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

Abstract

To provide a method for product fault root cause analysis based on knowledge graphs.SOLUTION: A method disclosed herein comprises acquiring product-associated design data including semi-structured data and / or unstructured data, acquiring product-associated production data including structured data, performing information extraction on the design data to acquire structured design data, performing a knowledge fusion of knowledge generated separately based on the production data and the structured design data to construct a target knowledge graph, and performing product fault root cause analysis tasks based on the target knowledge graph.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates generally to the field of knowledge graphs, and more particularly to a method and apparatus for performing root cause analysis of product defects based on knowledge graphs. [Background technology]

[0002] In the manufacturing industry, it is necessary to control the production process to improve product quality, reduce the product defect rate, and ensure production capacity at the same time. Therefore, when a product defect occurs on the production line, it is necessary to investigate the cause as early as possible to find the problem node and ensure the subsequent production efficiency. In the traditional manufacturing era, the root cause analysis of product defects was usually performed manually, relying on the experience of experts. However, as a result of the continuous development of manufacturing systems and manufacturing processes, the possible defect nodes and product occurrence paths have become increasingly complex, making it difficult to manually find the defect cause in a short time.

[0003] Since the beginning of the intelligent manufacturing era, attempts have been made to solve the problem of root cause analysis of product defects in an intelligent way, and the method using knowledge graphs has attracted attention. A knowledge graph is a structured, machine-understandable semantic knowledge database that helps realize knowledge modeling, data integration, data standardization, and provides extended functions including data access and query. At present, knowledge graphs are usually created based on data from production systems, such as data from manufacturing execution systems (MES) and SAP systems, so that root cause analysis of product defects is only based on the corresponding knowledge from the production side.

[0004] However, in actual production and manufacturing sites, the cause of product defects may not only be on the production side, but also on the design side. Data from the production side is generally structured data stored in relational databases such as Oracle, Impala, etc., while the original data from the design side is generally semi-structured or unstructured data, and there are barriers between heterogeneous data from multiple sources, so it is not easy for computers to directly process and analyze them. Therefore, an improved solution for building a knowledge graph based on heterogeneous data from multiple sources is needed to achieve a deeper and more comprehensive root cause analysis of product defects. Summary of the Invention [Problem to be solved by the invention]

[0005] In the manufacturing industry, root cause analysis of product defects is required when product defects occur on the production line. Considering the fact that the field of intelligent manufacturing is usually knowledge intensive, it is advantageous to introduce knowledge graph technology into the field of intelligent manufacturing to perform root cause analysis of product defects. A knowledge graph is a structured semantic network that allows for structuring, managing, and understanding a huge amount of information. In particular, entities extracted from existing data / raw data, relationships between entities, and entity attributes are used to form a network to construct a knowledge graph. Semantics contained in the knowledge graph are also used to perform inferences to identify possible defect paths and further quantify the probability of defect root causes.

[0006] Therefore, it is desirable to provide a method and apparatus for performing root cause analysis of product defects based on knowledge graphs, whereby entities, relationships between entities, and entity attributes can be determined based on heterogeneous data from multiple sources from all aspects from design to production. Here, the obtained entities, relationships between entities, and entity attributes can be used to create a knowledge graph, thereby avoiding the problem that the root cause analysis of product defects based on knowledge graphs only involves the production side and cannot fully identify the defect cause. Thus, instead of building the knowledge graph based only on data from the production system, the knowledge graph can be further built based on data from the design side, expert experience, etc. This has the advantage that the inferential ability of the knowledge graph can be utilized to derive the possible path of defect occurrence in a complete and detailed manner when a manufacturing defect occurs. [Means for solving the problem]

[0007] According to one aspect of the present invention, there is provided a method for performing root cause analysis of product defects based on knowledge graphs, the method comprising: Obtaining design data related to a product, the design data including semi-structured data and / or unstructured data; acquiring production data relating to the product, the production data including structured data; performing information extraction on the design data to obtain structured data; performing knowledge fusion of separately generated knowledge based on the production data and the structured design data to generate a target knowledge graph; performing a task of root cause analysis of product defects based on the target knowledge graph; Includes.

[0008] According to a further aspect of the present invention, there is provided a system for performing root cause analysis of product defects based on a knowledge graph, the system comprising: a storage device; and a processor connected to the storage device, the processor configured to cause one or more units to perform a method according to any one of the various embodiments of the present invention.

[0009] According to a further aspect of the invention, there is provided a computer readable medium having stored thereon a computer program comprising 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 invention. [Brief description of the drawings]

[0010] Various embodiments of the claimed subject matter will now be described, by way of example only, with reference to the figures, in which the same reference numerals are used in different figures to denote the same or similar parts. [Figure 1] 1 is a flow chart of modules and data for root cause analysis of product defects based on a knowledge graph according to one embodiment of the present invention. [Diagram 2] FIG. 2 illustrates an example design of a product component according to one embodiment of the present invention. [Diagram 3] FIG. 1 is a causal chain diagram for root cause analysis of a product defect using a knowledge graph created in accordance with one embodiment of the present invention. [Figure 4] 1 is a flow chart for root cause analysis of product defects based on a knowledge graph according to one embodiment of the present invention. [Diagram 5] FIG. 1 is a block diagram of an apparatus for root cause analysis of product defects based on knowledge graph according to one embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] In the following description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present invention. However, those skilled in the relevant field will appreciate that the present invention may be actually practiced without one or more of the specific details, or by using alternative methods, components, etc. In some embodiments, well-known structures and processes have not been shown or described in detail to avoid unnecessarily obscuring the understanding of the present invention.

