Complex production line reliability analysis method based on combination of fault operation and maintenance knowledge graph and fault tree

Abstract

The invention discloses a complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree, and the method comprises the steps: constructing a multi-level production line fault operation and maintenance knowledge graph based on a production line-equipment-fault-method, inputting fault information, carrying out the fault matching query of a production line layer, constructing an available equipment sub-graph, and carrying out the fault matching query of the equipment sub-graph; a production line fault tree is constructed by analyzing the logic relation between the production line faults and the factors causing the faults, and finally qualitative analysis is conducted on the production line fault tree, the production line fault rate is quantitatively calculated, and reliability analysis of the production line is achieved. The method is simple and easy to implement, reliability analysis of a current production system can be improved from an equipment level to a whole production line level considering coupling between equipment, a mode layer is reversely standardized and verified by utilizing characteristics of the production line and fault knowledge, production management personnel are helped to understand the rule of equipment or system fault generation, and the reliability of the production system is improved. And a reasonable maintenance strategy is formulated.

Description

Technical Field

[0001] The present invention belongs to the technical field of reliability analysis, and specifically relates to a complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree. Background Art

[0002] Aviation logistics equipment products are of many types and complex parts, and the manufacturing process involves a large number of processing equipment. For a complete production line, it usually includes a large number of processing personnel groups and various types of production equipment. In the daily maintenance system, a large amount of equipment maintenance operating procedures, accident handling plans, troubleshooting manuals, troubleshooting logs and other information in text form are stored. Most of this information is semi-structured and unstructured data with low processing efficiency. These unstructured data are difficult to be used for subsequent fault diagnosis. As a structured knowledge representation tool, knowledge graph can effectively integrate the multi-level, multi-source, and multi-cause information of production line faults, build a comprehensive and visual fault operation and maintenance knowledge graph, further study the production line fault and reliability reasoning analysis methods, and realize the accurate positioning of faults and production line health assessment. Traditional knowledge graph construction is non-automatic, time-consuming and requires a lot of manual work, which is not enough to meet the requirements of modern manufacturing models. Therefore, research on the automatic construction of high-quality fault operation and maintenance knowledge graphs is particularly important.

[0003] With the continuous improvement of the automation level of production lines, their production capacity has also been significantly improved. However, the highly integrated serial production mode makes the reliability of the production line face greater challenges: once a device in the production line fails or fails, the production activities of the entire production line may be interrupted. Therefore, for the production line layout in multi-product mixed line production, a series-parallel hybrid mode is often used to organize flexible production. At the same time, it is crucial for enterprises to continue to operate efficiently and make scientific and reasonable maintenance decisions based on the operating status of multiple devices in the production line. Production line reliability analysis is an important research direction to ensure the efficient operation of the production system. Different from the existing reliability analysis that focuses on a single device or independent system, the production line reliability analysis focuses on the synergy and overall performance of the entire production line. It aims to identify potential weak links and optimize system design and operation and maintenance strategies through in-depth analysis of each link in the production process.

[0004] Current reliability analysis methods mainly include fault tree analysis (FTA), Markov chain analysis, Monte Carlo simulation, and data-driven machine learning models. Among them, fault tree analysis is widely used because of its clear logic and easy to understand characteristics. The production line is a complex system composed of multiple equipment and process links. Fault tree analysis can effectively identify fault correlations and key nodes between equipment. Existing reliability analysis methods are often targeted at a certain system or a specific device. However, in fact, there are often certain correlations and coupling effects between failures of equipment. When a device component fails, it may affect other components, and even trigger a chain reaction in extreme cases. Therefore, the reliability analysis method of complex production line systems with equipment coupling and series-parallel mixed modes is innovative and has certain theoretical significance and practical application value. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a complex production line reliability analysis method that combines a fault operation and maintenance knowledge graph with a fault tree. The reliability analysis of complex production lines with mixed series and parallel modes is performed by constructing a production line fault operation and maintenance knowledge graph and based on the fault tree.

