product may be obtained, wherein the production data includes structured data. The operation of step S402 may be performed by, for example, referring to Figure 1 The described production data storage unit M105 is implemented.

[0077] In one embodiment, the production data comes from one or more of the following: a manufacturing execution system (MES), a SAP system, and production statistics. The production statistics may include production records from a factory, production records from a plant, production records from a production line, and / or production records from equipment.

[0078] In step S403, an optical character recognition operation may be optionally applied to the design data to convert the text in the design data into a computer-readable text format. The operation of step S403 may be performed by, for example, referring to Figure 1 The design data preprocessing unit M102 described above is implemented.

[0079] In step S404, information extraction may be performed on the design data to obtain structured design data. The operation of step S403 may be performed by, for example, referring to Figure 1 The design data preprocessing unit M102 described above is implemented.

[0080] In one embodiment, performing information extraction on the design data includes performing entity extraction and relationship extraction on the design data to identify entities and relationships between entities from the design data. For example, entity extraction can be implemented using one or more of the following: rule-based methods, such as regular matching, etc.; statistical model-based methods, such as hidden Markov models, conditional Markov models, maximum entropy models, and conditional random field models, etc.; and deep learning-based methods, such as methods based on convolutional neural networks, recurrent neural networks, and neural networks that introduce attention mechanisms, such as LSTM-CRF, LSTM-CNN-CRF, Transformer, etc. For example, relationship extraction can be implemented using one or more of the following: template-based relationship extraction methods; supervised learning-based relationship extraction methods, such as converting relationship extraction into a classification problem, training models such as support vector machines and naive Bayes; weakly supervised learning-based relationship extraction methods, such as Bootstrapping methods.

[0081] In one embodiment, entity extraction and relation extraction are performed separately in sequence. In another embodiment, entity extraction and relation extraction are performed jointly, for example, using methods such as BERT, LSTM-RNN, DGCNN, etc.

[0082] In a further embodiment, performing the information extraction on the design data further includes performing attribute extraction on the design data to identify the attributes of entities from the design data. For example, attribute extraction can be implemented using one or more of the following: unsupervised methods, such as rule-based slot filling methods, clustering-based methods, etc.; semi-supervised methods based on dependency relationships, such as using PageRank algorithms, etc.; sequence labeling methods based on deep learning, such as CRF models, BI-GRU+CRF models, etc.; attribute extraction methods based on meta-patterns, etc. For another example, the attribute extraction operation can be implemented as a certain instance of the relationship extraction operation, and thus can be implemented using a similar method of relationship extraction.

[0083] In one embodiment, performing the information extraction on the design data includes one or more of the following tasks: word segmentation, vocabulary query, syntactic analysis, corpus processing, named entity recognition, term extraction, part-of-speech tagging, word meaning role tagging and entity relationship recognition, sequence tagging.

[0084] In one embodiment, the structured design data is organized in the form of design metadata with corresponding design drawing numbers. The design drawing number may correspond to the index of a data table, and the design metadata may correspond to the headers of multiple data tables, wherein each data table is regarded as a class, each row in the table represents an instance or object of the class, and each column in the table represents the attributes of the class. And wherein the design metadata may be extracted from the design data via an information extraction operation, and the design metadata may correspond to the extracted entities, relationships and / or attributes. For example, the design metadata may include one or any multiple of the following: product name, product number, component quantity, component name, component number, component shape, component material, component size, assembly sequence, etc.

[0085] Optionally, in step S405, one or more of expert experience, procurement data, material data, and acceptance data may also be obtained. The operation of step S405 may be performed by, for example, referring to Figure 1 The expert experience storage unit M107 described above is used to implement the above-described data. The data from various sources are structured, semi-structured or unstructured data, and can be processed with reference to the above-described related descriptions.

[0086] In step S406, the knowledge generated based on the production data and the structured design data can be fused to construct a target knowledge graph. Optionally, the data generated based on one or more of expert experience, procurement data, material data, and acceptance data can be further fused to construct a target knowledge graph. The operation of step S406 can be, for example, referred to Figure 1 The design knowledge generating unit M103, production knowledge generating unit M106, and expert knowledge generating unit M109 described above can be implemented. In addition, the operation of step S406 can also be performed by, for example, referring to Figure 1 Optionally, the operation of step S406 can also be performed by, for example, referring to Figure 1 The described knowledge graph generation unit M110 is implemented.

[0087] In one embodiment, the production data and the structured design data can be converted into one of the following formats: Resource Description Framework (RDF) triple format, RDF quad format with time information, RDFS triple format, RDFS quad format with time information, so as to obtain the knowledge based on the production data and the structured design data.