[0012] In the field of manufacturing, when a product defect occurs on the production line, it is important to determine the root cause as early as possible, find the problem node, and ensure the subsequent production efficiency. With the development of various intelligent manufacturing technologies, it is no longer possible to rely solely on engineers to analyze and identify the root cause of product defects, but instead, computers are used to extract information from huge amounts of data, which can assist in decision-making in analyzing the root cause of product defects. A knowledge graph can organize data from various sources into a network in the form of entities, relationships between entities, and entity attributes. It is a knowledge representation that is human-recognizable and machine-handleable, and is frequently used in root cause analysis of product defects because of its knowledge integration and inferential capabilities.

[0013] It often happens that the possible root causes of a product defect are on the production side, for example a product defect due to a malfunction of the system that manufactures the product. Therefore, it is a common practice today to create a knowledge graph based on data from the production side, for example data from Manufacturing Execution Systems (MES) and SAP systems. Data from these systems is generally structured and stored in various relational databases such as Oracle, Impala, etc., so a computer has no problem extracting the information and mapping it to an ontology to create a knowledge graph.

[0014] However, in actual production sites, product defects may occur on the design side, for example, product defects due to incorrect component dimensions in the design or incorrect material identification. Furthermore, product defects caused by the design side actually have a more widespread impact on the entire production process, so it is essential to create a knowledge graph based on design-related data. However, product design-related data is usually semi-structured or unstructured, and useful information cannot be obtained directly from a computer, as in structured data. In some cases, design-related data may be stored in semi-structured data formats, such as Extensible Markup Language (XML) format, JSON format, and Hypertext Markup Language (HTML) format. In other cases, design-related data may be stored in unstructured data formats, such as CAD format, JPG format, all applicable image formats, and PDF format.

[0015] For this purpose, the present invention proposes a method for root cause analysis of product defects based on knowledge graph, which does not rely only on production data, but more advantageously utilizes heterogeneous data from multiple sources, such as design and expert experience, thereby creating a knowledge graph, and deriving a more complete and detailed possible defect path when product defects occur, thereby providing more comprehensive decision support for root cause analysis of product defects. The following provides a specific description of the proposed method for root cause analysis of product defects based on knowledge graph, but those skilled in the art will understand that the present invention is not limited to a specific field and can be applied to any production line or production process.

[0016] 1 is a flow chart of modules and data for root cause analysis of product defects based on knowledge graph according to one embodiment of the present invention, where the dashed boxes indicate optional modules. Each module provided in this embodiment may be implemented in software and / or hardware and may be integrated into a data processing device capable of executing this embodiment or may be distributed among multiple data processing devices capable of executing this embodiment.

[0017] Referring to FIG. 1, design data D101 belonging to a product are obtained, and these design data D101 are input and stored in a design data storage unit M101 in an appropriate format, where these design data D101 can be stored as semi-structured data and / or unstructured data.

[0018] In one embodiment, the design data D101 related to the product may include product requirement documents, product assembly drawings, component design documents, component design drawings, circuit design files, circuit diagrams, design logs, etc. In particular, the circuit design files and design protocols can be stored in an XML (Extensible Markup Language) format or an HTTP (Hypertext Markup Language) format, which have a semi-structured data structure. Among them, the product requirement documents and individual element design documents can be stored in any suitable office document or text format, and the product assembly drawings, component design drawings, and circuit diagrams can be stored in CAD or image formats, such as JPG or PDF formats, which have unstructured data structures. The above-mentioned design data sources are used only as examples and do not limit the scope of the present invention.

[0019] FIG. 2 shows a diagram illustrating an example of a design of a product component according to one embodiment of the present invention. In FIG. 2, the design drawing of the product in this example is marked with a design drawing identification number ID in the upper left corner, the shape and size of the component are shown in the center, the lower left corner of field 201 usually contains text annotations of the design, and the central fields 202 and 203 usually contain dimensions of the component. For example, field 202 can number the axis dimensions of an elliptical component, and field 203 can number the sphere diameter of a spherical component. Thus, a design drawing generally contains graphic information, numerical information, and text information, and the positions of the graphic, numerical information, and text vary depending on the design drawing. To create a knowledge graph, additional processing is required to obtain useful information that can be read by a computer from a similarly large component diagram.

[0020] With further reference to FIG. 1, semi-structured and / or unstructured design data D102 is input to a design data pre-processing unit M102 for pre-processing, whereby it is transformed into structured data D103.

[0021] In one embodiment, if the design data D102 is semi-structured data, the design data pre-processing unit M102 performs an information extraction process on the design data D102.