[0006] The technical solution adopted by the present invention is: a complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree, and the specific steps are as follows:

[0007] S1. Construct a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method;

[0008] S11. Build the knowledge graph model layer for production line failure operation and maintenance;

[0009] First, we sorted out the relevant concepts of production line fault operation and maintenance knowledge, obtained the fault text through the data source for summary and refinement, determined the entities and attributes of fault operation and maintenance knowledge, and then used Protégé ontology modeling to perform detailed data definitions of each entity and attribute to obtain the production line fault knowledge ontology. Finally, we used the constructed ontology to export "RDF / XML" in Protégé and further imported it into the graph database to realize the construction of the production line fault operation and maintenance knowledge graph model layer.

[0010] Among them, the data sources include: manufacturing information system, fault log and equipment manual; the fault operation and maintenance knowledge entities in the production line fault operation and maintenance knowledge graph model layer include: equipment, components, processes, fault phenomena, fault causes, and solutions; the fault operation and maintenance knowledge attributes in the production line fault operation and maintenance knowledge graph model layer include: product, team, name, specification model, manufacturer, equipment code, purchase date, start time, end time, alarm number, and alarm text.

[0011] S12. Based on the guidance of the pattern layer constructed in step S1, knowledge is extracted from the fault text to construct a data layer of the production line fault operation and maintenance knowledge graph;

[0012] First, the preprocessed fault operation and maintenance raw data is labeled. After the data is labeled, it is divided into training set, validation set, and test set, and input into the built BERT-BILSTM-CRF entity extraction model to automatically identify and extract entities of production line fault operation and maintenance knowledge, and then use the template-based method to extract relationships to obtain the data layer of the production line fault operation and maintenance knowledge graph.

[0013] S13. Based on step S12, fault knowledge triples are obtained through entity extraction and relationship extraction, the triples are fused, and stored and visualized through Neo4j to complete the construction of a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method.

[0014] Among them, the graph database in the production line fault operation and maintenance knowledge graph data layer stores data in a graph structure, and the Neo4j graph database is selected to store and visualize the extracted triples. In the knowledge graph, entities are represented by nodes, the relationships between entities are represented by edges, and attributes are used to describe the characteristics of nodes or edges.

[0015] S2. Based on the production line fault operation and maintenance knowledge graph constructed in step S1, the fault information is input to perform production line fault matching query and construct an available equipment subgraph, that is, query the current product process data to obtain the available equipment subgraph for each process;

[0016] S3. Based on step S2, a production line fault tree is constructed by analyzing the logical relationship between the production line failure and various factors causing the failure;

[0017] S4. Based on step S3, a qualitative analysis is performed on the production line fault tree and a quantitative calculation is made on the production line failure rate to achieve reliability analysis of the production line.

[0018] Furthermore, the step S12 is specifically as follows:

[0019] S121. Annotate the pre-processed original fault operation and maintenance data;

[0020] The original fault operation and maintenance data is the fault text in step S11; the characteristics of the original fault operation and maintenance data include: structured, semi-structured, and unstructured; and the source of the original fault operation and maintenance data includes: manufacturing information system, PDF document; the original fault operation and maintenance data is preprocessed as follows:

[0021] (1) Preprocessing of raw data of fault operation and maintenance obtained by the manufacturing information system;

[0022] The original data of fault operation and maintenance acquired by the manufacturing information system includes: equipment fault logs and equipment ledgers; they are structured data stored in relational databases, with clear data meanings and standardized naming. The preprocessing includes: missing value processing, outlier processing, and date and time formatting;

[0023] Among them, the missing values ​​and abnormal values ​​in the equipment fault log and equipment ledger are processed by deleting the missing values. For the date and time, the format is unified into the ISO8601 standard format.

[0024] (2) Preprocessing of the original fault operation and maintenance data obtained from PDF documents;

[0025] The original fault operation and maintenance data obtained from the PDF document is the semi-structured data and unstructured data in the equipment manual. The data is displayed in two forms: table and plain text. The preprocessing is as follows:

[0026] For data in the form of tables, important data is integrated into one-dimensional text sequences by row, and then knowledge is extracted by named entity recognition. For data in the form of plain text, the acquired fault-related text is cleaned and annotated, and data including line breaks, spaces, useless characters, and ambiguous records are deleted.