[0088] In a further embodiment, an ontology library can be established based at least on the design data and the production data (optionally, one or more of expert experience, procurement data, material data, and acceptance data), and the ontology library includes multiple ontologies; and then the production data and the structured design data in RDF or RDFS triple format or RDF or RDFS quad format are mapped to the multiple ontologies to obtain the knowledge based on the production data and the structured design data. For example, the ontology can be developed in a bottom-up and / or top-down manner based on the original data associated with the product. For example, the ontology library includes one or more of the following: design ontology, product ontology, equipment ontology, material ontology, physical asset ontology, process segment ontology, expert ontology, procurement ontology, and acceptance ontology.

[0089] In one embodiment, when the knowledge based on the production data and the structured design data is RDF triple or RDF quad data, knowledge fusion includes one or more of the following tasks: data cleaning, entity alignment, entity matching, entity disambiguation, relationship deduplication, etc. In this embodiment, the fused knowledge is mapped to the multiple ontologies to construct the target knowledge graph.

[0090] In another embodiment, when the knowledge based on the production data and the structured design data is mapped to the multiple ontologies, knowledge fusion includes one or more of the following tasks: data cleaning, ontology alignment, ontology matching, entity alignment, entity matching, entity disambiguation, relationship deduplication, etc. In this embodiment, the fused knowledge is the target knowledge graph.

[0091] In an optional embodiment, similar processing can be further performed based on one or more of expert experience, procurement data, material data, and acceptance data with reference to the above description to construct the target knowledge graph.

[0092] In a further embodiment, during knowledge fusion, equivalent ontologies / entities, conflicting ontologies / entities, repeated relationships between ontologies / entities, and / or attributes that can be merged can be identified through expert annotation.

[0093] In step S407, the product failure root cause analysis task can be performed based on the target knowledge graph. The operation of step S405 can be performed by, for example, referring to Figure 1 The described knowledge graph is implemented by the application unit M111.

[0094] In one embodiment, performing a product failure root cause analysis task based on the target knowledge graph includes: taking at least entities in the target knowledge graph as nodes and the relationships between the entities as connecting edges, calculating the weights of each connecting edge; and further performing the product failure root cause analysis task using a depth-first search based on the weights of each connecting edge. For example, the weight of each connecting edge can be calculated based on one of degree centrality, eigenvector centrality, betweenness centrality, or connection centrality. For example, the product failure root cause analysis task includes a product failure causal chain generation service, a product failure root cause search service, and / or a product failure root cause automatic generation service.

[0095] Optionally, in step S408, the target knowledge graph may be trained using a machine learning model, and then the product failure root cause analysis task of step S407 may be performed based on the trained target knowledge graph. The operation of step S408 is, for example, Figure 1 The described knowledge graph training unit M112 is used to implement it.

[0096] In one embodiment, training the target knowledge graph using a machine learning model may include: generating a graph using at least entities in the target knowledge graph as nodes and relationships between the entities as connecting edges; representing the nodes and the connecting edges as vectors, and training the target knowledge graph using a graph neural network model. For example, the graph neural network model may be one or more of a graph convolutional network, a graph autoencoder, a graph generation network, a graph recurrent network, or a graph attention network.

[0097] In one embodiment, training the target knowledge graph using a machine learning model may include: using the difference between the most likely fault path or fault node output by the target knowledge graph and the actual fault path or fault node as a loss function, and training the target knowledge graph by minimizing the loss function. For example, the difference between the most likely fault path or fault node output and the label may be measured by cosine similarity, log-likelihood similarity, Euclidean distance, Manhattan distance, etc.

[0098] Figure 5 A block diagram of a device that can perform root cause analysis of product failures based on a knowledge graph according to an embodiment of the present invention is shown.

[0099] The example computing device 500 includes an internal communication bus 502 and a processor (e.g., a central processing unit (CPU)) 504 connected to the internal communication bus 502, the processor 504 being used to execute instructions stored in a memory 506 to implement the method for performing root cause analysis of product failures based on a knowledge graph described in detail above. The memory 506 is suitable for tangibly embodying computer program instructions and data, and may include various forms of memory, including, for example, semiconductor memory devices such as EPROM, EEPROM, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM disks, etc. The computing device 500 may also include an input / output (I / O) interface 508, so that various I / O devices (e.g., a cursor control device such as a mouse, a keyboard, etc.) can be coupled to the computing device 500 through the I / O interface 508 to allow a user to apply various commands and input data. The computing device 500 may also include a display unit 510 for displaying a graphical user interface.

[0100] The computer program may include instructions that can be executed by a computer, and the instructions are used to cause the processor 504 of the computing device 500 to perform the method of performing product failure root cause analysis based on the knowledge graph of the present invention. The program can be recorded on any data storage medium including a memory. For example, the program can be implemented in digital electronic circuits, or in computer hardware, firmware, software, or a combination thereof. The process / method steps described in the present invention can be performed by a programmable processor that executes program instructions to perform methods, steps, operations by operating on input data and generating outputs.