[0022] In one embodiment, the information extraction process includes one or more of the following steps: an entity extraction step, a relationship extraction step, and an attribute extraction step. Among these, the entity extraction process identifies entities based on the design data D102. For example, the entities "component 1", "component 2", and "component 3" can be identified by the entity extraction process based on the description "the shape of component 1 is an ellipse, and component 1 includes component 2 and component 3". Among these, the relationship between entities is identified by a relationship extraction process from the design data D102. For example, the relationship extraction process can identify a relationship of "include" between "component 1", "component 2", and "component 3" based on the description "the shape of component 1 is an ellipse, and component 1 includes component 2 and component 3". Among these, attribute information of an entity is identified by an attribute extraction process based on design data D102. For example, the entity extraction process can identify the attribute "shape" of "component 1" based on the description "the shape of component 1 is an ellipse, and component 1 includes component 2 and component 3," and can also identify that component 1 has a value of "elliptical shape." In some embodiments, the attribute extraction process may be executed as an instance of a relationship extraction process.

[0023] In one embodiment, entity extraction may be performed using any one or any number of the following elements: rule-based approaches such as normal matching; statistical model-based approaches such as hidden Markov models, conditional Markov models, maximum entropy models, conditional random field models; and deep learning-based approaches such as convolutional neural networks, recurrent neural networks, and neural networks incorporating attention mechanisms such as LSTM-CRF, LSTM-CNN-CRF, transformers, etc.

[0024] In one embodiment, relationship extraction may be performed using any one or any number of the following elements: template-based approaches for relationship extraction; approaches for relationship extraction based on supervised learning (e.g., transforming relationship extraction into a classification problem and using learning models such as support vector machines and naive Bayes); approaches for relationship extraction based on weakly supervised learning (e.g., bootstrap approaches).

[0025] In one embodiment, entity extraction and relationship extraction are performed separately in sequence, while in another embodiment, entity extraction and relationship extraction are performed together, for example using methods such as BERT, LSTM-RNN, DGCNN, etc.

[0026] In one embodiment, attribute extraction may be performed using any one or any number of the following elements: unsupervised learning based approaches (e.g., rule-based slot-filling approaches, clustering-based approaches), dependency-based semi-supervised learning based approaches such as the PageRank algorithm, deep learning based approaches for sequence annotation such as the CRF model, the BI-GRU+CRF model, etc.

[0027] In one embodiment, information extraction may be performed using one or more of the following elements: word segmentation, word list search, syntactic parsing, corpus processing, named entity recognition, term extraction, part-of-speech tagging, vocabulary word role annotation, entity relationship recognition, and sequence labeling.

[0028] In another embodiment, if the design data D102 is unstructured data, the design data pre-processing unit M102 applies an optical character recognition process to convert text in the design data into a computer-readable text format before performing the information extraction process with the design data D102. After the unstructured design data D102 is converted into a computer-readable text format, the information extraction process can continue with reference to the above description.

[0029] In one embodiment, the structured data D103 may be stored in a relational database. In a further embodiment, the structured data D103 is configured in the form of design metadata with a corresponding design drawing number, where the design drawing number may correspond to an index of a data table, and the design metadata may correspond to a header of the data tables. Each data table is also considered as a class, each row of the table represents an instance or object of the class, and each column of the table represents an attribute of the class. In this case, the design metadata may be extracted from the design data D102 by an information extraction process via the design data pre-processing unit M102, and the design metadata may correspond to the extracted entities, relationships, and / or attributes. For example, the design metadata includes any one or more of the following elements: product name, product number, component number, component name, component number, component shape, component material, component size, and assembly sequence.

[0030] Still referring to FIG. 1, structured data D103 is input to a design knowledge generation unit M103, which results in design knowledge D104.

[0031] In one embodiment, the design knowledge D104 can be obtained by converting the structured data D103 into Resource Description Framework (RDF) or RDFS format.

[0032] Specifically, the structured data D103 can be converted into a Resource Description Framework (RDF) format or an RDFS format. For example, an RDF or RDFS triplet can configure the mapped data in a {subject-predicate-object} format. Also, an RDF or RDFS triplet can configure the mapped data in, for example, a {subject-attribute-attribute value} format. Since the assembly sequence of product components has an important effect on root cause analysis, time order information can be added to an RDF or RDFS triplet to obtain an RDF or RDFS quadruple.

[0033] In a further embodiment, the design knowledge D104 can be obtained by mapping the structured data D103 to an ontology in a created ontology library. In this case, the design knowledge D104 can be a knowledge graph created based on the design data D101.

[0034] The process of creating a knowledge graph first includes developing an ontology based on design data related to a product. An ontology can generally be understood as a conceptualized description of a particular domain (e.g., automobile parts production) and represents knowledge data that can be understood by experts in terms of classes and relationships. A design-related ontology can be developed in a bottom-up manner based on design data related to a product. To develop the above-mentioned ontology, various ontology development methods can be used, such as Common-KADS, TOVE, IDEF5, METHONTOLOGY, etc., which are used to develop ontologies. The ontology development can be realized using Web Ontology Language (OWL). Furthermore, in the ontology development, the ontology can also be modified or improved in a top-down manner based on the knowledge of experts by incorporating the experience of experts.

[0035] As already mentioned, the structured data D103 can be converted into the form of RDF triples or RDF quadruples, and then the design knowledge D104 can be obtained by fusing the data of the RDF triples or RDF quadruples with the created ontology OWL. In particular, the data fields of the RDF triples or RDF quadruples can be mapped to the created ontology using R2RML mapping (RDB to RDF Mapping Language). Furthermore, other mapping methods can also be used, such as, for example, direct mapping (D2RQ).

[0036] Referring to FIG. 1, production data D105 belonging to a product are obtained, and these production data D105 are input and stored in a production data storage unit M105 in a suitable format, where these production data D105 can be stored as structured data.