[0027] After performing the corresponding data preprocessing operations, the production line fault operation and maintenance knowledge graph model layer constructed in step S11 is combined. The BIO method is used to annotate the important fault-related information. The BIO method expresses the category and position of the named entity of the character sequence as follows:

[0028]

[0029] The important fault-related information includes: fault phenomenon, fault cause, and solution; item i Represents the sequence of named entity pairs in the i-th production line fault text, which completely records the position and type of each named entity character in the production line fault text. Indicates the position of the jth character in the nth named entity in the i-th production line fault text sequence in the production line fault text; Represents the named entity category of the jth character in the nth named entity in the i-th production line fault text sequence. The collected text is annotated using annotation tools to obtain a data set of record number, entity category, starting character number, ending character number, and entity content.

[0030] S122: Based on step S121, the labeled data is divided into a training set, a validation set, and a test set, and input into the built BERT-BILSTM-CRF entity extraction model. The training set and the validation set are used to train the model, and then the experimental parameters are adjusted. The optimal model during training is used to automatically identify and extract entities from the test set data for production line fault operation and maintenance knowledge.

[0031] Among them, the data division ratio is set according to the actual situation; when training the model with the training set and the validation set, the three indicators of precision P, recall rate R and the harmonic mean F1 of the two are selected to evaluate the performance of the model. The calculation formulas of the three indicators are as follows:

[0032]

[0033] Among them, TP represents the number of production line fault operation and maintenance knowledge entities correctly identified by the entity extraction model; FP represents the number of production line fault operation and maintenance knowledge entities incorrectly identified by the knowledge extraction model; and FN represents the number of production line fault operation and maintenance knowledge entities that are not identified.

[0034] Entity extraction is achieved through the BERT-BILSTM-CRF algorithm model, that is, based on the BILSTM-CRF model, the BERT distributed input representation method is adopted. First, the input text sequence is converted into the corresponding word vector representation through the pre-trained BERT model. Based on the BERT output, a bidirectional LSTM layer is introduced.

[0035] The hidden states from the bidirectional LSTM are then used as features for the subsequent CRF layer. Finally, the CRF layer is used to infer the category of each word.

[0036] Among them, the input vector E of the BERT model is formed by the sum of the word vector, position vector and segment vector embedding , the expression is as follows:

[0037] E embedding =E wordpiece +E segment +E position

[0038] Among them, E wordpiece Represents word vector, E segment represents the segment vector, E position Represents the position vector; in the BERT structure, the self-attention mechanism is introduced, and the calculation expression is as follows:

[0039]

[0040] Among them, Q, K, V represent word vector matrices, T represents the transpose operation, and d kRepresents the dimension of the input vector. A single-layer bidirectional long short-term memory network BILSTM is composed of two long short-term memory networks, one of which is responsible for processing the forward information of the input sequence, and the other is responsible for processing the reverse information of the input sequence. After both the forward and reverse sequences are processed, the output results of the two LSTMs are spliced ​​together to obtain the final BILSTM output result.

[0041] S123. Based on step S122, use a template-based method to extract relationships and construct a production line fault knowledge graph data layer;

[0042] After performing named entity recognition on the fault text, the node part of the knowledge graph has been constructed. The entities obtained in step S122 are subjected to row relationship extraction using a template-based method, that is, querying in turn whether the matches contain a given entity pair. After a successful match, the corresponding triples are output, and finally the knowledge triples, that is, the edges of the knowledge graph, are obtained.

[0043] Among them, the relationship between the entity pair "Parts-Device" is "Belong_to", the relationship between the entity pair "Fault_phenomenon-Parts" is "Located", the relationship between the entity pair "Fault_cause-Fault_phenomenon" is "Cause", and the relationship between the entity pair "Solution-Fault_cause" is "Solve"; the knowledge triple records the subject and object and the relationship between them, in the form of <entity 1, relationship, entity 2>.