[0101] It should be noted that "based on" in the description herein means "at least based on", rather than "only based on". In addition to what is described herein, various modifications may be made to the disclosed embodiments and implementations without departing from the scope of the disclosed embodiments and implementations. Therefore, the descriptions and examples herein should be interpreted as illustrative rather than limiting. The scope of the present invention should be measured only by reference to the claims.

Claims

1. A method for product failure root cause analysis based on knowledge graph, comprising: Acquire design data associated with a product, wherein the design data includes semi-structured data and / or unstructured data; Acquire production data associated with a product, wherein the production data includes structured data; performing information extraction on the design data to obtain structured design data; Performing knowledge fusion on the knowledge generated based on the production data and the structured design data to construct a target knowledge graph; and A product failure root cause analysis task is performed based on the target knowledge graph.

2. The method according to claim 1, further comprising: Generate expert knowledge based on expert experience; The knowledge generated based on the production data and the structured design data are respectively fused with the expert knowledge to construct the target knowledge graph.

3. The method according to claim 1, further comprising: Prior to performing the information extraction on the design data, an optical character recognition operation is applied to the design data to convert text in the design data into a computer-readable text format.

4. The method according to claim 1, performing the information extraction on the design data comprises: Entity extraction and relationship extraction are performed on the design data to identify entities and relationships between entities from the design data.

5. The method according to claim 4, performing the information extraction on the design data further comprises: Attribute extraction is performed on the design data to identify attributes of entities from the design data.

6. According to the method of claim 1, performing the information extraction on the design data includes one or more of the following tasks: word segmentation, vocabulary query, syntactic analysis, corpus processing, named entity recognition, term extraction, part-of-speech tagging, word meaning role tagging and entity relationship recognition, sequence tagging.

7. The method according to claim 1, wherein: The structured design data is organized in the form of design metadata having corresponding design drawing numbers.

8. The method according to claim 7, wherein: The design metadata corresponds to one or more of entities, relationships between entities, and attributes of entities obtained by performing the information extraction; and / or The design metadata includes one or any multiple of the following: product name, product number, component quantity, component name, component number, component shape, component material, component size, and assembly sequence.

9. The method according to claim 1, further comprising: The production data and the structured design data are respectively converted into one of the following formats: a Resource Description Framework (RDF) triple format, an RDF quadruple format with time information, an RDFS format, or an RDFS quadruple format with time information, so as to obtain the knowledge generated based on the production data and the structured design data, respectively.

10. The method according to claim 9, further comprising: Establishing an ontology library based at least on the design data and the production data, wherein the ontology library includes a plurality of ontologies; The production data and the structured design data in RDF or RDFS triple format or RDF or RDFS quad format are mapped to the multiple ontologies to obtain the knowledge generated based on the production data and the structured design data, respectively.

11. According to the method of claim 10, the ontology library includes one or more of the following: design ontology, product ontology, equipment ontology, material ontology, physical asset ontology, process segment ontology, expert ontology, procurement ontology, and acceptance ontology.

12. The method according to claim 1, wherein: The design data comes from one or any combination of the following: product requirement documents, product assembly drawings, component design documents, component design drawings, circuit design files, circuit graphic drawings, and design logs.

13. The method according to claim 1, wherein: The production data comes from one or more of the following: a manufacturing execution system (MES), a SAP system, and production statistics.

14. According to the method of claim 1, performing the product failure root cause analysis task based on the target knowledge graph comprises: At least using entities in the target knowledge graph as nodes, using relationships between the entities as connecting edges, and calculating weights of each connecting edge; The product failure root cause analysis task is further performed using a depth-first search based on the weights of the respective connection edges.

15. The method according to claim 14, wherein the weight of each connecting edge is calculated based on one of degree centrality, eigenvector centrality, betweenness centrality or connectivity centrality.

16. The method according to claim 1, further comprising: Generate a graph using at least entities in the target knowledge graph as nodes and relationships between the entities as connecting edges; Representing the nodes and the connecting edges as vectors, and training the target knowledge graph using a graph neural network model; as well as The product failure root cause analysis task is performed based on the trained target knowledge graph.

17. The method according to claim 16, wherein: The graph neural network model includes one of the following: a graph convolutional network, a graph autoencoder, a graph generation network, a graph recurrent network, or a graph attention network.

18. A system for analyzing the root cause of product failure based on knowledge graph, comprising: Memory; A processor is coupled to the memory, the processor being configured to cause one or more units to perform the method according to any one of claims 1-17.

19. A computer readable medium storing a computer program comprising instructions which, when executed by a processor, cause one or more units to perform the method according to any one of claims 1 to 17.

Images