[0037] In one embodiment, the production data D105 related to the product may include any production related records, such as data from a factory, a production line, or a production facility, including data from a manufacturing execution system (MES), data from a SAP system, production statistics (factory data, equipment data, production line data, device data, etc.). The production data D105 is typically structured data having a format that can be easily processed by a computer, for example, data from an MES system generally has a unified format of machine={machine type, machine status...}. The production data D105 may 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.

[0038] Still referring to FIG. 1, the structured data D106 is input to a production knowledge generation unit M106, which results in production knowledge D107.

[0039] In one embodiment, similar to the design knowledge D104, the production knowledge D107 can be obtained by converting the structured data D106 into the Resource Description Framework (RDF) format or RDFS format.

[0040] In a further embodiment, similar to the design knowledge D104, the production knowledge D107 can be obtained by mapping the structured data D106 to an ontology in a created ontology library. In this case, the production knowledge D107 can be a knowledge graph created based on the production data D105.

[0041] The process of creating a knowledge graph first includes developing an ontology based on production data related to a product. An ontology can generally be understood as a conceptualized description of a particular domain (e.g., automobile parts production) and represents knowledge data that can be understood by experts in terms of classes and relationships. An ontology related to production can be developed in a top-down manner based on production data related to a product. To develop the above-mentioned ontology, various ontology development methods can be used, such as Common-KADS, TOVE, IDEF5, METHONTOLOGY, etc., which are used to develop ontologies. Ontology development can be achieved using Web Ontology Language (OWL). Furthermore, in ontology development, the ontology can also be modified or improved in a top-down manner based on expert knowledge by incorporating the experts' experience.

[0042] The process of acquiring production knowledge D107 using a module M106 for generating production knowledge based on structured data D106 can be referred to in the description of the process for acquiring design knowledge D104 based on structured data D103, so it will not be described repeatedly here.

[0043] In the root cause analysis process of actual product defects, design data and production data can make the possible defect-related paths between products, components, and production equipment more obvious, while expert experience can provide actual defect-related paths for root cause analysis of past product defects, so incorporating expert experience is also advantageous for root cause analysis of product defects.

[0044] Still referring to FIG. 1, expert data D108 relating to the expert's experience is optionally obtained, and the expert data D108 is inputted into and stored in an expert data storage unit M107 in a suitable format.

[0045] In some cases, the expert data D108 may include defect detection records, production equipment manuals, maintenance history documents, etc., where the expert data D108 may be stored as structured data, semi-structured data, and / or unstructured data. The above-mentioned expert data sources are used only as examples and are not intended to limit the scope of the present invention.

[0046] In some cases, the expert knowledge D111 may be RDF triplet or RDF quadruple data. In this case, the expert knowledge D111 may be a knowledge graph created based on the expert data D108. The process of using the expert knowledge generating unit M109 to obtain the expert knowledge D111 based on the structured data D109 can refer to the description of the process for obtaining the design knowledge D104 based on the structured data D103, and will not be repeated here. The process of using the expert data pre-processing unit M108 and the expert knowledge generating unit M109 to obtain the expert knowledge D111 based on the semi-structured data or unstructured data D109 can refer to the description of the process for obtaining the design knowledge D104 based on the semi-structured data and / or unstructured data D102, and will not be repeated here.

[0047] Still referring to FIG. 1, design knowledge D104 and manufacturing knowledge D107, and optionally expert knowledge for knowledge fusion D111, are input to a knowledge fusion unit M104, which generates fusion knowledge D112.

[0048] The purpose of knowledge fusion is to integrate knowledge at various levels (e.g., concept level and data level). In one embodiment, knowledge fusion may include any one or more of the following processes: data pre-processing, partitioning, load balancing, record linking, result evaluation, and / or result output.

[0049] When the design knowledge D104 and production knowledge D107, and the optional expert knowledge D111, are RDF triplet or RDF four-element data, the knowledge fusion aims to find equivalent instances among data from different sources. In one embodiment, the knowledge fusion includes any one or more of the following tasks: data cleansing, entity alignment, entity matching, entity disambiguation, and relationship deduplication. In this case, the fusion knowledge D112 may be RDF triplet or RDF four-element data.

[0050] When the design knowledge D104, the production knowledge D107, and the optional expert knowledge D111 are knowledge graphs created based on the design data D101, the production data D105, and the expert data D108, the purpose of the knowledge fusion is to find equivalent instances, equivalences, or concepts or attributes including relationships between multiple knowledge graphs. In one embodiment, the knowledge fusion includes any one or more of the following tasks: data cleansing, ontology alignment, ontology matching, entity alignment, entity matching, entity disambiguation, and relationship deduplication. In this case, the fusion knowledge D112 is a complete knowledge graph that integrates the design knowledge D104, the production knowledge D107, and the expert knowledge D111 (optional).

[0051] In other embodiments, other methods may also be used in the knowledge fusion process to incorporate expert experience, thereby further modifying or improving the knowledge fusion process. For example, equivalent ontologies / entities, conflicting ontologies / entities, relationships between overlapping ontologies / entities, and / or attributes that can be merged may be identified by expert labeling.