[0044] Furthermore, the step S2 is specifically as follows:

[0045] Based on the knowledge graph constructed in step S1, first connect to the graph database and input known information, namely product information and faulty equipment information. Then query the product's processes and the teams to which the faulty equipment belongs respectively. Then query the processing areas required for the product process and the processing areas to which the teams belong. Finally, query the available equipment in the processing area and obtain the sub-graph of available equipment for each process.

[0046] Furthermore, the step S3 is specifically as follows:

[0047] Assume that Z independent events constitute the fault tree, and the faults between events are independent of each other. Let φ(X) represent the structure function of the fault tree, and the variable x i The state variable representing the bottom event of the fault tree, φ(X) takes the value of 0 or 1, indicating that the top event fails or is normal, x i The value of is 0 or 1, indicating that the subcomponent fails or is normal. Then the structural function expression of the fault tree is as follows:

[0048]

[0049] The logical symbols of the fault tree include: "AND gate", "OR gate", and "NOT gate". The structural function expression of the fault tree "AND gate" is as follows:

[0050]

[0051] When x i When only 0 or 1 is taken, the expression is as follows:

[0052]

[0053] When the parallel system corresponds to the "AND gate" structure of the fault tree, that is, when all subcomponents fail, the system fails, that is, when all available equipment in parallel with the production line fails, the production line fails.

[0054] The structure function expression of the fault tree "OR gate" is as follows:

[0055]

[0056] When x i When only 0 or 1 is taken, the expression is as follows:

[0057]

[0058] When the parallel system uses the "AND gate" structure of the fault tree, that is, when all subcomponents fail, the system fails, that is, when all available equipment in parallel with the production line fails, the production line fails.

[0059] Sort out the logical relationships in the fault tree, connect them with the corresponding "OR gate" and "AND gate" respectively, and finally get a complete production line fault tree.

[0060] Furthermore, the step S4 is specifically as follows:

[0061] To analyze the reliability of a production line, we first analyze its internal mechanical equipment and use the mean time between failures (MTBF) and failure rate as two indicators to evaluate the reliability of the production equipment.

[0062] The mean time between failures (MTBF) refers to the average time that a mechanical device experiences during continuous normal operation, that is, the average time interval from the occurrence of one failure to the occurrence of the next failure. The MTBF calculation expression is as follows:

[0063]

[0064] Among them, r i represents the failure frequency of the i-th equipment in the evaluation period, t i It represents the actual working time of the i-th equipment during the evaluation period, and N represents the number of equipment sampled.

[0065] First, we conduct a qualitative analysis of the minimum cut set, and the expression is as follows:

[0066]

[0067] Among them, c j represents the jth minimum cut set, x i Represents the i-th bottom event in the j-th minimum cut set. The fault tree structure function expression is as follows, which means that when a minimum cut set occurs in the fault tree, the top event occurs.

[0068]

[0069] Each station of the production line is composed of multiple devices in parallel. Each device can complete the corresponding processing steps. At the same time, each station is connected in series to form the entire production line, forming a complex series-parallel system for production and manufacturing. Assume that the system has Z minimum cut sets, namely E1, E2, ..., E Z , then the probability P(T1) of the top-level event T1 of system failure is expressed as follows:

[0070]

[0071] in, It means that the input variables are first processed as non-intersecting variables and then the Boolean algebra "join" operation is performed, that is, The superscript ' indicates that the event does not occur.

[0072] After evaluating the failure rate of the production line, the importance of each equipment unit is calculated. The key importance is the key indicator of importance, which is used to measure the importance of the impact of the bottom event on the top event and identify the weak links of the system. The calculation expression is as follows:

[0073]

[0074] Among them, I cr (j) represents the critical importance of the jth equipment unit, Q1 represents the system failure probability, q j represents the failure probability of equipment unit j.