[0052] In other embodiments, the knowledge data input to the knowledge fusion unit may include knowledge data from other sources in addition to the exemplary data mentioned above, including but not limited to procurement knowledge from procurement data, knowledge from material data, and receipt knowledge from receipt data. The data sources are used as examples only and do not limit the scope of the present invention. Depending on whether the data from various sources is structured, semi-structured, or unstructured, the corresponding knowledge data can be obtained by performing similar processing with reference to the above description regarding design data, production data, and / or expert data.

[0053] In general, due to the confidentiality of various product-related data, encryption processing is required when acquiring raw data, thereby ensuring the robustness of the system, especially for raw data acquired in electronic form from clouds such as Azure Cloud. For example, when acquiring production data from MES systems or SAP systems from clouds, the raw data is almost always downloaded in the form of "key-value pairs" and at least the confidential data contained therein is encrypted.

[0054] Still referring to FIG. 1, the fusion knowledge data D112 is optionally input to a knowledge graph generation unit M110, thereby generating a knowledge graph D113.

[0055] In one embodiment, the fusion knowledge data D112 may include design knowledge D104, production knowledge D107, and optionally expert knowledge D111. In a further embodiment, the fusion knowledge data D112 may include procurement knowledge, material knowledge, and acceptance knowledge. If the fusion knowledge data D112 is RDF triplet data or RDF quadruple data, the data is mapped to ontologies in the created ontology library, thereby creating a complete knowledge graph D113. As described above, the ontology library can be built in a top-down and / or bottom-up manner based on various raw data related to the product. In one embodiment, the created ontology library includes any one or more of the following ontologies: design ontology, product ontology, equipment ontology, material ontology, real asset ontology, process segment ontology, expert ontology, procurement ontology, and acceptance ontology.

[0056] Through the above-mentioned method, the complete knowledge graph can integrate heterogeneous data from multiple sources, such as the design side, the production side, and the experience of experts, to derive more complete and detailed possible defect paths when product defects occur, and provide more comprehensive decision support for root cause analysis of product defects.

[0057] In one embodiment, the created 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 via a query language such as the SPARQL language, thus providing assistance in root cause analysis of product defects. In a further embodiment, the created knowledge graph can be visualized, where for example, the entities of the knowledge graph D113 are used as nodes and the relationships between entities are used as connecting edges to generate the graph. Also, attributes can be designed as entities, and attribute relationships between entities and entity attributes can be designed as relationships between entities, so that they can be handled in a similar manner.

[0058] Furthermore, the created knowledge graph can be used to provide a more comprehensive application. Still referring to Figure 1, the created knowledge graph D113 is input to a knowledge graph application unit M111, which performs a task for root cause analysis of product defects.

[0059] In one embodiment, the knowledge graph application unit M111 can provide a service for creating a causal chain of product defects. The service for creating a causal chain of product defects can visualize all possible defect cause paths based on, for example, defective products, components, or defect phenomena queried by a user. In a further embodiment, the knowledge graph application unit M111 can visualize all possible defect cause paths based on a knowledge graph stored in a graph database and calculate a weight for each connection edge, where a connection edge with a higher weight indicates a higher defect probability / defect frequency. In one embodiment, the weight for each connection edge can be calculated based on degree centrality, eigenvector centrality, betweenness centrality, or connection centrality.

[0060] FIG. 3 is a causal chain diagram for root cause analysis of product defects using a knowledge graph created according to one embodiment of the present invention. As shown in FIG. 3, each point represents an entity, and each connection edge represents a relationship between the entities. For example, node A represents product A, node MA represents production plant A, node PA represents production equipment A, node DA represents design document A, and the reference number with suffix B is expressed similarly to the reference number with suffix A. In the example of FIG. 3, the attributes of the product / component contained in the design file are also designed as entities, where Ma represents material, Si represents size, and Dr represents drawing. The existence of a connection edge between nodes indicates the possibility of the existence of a defect path, and a connection edge with a high weight indicates a high defect probability / defect frequency.

[0061] In one embodiment, a depth-first search (DFS) can be performed in descending order of the weights of the connection edges until the real root cause of the defect is found. For example, when performing root cause analysis of a defect on product B, the manufacturing plant MB is reached first along a connection edge with a weight of 0.6. If the manufacturing plant MB is not the cause of the defect, the design document DB is reached along a connection edge with a weight of 0.5. To determine whether the design document DB is the cause of the defect, the dimensions, materials, and drawings in the design document DB need to be checked in order from the highest weight to the lowest. If the design document DB is not the cause of the defect, the design document DA is checked with a weight of 0.3. A similar process should be continued until the exact cause of the defect is found.

[0062] In one embodiment, the knowledge graph application unit M111 can provide a search service for root causes of product defects. For example, multiple possible defect cause nodes can be provided based on defective products, defective components, or defect symptoms queried by a user. In one embodiment, a user can execute a query using the SPARQL language. In one embodiment, the returned defect cause nodes can be sorted and filtered according to the order of the weights of the connecting edges from highest to lowest, as described above.

[0063] In one embodiment, the knowledge graph application unit M111 can provide an automatic creation service of root causes of product defects. For example, if a defect occurs in a production line, the knowledge graph application unit M111 can provide a user with an inferred root cause node for the product defect based on the identified defect. In one embodiment, multiple possible defect cause nodes can be provided to the user, and the provided defect cause nodes can be sorted and filtered in descending order according to the weighting of the connecting edges as described above.

[0064] In the above embodiment, a method is described for constructing a knowledge graph using heterogeneous data from multiple sources related to a product, and providing a number of applications for root cause analysis of product defects based on the constructed knowledge graph. Based on this, in order to further improve the defect root cause analysis ability of the constructed knowledge graph, the present invention further provides a method for training the constructed knowledge graph using a machine learning model.

[0065] Still referring to FIG. 1, the created knowledge graph D113 is optionally input to a knowledge graph learning unit M112, which learns it using a machine learning model.

[0066] In one embodiment, the created knowledge graph D113 is input to a machine learning model, and supervised learning is performed based on the most likely fault path or fault node output. For example, the created knowledge graph D113 is input to the machine learning model in the form of a graph, where 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. Here, for example, as described above, the most likely fault path or fault node is filtered according to the order of the weight of the connecting edge from the largest to the smallest. In one embodiment, the machine learning model may be, for example, one or more graph neural network models, such as a graph convolution network, a graph autoencoder, a graph generation network, a graph recurrent network, or a graph attention network.

[0067] During the learning process, the machine learning model can use the real defect path or defect node as the label, where the difference between the output most likely defect path or defect node and the label is used as the loss function, and the learning of the knowledge graph D113 is achieved by minimizing the loss function. For example, the difference between the output most likely defect path or defect node and the label can be measured by cosine similarity, log-likelihood similarity, Euclidean distance, Manhattan distance, etc. Then, the learned knowledge graph D114 is input to the knowledge graph application unit M111 to perform a task for root cause analysis of product defects.

[0068] 4 is a flow chart of root cause analysis of product defects based on knowledge graph according to one embodiment of the present invention, in which the dashed box shows the optional process. Each step provided in this embodiment can be realized by one or more devices having computing capabilities in the form of software and / or hardware.

[0069] In step S401, design data related to a product can be obtained, where the design data includes semi-structured data and / or unstructured data, where the process of step S401 can be realized by, for example, the design data storage unit M101 described with reference to FIG.

[0070] In one embodiment, the design data is obtained from any one or more of the following elements: product requirement documents, product assembly drawings, component design documents, component design drawings, circuit design files, circuit diagrams, and design logs. For example, the circuit design files and design protocols can be stored in an XML (Extensible Markup Language) or HTTP (Hypertext Markup Language) format, which have a semi-structured data structure. For example, the product requirement documents and component design documents can be stored in any suitable office document or text format, and the product assembly drawings, component design drawings, and circuit diagrams can be stored in a CAD or image format, such as a JPG or PDF format, which have an unstructured data structure.

[0071] In step S402, production data related to the product can be obtained, where the production data includes structured data, in which case the process of step S402 can be realized, for example, by the production data storage unit M105 described with reference to FIG.

[0072] In one embodiment, the production data is obtained from one or more of the following areas: Manufacturing Execution System (MES), SAP system, and production statistics. The production statistics data may include factory production records, equipment production records, production line production records, and / or equipment production records.

[0073] In step S403, an optical character recognition process may optionally be applied to the design data to convert text in the design data into a computer-readable text format, where the process of step S403 may be implemented, for example, by design data pre-processing unit M102 described with reference to FIG.

[0074] In step S404, information extraction can be performed on the design data to obtain structured data. In this case, the process of step S403 can be realized, for example, by the design data pre-processing unit M102 described with reference to FIG.

[0075] In one embodiment, performing information extraction from the design data further includes performing entity extraction from the design data, whereby entities and relationships between entities can be identified from the design data. For example, the entity extraction may be performed using any one or any number of the following elements: rule-based approaches such as normal matching; statistical model-based approaches such as hidden Markov models, conditional Markov models, maximum entropy models, conditional random field models, and deep learning-based approaches such as convolutional neural networks, recurrent neural networks, and neural network-based approaches that introduce attention mechanisms such as LSTM-CRF, LSTM-CNN-CRF, transformers, etc. For example, the relationship extraction may be performed using any one or any number of the following elements: template-based approaches for relationship extraction; approaches for supervised learning-based relationship extraction (e.g., transforming relationship extraction into a classification problem and learning models such as support vector machines and naive Bayes); approaches for weakly supervised learning-based relationship extraction (e.g., bootstrap approaches).

[0076] In one embodiment, entity extraction and relationship extraction are performed separately in sequence, while in another embodiment, entity extraction and relationship extraction are performed together, for example using methods such as BERT, LSTM-RNN, DGCNN, etc.

[0077] In a further embodiment, performing information extraction from the design data further includes performing attribute extraction from the design data, whereby attributes of entities can be identified from the design data. For example, the attribute extraction may be performed using any one or any number of the following elements: unsupervised learning based approaches (e.g., rule-based slot-filling approaches, clustering based approaches), dependency-based semi-supervised learning based approaches such as the PageRank algorithm, deep learning based approaches for sequence annotation such as the CRF model, BI-GRU+CRF model, etc. In a further example, the attribute extraction process may be performed as a specific instance of a relationship extraction process and thus can be performed using similar approaches to relationship extraction.

[0078] In one embodiment, performing the information extraction process from the design data includes one or more of the following tasks: word segmentation, word list search, syntactic parsing, corpus processing, named entity recognition, term extraction, part-of-speech tagging, vocabulary word role annotation, entity relationship recognition, and sequence labeling.