[0075] In summary, all possible failure modes that may lead to production line failures can be qualitatively identified, that is, the minimum cut set MCS of the fault tree can be obtained; at the same time, based on the obtained minimum cut set, the weak links of the system can be determined, and the probability of production line failure and the critical importance of each equipment unit can be quantitatively calculated.

[0076] Beneficial effects of the present invention: The method of the present invention constructs a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method, inputs fault information to perform production line layer fault matching query, constructs an available equipment subgraph, and then constructs a production line fault tree by analyzing the logical relationship between the production line fault and the various factors that cause the fault. Finally, the production line fault tree is qualitatively analyzed and the production line failure rate is quantitatively calculated to achieve reliability analysis of the production line. The method of the present invention is simple and easy to implement, and can upgrade the reliability analysis of the current production system from the equipment level to the entire production line level that considers the coupling between equipment. By utilizing the characteristics of the production line and the fault knowledge to reversely standardize and verify the pattern layer, a production line fault operation and maintenance knowledge graph pattern layer design scheme based on ontology and a set of production line fault operation and maintenance knowledge graph construction processes are proposed, which are helpful for production managers to understand the laws of equipment or system failures and formulate reasonable maintenance strategies; compared with the existing knowledge graph, there is more information on the production line layer and a multi-level structure. In addition, the knowledge graph is used to obtain the fault tree model of the production line for reliability analysis and calculation. Compared with the existing equipment-level and single-system fault tree analysis, it can more accurately reveal the equipment coupling effect and key node bottlenecks in the production process, thereby providing more targeted operation and maintenance strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a flow chart of a complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree according to the present invention.

[0078] Figure 2 A framework diagram is constructed for the production line fault operation and maintenance knowledge graph in an embodiment of the present invention.

[0079] Figure 3 It is a schematic diagram of the knowledge graph model layer of production line fault operation and maintenance in an embodiment of the present invention.

[0080] Figure 4 Schematic diagram of the BERT-BILSTM-CRF entity extraction model in an embodiment of the present invention.

[0081] Figure 5 This is a schematic diagram of a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method in an embodiment of the present invention.

[0082] Figure 6 This is a flow chart of production line layer fault matching in an embodiment of the present invention.

[0083] Figure 7 This is a subgraph of the available equipment query results for the "milling profile" process in an embodiment of the present invention.

[0084] Figure 8 Schematic diagram of the production line structure in an embodiment of the present invention.

[0085] Fig. 9 Schematic diagram of a production line fault tree in an embodiment of the present invention. DETAILED DESCRIPTION

[0086] The method of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0087] like Figure 1 As shown, a flowchart of a complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree of the present invention, the specific steps are as follows:

[0088] S1, such as Figure 2 As shown, a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method is constructed;

[0089] In the process of production line reliability analysis based on production line fault operation and maintenance knowledge graph and fault tree, the construction of knowledge graph and production line fault tree based on knowledge graph are crucial, which provides an effective solution for reliability analysis not only at the equipment level but also at the more complex production line level.

[0090] S11. Build the knowledge graph model layer for production line failure operation and maintenance;

[0091] The model layer of the production line fault operation and maintenance knowledge graph is constructed as follows: Figure 3 As shown in the figure, firstly, the relevant concepts of production line fault operation and maintenance knowledge are sorted out, the fault text is obtained through the data source for summary and refinement, and the entities and attributes of fault operation and maintenance knowledge are determined (the scope of production line fault knowledge domain and the purpose of knowledge graph construction are determined, and the relevant concepts in the manufacturing field and the relationship between concepts are extracted to form a knowledge system in the manufacturing field). Then, with the help of Protégé ontology modeling, detailed data definitions of each entity and attribute are performed to obtain the production line fault knowledge ontology. Finally, the constructed ontology is used to export "RDF / XML" in Protégé, and further imported into the graph database to realize the construction of the production line fault operation and maintenance knowledge graph model layer.