[0079] In one embodiment, the structured design data is in the form of design metadata with a corresponding design drawing number, where the design drawing number may correspond to an index of a data table, and the design metadata may correspond to a header of the data tables. Each data table is considered as a class, each row of the table represents an instance or object of the class, and each column of the table represents an attribute of the class. In this case, the design metadata may be extracted by an information extraction process, and the design metadata may correspond to the extracted entities, relationships, and / or attributes. For example, the design metadata may include any one or more of the following elements: product name, product number, component number, component name, component number, component shape, component material, component size, and assembly sequence.

[0080] Optionally, in step S405, any one or more elements of expert experience, procurement data, material data, and / or receipt data may also be obtained. In this case, the process of step S405 may be realized, for example, by the expert experience storage unit M107 described with reference to Fig. 1. In this case, the data from various sources may be structured, semi-structured, or unstructured data, which may be processed with reference to the corresponding above description.

[0081] In step S406, knowledge fusion of the knowledge generated based on the production data and the structured design data is performed, so that a target knowledge graph can be generated. Optionally, this knowledge fusion can be further performed on data based on any one or more of expert experience, procurement data, material data, and receiving data to generate a target knowledge graph. Here, the process of step S406 can be realized by, for example, the design knowledge generating unit M103, the production knowledge generating unit M106, and the expert knowledge generating unit M109 described with reference to FIG. 1. Furthermore, the process of step S406 can be realized by, for example, the knowledge fusion unit M104 described with reference to FIG. 1. Optionally, the process of step S406 can be realized by, for example, the knowledge graph generating unit M110 described with reference to FIG. 1.

[0082] In one embodiment, the production data and the structured design data are converted into one of the following formats, namely Resource Description Framework (RDF) triplet format, RDF four-element format with temporal information, RDFS triplet format or RDFS four-element format with temporal information, respectively, so as to obtain knowledge generated based on the production data or the structured design data, respectively.

[0083] In the following embodiment, an ontology library may be established, the ontology library including a plurality of ontologies based on at least design data and production data (optionally, any one or more of the following elements: expert experience, procurement data, material data, and receipt data). The production data and the structured design data are then mapped to the plurality of ontologies in RDF or RDFS triplet format or RDF or RDFS four-element format to obtain the production data based on the knowledge of the production data and the structured design data. For example, the ontologies may be developed in a bottom-up and / or top-down manner based on the raw data related to production. For example, the ontology library may include any one or more of the following ontologies: design ontology, product ontology, equipment ontology, material ontology, real asset ontology, process segment ontology, expert ontology, procurement ontology, and receipt ontology.

[0084] In one embodiment, when the knowledge based on the production data and the structured design data is RDF triplet data or RDF four-element data, the knowledge fusion includes any one or more of the following tasks: data cleansing, entity alignment, entity matching, entity disambiguation, and relationship deduplication. In this embodiment, the fused knowledge is reflected in multiple ontologies, thereby creating a target knowledge graph.

[0085] In a further embodiment, where knowledge based on production data and structured design data is reflected in multiple ontologies, knowledge fusion includes any one or more of the following tasks: data cleansing, entity alignment, entity matching, entity disambiguation, relationship deduplication, In this embodiment, the fused knowledge is the target knowledge graph.

[0086] In an optional embodiment, continuing with the above description, a similar process can be performed based on any one or more of expert experience, procurement data, material data, and / or receiving data to thereby create a target knowledge graph.

[0087] In a further embodiment, during knowledge fusion, equivalent ontologies / entities, conflicting ontologies / entities, relationships between overlapping ontologies / entities, and / or attributes that can be merged can be identified by expert labeling.

[0088] In step S407, a task of analyzing the cause of a product defect can be executed based on the target knowledge graph. The process in step S405 can be realized by, for example, the knowledge graph application unit M111 described with reference to FIG.

[0089] In one embodiment, performing a task of root cause analysis of product defects based on the target knowledge graph includes using at least entities in the target knowledge graph as nodes, using relationships between the entities as connecting edges, and calculating a weight for each connecting edge, and further includes using a depth-first search, thereby performing the task of root cause analysis of product defects based on the weight for each connecting edge. For example, the weight for each connecting edge may be calculated based on degree centrality, eigenvector centrality, betweenness centrality, or connection centrality. For example, the task of root cause analysis of product defects may include a service for generating a causal chain for product defects, a service for searching for root causes of product defects, and / or a service for automatically creating root causes for product defects.

[0090] Optionally, in step S408, the machine learning model can be used to learn the target knowledge graph, and then the task of analyzing the root cause of the product defect from step S407 can be performed based on the learned target knowledge graph. The process of step S408 can be realized by, for example, the knowledge graph learning unit M112 described with reference to FIG. 1.

[0091] In one embodiment, using the machine learning model for learning the target knowledge graph includes 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 connecting edges as vectors, and using a graph neural network model to learn the target knowledge graph. For example, the graph neural network model may be any one or any combination of a graph convolutional network, a graph autoencoder, a graph generative network, a graph recurrent network, or a graph attention network.

[0092] In one embodiment, using the machine learning model for learning the target knowledge graph includes using a difference between the most likely fault path or fault node output from the target knowledge graph and the real fault path or fault node as a loss function, and learning the target knowledge graph is achieved by minimizing the loss function. For example, the difference between the most likely fault path or fault node output and the label can be measured by cosine similarity, log-likelihood similarity, Euclidean distance, Manhattan distance, etc.

[0093] FIG. 5 is a block diagram of an apparatus for root cause analysis of product defects based on knowledge graph according to one embodiment of the present invention.

[0094] The exemplary 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, which is used to execute commands stored in a storage device 506 to realize the method for root cause analysis of product defects based on knowledge graphs as described in detail above. The storage device 506 is suitable for delivering computer program commands and computer program data and may include various forms of storage devices, such as semiconductor storage devices such as EPROM, EEPROM and flash memory, data carriers such as internal hard disks and removable disks, magneto-optical data carriers, CD-ROM data carriers, etc. Furthermore, the computing device 500 may further include an input / output (I / O) interface 508, which allows a user to issue various commands and input data by connecting various I / O devices (e.g., cursor control devices such as a mouse, keyboard, etc.) to the computing device 500 via the I / O interface 508. Furthermore, the computing device 500 may also include a display unit 510 for displaying a graphical user interface.

[0095] The computer program may include computer executable instructions that are used to cause the processor 504 of the computing device 500 to execute the method for root cause analysis of product defects based on knowledge graph of the present invention. The program may be stored on a respective data carrier including a memory device. For example, the program may be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or combinations thereof. The steps of the process / method described in the present invention are performed by a programmable processor executing program commands, which may cause the method, process, operation to be performed by processing input data and generating output.

[0096] It should be noted that "based on" in this specification means "based at least on" and not "based only on". In addition to what is described in this specification, various modifications can be made to the disclosed embodiments and implementations without departing from the scope of the disclosed embodiments and implementations. Accordingly, the descriptions and examples contained herein should be interpreted in an illustrative sense and not in a limiting sense. The scope of the present invention should be determined solely by reference to the claims.

Claims

1. 1. A method for performing root cause analysis of product defects based on a knowledge graph, comprising: Obtaining design data relating to a product, the design data including semi-structured data and / or unstructured data; obtaining production data associated with the product, the production data including structured data; performing information extraction on the design data to obtain structured design data; performing knowledge fusion of separately generated knowledge based on the production data and the structured design data to generate a target knowledge graph; performing a task of root cause analysis of product defects based on the target knowledge graph; The method includes:

2. Generating expert knowledge based on the experience of experts; fusing separately generated knowledge based on the production data and the structured design data with the expert knowledge to construct the target knowledge graph; The method of claim 1 further comprising:

3. 10. The method of claim 1 , further comprising applying an optical character recognition process to the design data to convert text within the design data into a computer readable text format prior to performing the information extraction on the design data.

4. The method of claim 1 , wherein the information extraction performed 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.

5. The method of claim 4 , wherein the information extraction performed on the design data includes performing attribute extraction on the design data to identify attributes of entities from the design data.

6. 2. The method of claim 1, wherein the information extraction performed on the design data includes one or more of the following tasks: word segmentation, word list search, syntactic parsing, corpus processing, named entity recognition, term extraction, part of speech tagging, vocabulary word role annotation, entity relationship recognition, sequence labeling.

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

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

9. 2. The method of claim 1, further comprising converting the production data and the structured design data separately into one of the following formats: Resource Description Framework (RDF) three-element format, RDF four-element format with time information, RDFS format, or RDFS four-element format with time information, thereby obtaining knowledge that is generated separately based on the production data and the structured design data.

10. creating an ontology library including a plurality of ontologies based on at least the design data and the production data; mapping the production data and the structured design data having an RDF or RDFS three-element format or an RDF or RDFS four-element format to the multiple ontologies to obtain separately generated knowledge based on the production data and the structured design data; The method of claim 9 further comprising:

11. 11. The method of claim 10, wherein the ontology library includes any one or more of the following elements: a design ontology, a product ontology, an equipment ontology, a material ontology, a real asset ontology, a process segment ontology, an expert ontology, a procurement ontology, and an acquisition ontology.

12. 2. The method of claim 1 , wherein the design data is derived from any one or more of the following elements: product requirements documents, product assembly drawings, component design documents, component design drawings, circuit design files, circuit diagrams, and design logs.

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

14. The root cause analysis of the product defect performed based on the target knowledge graph includes: using at least the entities in the target knowledge graph as nodes and relationships between the entities as connecting edges; Calculating a weight for each connecting edge; further performing a task of root cause analysis of product defects using a depth-first search based on the weighting of each connecting edge; The method of claim 1 , comprising:

15. The method of claim 14 , wherein the weighting of each connection edge is calculated based on one of degree centrality, eigenvector centrality, betweenness centrality, or connection centrality.

16. using at least the entities in the target knowledge graph as nodes and relationships between the entities as connecting edges to generate the graph; representing the nodes and connecting edges as vectors; learning the target knowledge graph using a graph neural network model; performing a task of root cause analysis of product defects based on the learned target knowledge graph; The method of claim 1 further comprising:

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

18. 1. A system for performing root cause analysis of product defects based on a knowledge graph, comprising: A storage device; a processor connected to the storage device; Equipped with A system, wherein the processor is configured to cause one or more units to perform the method of any one of claims 1 to 17.

19. A computer readable medium having stored thereon a computer program comprising instructions which, when executed by a processor, cause one or more units to carry out a method according to any one of claims 1 to 17.

Images