[0092] Among them, the data sources include: manufacturing information system, fault log and equipment manual; the fault operation and maintenance knowledge entities (core ontology elements) in the production line fault operation and maintenance knowledge graph model layer include: equipment, components, processes, fault phenomena, fault causes, and solutions; the fault operation and maintenance knowledge attributes in the production line fault operation and maintenance knowledge graph model layer include: product, team, name, specification model, manufacturer, equipment code, purchase date, start time, end time, alarm number, and alarm text.

[0093] S12. Based on the guidance of the pattern layer constructed in step S1, knowledge is extracted from the fault text to construct a data layer of the production line fault operation and maintenance knowledge graph;

[0094] BERT-BILSTM-CRF entity extraction model Figure 4 As shown in the figure, the preprocessed fault operation and maintenance raw data is first labeled. After the data is labeled, it is divided into training set, validation set, and test set, and input into the built BERT-BILSTM-CRF entity extraction model to automatically identify and extract entities of production line fault operation and maintenance knowledge, and then use the template-based method to extract relations to obtain the data layer of the production line fault operation and maintenance knowledge graph.

[0095] S13. Based on step S12, fault knowledge triples are obtained through entity extraction and relationship extraction, the triples are fused, and stored and visualized through Neo4j to complete the construction of a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method.

[0096] Among them, considering that both the knowledge graph and the graph database use similar data models, this similarity enables the graph database to store and represent the data of the knowledge graph more naturally. At the same time, the production line fault operation and maintenance knowledge graph involves complex relationships between entities, such as the relationship between equipment, processes, teams, and fault phenomena. The graph database in the production line fault operation and maintenance knowledge graph data layer stores data in a graphical structure, which can effectively process and query these complex relationships, so that the data of the knowledge graph can be more intuitively and effectively managed and analyzed. Therefore, the Neo4j graph database is selected to store and visualize the extracted triples, thereby constructing a graph like Figure 5 The production line fault operation and maintenance knowledge graph based on the multi-level production line-equipment-fault-method is shown. The production line layer describes the overall structure of the production process, which consists of multiple processes. There is a process dependency between the processes, and each process corresponds to different tasks. Specific equipment is required to complete these tasks. Each device contains multiple attribute information, and the equipment can be further divided into components. The fault layer and the method layer are mainly the faults that may occur in the equipment and the corresponding solutions. The relationship between these levels is inseparable. The method layer is to solve the faults of the fault layer, and these faults often occur in specific equipment or their components at the equipment layer. When the equipment fails, it will also affect the completion of the corresponding process at the production line layer. In the knowledge graph, entities are represented by nodes, the relationship between entities is represented by edges, and attributes are used to describe the characteristics of nodes or edges.

[0097] S2. Based on the production line fault operation and maintenance knowledge graph constructed in step S1, the fault information is input to perform production line fault matching query and construct an available equipment subgraph, that is, query the current product process data to obtain the available equipment subgraph for each process;

[0098] S3. Based on step S2, a production line fault tree is constructed by analyzing the logical relationship between the production line fault and various factors causing the fault (including the coupling effect between devices);

[0099] S4. Based on step S3, a qualitative analysis is performed on the production line fault tree and a quantitative calculation is made on the production line failure rate to achieve reliability analysis of the production line.

[0100] In this embodiment, step S12 is specifically as follows:

[0101] S121. Annotate the pre-processed original fault operation and maintenance data;

[0102] The original fault operation and maintenance data is the fault text in step S11; the characteristics of the original fault operation and maintenance data include: structured, semi-structured, and unstructured; and the source of the original fault operation and maintenance data includes: manufacturing information system, PDF document; the original fault operation and maintenance data is preprocessed as follows:

[0103] (1) Preprocessing of raw data of fault operation and maintenance obtained by the manufacturing information system;

[0104] The original data of fault operation and maintenance acquired by the manufacturing information system include: equipment fault logs and equipment ledgers; they are structured data stored in relational databases, with clear meanings and relatively standardized naming. The preprocessing includes: missing value processing, outlier processing, and date and time formatting;

[0105] Among them, the number of missing values ​​and abnormal values ​​in equipment failure logs and equipment ledgers is not large, which may be caused by human data entry errors. Therefore, the method of deleting missing values ​​is used to deal with them, which will not affect the overall data characteristics. For date and time, the format is unified into the ISO8601 standard format, which is convenient for sorting and comparison.

[0106] (2) Preprocessing of the original fault operation and maintenance data obtained from PDF documents;

[0107] The original data of fault operation and maintenance obtained from PDF documents are semi-structured data and unstructured data in the equipment manual. The data is displayed in two forms: table and plain text. These two forms of data are also the main objects of knowledge extraction. The preprocessing is as follows:

[0108] For data in the form of tables, considering the diversity of table formats in equipment manuals, table data has important structural information and cannot be directly extracted by mapping. Therefore, the important data is integrated into a one-dimensional text sequence by row, and then the knowledge is extracted by named entity recognition. For data in the form of plain text, the acquired fault-related text is cleaned and annotated. Since the corpus contains many useless characters, in order to improve the quality of algorithm training, line breaks, spaces, useless characters and data with ambiguous records are deleted.

[0109] After performing the corresponding data preprocessing operations, the production line fault operation and maintenance knowledge graph model layer constructed in step S11 is combined. The BIO method is used to annotate important fault-related information (BIO is a commonly used sequence annotation method that can avoid the problem of partial annotation information loss in the IOB method and the problem of increasing the number of predicted labels in the IBOES method). The BIO method expresses the category and position of the named entity of the character sequence as follows:

[0110]

[0111] The important fault-related information includes: fault phenomenon, fault cause, and solution; item i Represents the sequence of named entity pairs in the i-th production line fault text, which completely records the position and type of each named entity character in the production line fault text. Indicates the position of the jth character in the nth named entity in the i-th production line fault text sequence in the production line fault text; Represents the named entity category of the jth character in the nth named entity in the ith production line fault text sequence. Use annotation tools to annotate the collected text to obtain a data set of record number, entity category, starting character number, ending character number, and entity content. In this embodiment, the BIO annotation method is used to represent the entities in the production line fault field. According to the constructed production line fault operation and maintenance knowledge graph model layer, a total of 5 entities are set, including equipment, components, phenomenon, reason and solution, with a total of 10 labels.

[0112] S122: Based on step S121, the labeled data is divided into a training set, a validation set, and a test set, and input into the built BERT-BILSTM-CRF entity extraction model. The training set and the validation set are used to train the model, and then the experimental parameters are adjusted. The optimal model during training is used to automatically identify and extract entities from the test set data for production line fault operation and maintenance knowledge.

[0113] The data division ratio is set according to the actual situation. In this embodiment, the division ratio is 7:2:1. When training the model with the training set and the validation set, the three indicators of precision P, recall rate R and the harmonic mean F1 of the two are selected to evaluate the performance of the model. The calculation formulas of the three indicators are as follows:

[0114]

[0115]

[0116] Among them, TP represents the number of production line fault operation and maintenance knowledge entities correctly identified by the entity extraction model; FP represents the number of production line fault operation and maintenance knowledge entities incorrectly identified by the knowledge extraction model; and FN represents the number of production line fault operation and maintenance knowledge entities that are not identified.

[0117] Entity extraction is achieved through the BERT-BILSTM-CRF algorithm model, that is, based on the BILSTM-CRF model, the BERT distributed input representation method is adopted (existing distributed input representation methods such as Skip-gram, Word2Vec, etc., have excellent computational efficiency and can quickly and effectively map words to unique dense word vectors, but there are certain challenges in dealing with polysemy problems. In contrast, the BERT model uses dynamic word embedding technology to extract semantic features at different levels and effectively solve the polysemy problem).

[0118] First, the input text sequence is converted into the corresponding word vector representation through the pre-trained BERT model. Based on the BERT output, a bidirectional LSTM layer is introduced (this layer captures the contextual information in the sequence, helps accurately locate entity boundaries and distinguish different entity categories).

[0119] The hidden states from the bidirectional LSTM are then used as features for the subsequent CRF layer. Finally, the CRF layer is used to infer the category of each word.

[0120] Among them, the input vector E of the BERT model is formed by the sum of the word vector, position vector and segment vector embedding , the expression is as follows:

[0121] E embedding =E wordpiece +E segment +E position

[0122] Among them, E wordpiece Represents word vector, E segment represents the segment vector, E positionRepresents the position vector; segment vector is used to distinguish different sentences; position vector is used to introduce the relative position information of words in the sequence in the subsequent self-attention model, and provide the model with information about the position of words in the sequence through position embedding; word vector improves the model's processing ability for multi-language and complex context language structures by dividing words into subwords; in the BERT structure, the self-attention mechanism is introduced, and the calculation expression is as follows:

[0123]

[0124] Among them, Q, K, V represent word vector matrices, T represents the transpose operation, and d k Represents the dimension of the input vector. The attention mechanism can improve the weight parameters, highlight the important knowledge in the text, and focus on the extraction of important knowledge.

[0125] The single-layer bidirectional long short-term memory network (BILSTM) is composed of two long short-term memory networks, one of which is responsible for processing the forward information of the input sequence, and the other is responsible for processing the reverse information of the input sequence. After both the forward and reverse sequences are processed, the output results of the two LSTMs are spliced ​​together to obtain the final BILSTM output result. In this embodiment, for the text "spindle vibration", the forward LSTM inputs "spindle", "axis", "vibration", and "motion" in sequence to obtain four vectors {h L0 ,h L1 ,h L2 ,h L3}, the backward LSTM inputs "movement", "vibration", "axis", and "main" in sequence to obtain four vectors {h R0 ,h R1 ,h R2 ,h R3}Finally, the forward and backward hidden vectors are concatenated to obtain {[h L0 ,h R3 ],[h L1 ,h R2 ],[h L2 ,h R1 ],[h L3 ,h R0 ]}.

[0126] S123. Based on step S122, use a template-based method to extract relationships and construct a production line fault knowledge graph data layer;

[0127] Considering that the knowledge graph constructed is a production line fault operation and maintenance knowledge graph, it has a strong degree of professionalism and requires high accuracy. At the same time, the record format of engineering documents related to fault operation and maintenance is relatively uniform, so a template-based method is used for relationship extraction. After the fault text named entity recognition is performed, the node part of the knowledge graph has been constructed. The entity obtained in step S122 is extracted using a template-based method, that is, the match is queried in turn to see whether it contains a given entity pair. After a successful match, the corresponding triple is output, and finally the knowledge triple is obtained, that is, the edge of the knowledge graph.

[0128] Among them, the relationship between the entity pair "Parts-Device" is "Belong_to", the relationship between the entity pair "Fault_phenomenon-Parts" is "Located", the relationship between the entity pair "Fault_cause-Fault_phenomenon" is "Cause", and the relationship between the entity pair "Solution-Fault_cause" is "Solve"; the knowledge triple records the subject and object and the relationship between them, in the form of <entity 1, relationship, entity 2>.

[0129] In this embodiment, step S2 is specifically as follows:

[0130] Fault matching means that if an abnormality is found in related equipment or components during the operation of the production line, the fault phenomenon and the source of the abnormality are analyzed to match the fault cause, solution, related processing area and product that may cause the equipment to produce the fault phenomenon.

[0131] Through a production line layer matching method based on graph data, based on the knowledge graph constructed in step S1, fault query is performed to further obtain the query subgraph. First, connect to the graph database, input known information, namely product information and faulty equipment information, and then query the processes of the product and the team to which the faulty equipment belongs, and then query the processing area required for the product process and the processing area to which the team belongs, and finally query the available equipment in the processing area to obtain the subgraph of available equipment for each process. The fault matching query process for production line reliability analysis in this embodiment is shown in Table 1.

[0132] Table 1

[0133]

[0134] The production line in this embodiment is a small batch mixed line for producing product 1 and product 2. First, the process flow of the two products of the production line is determined. The two product processes are divided into four parts, including forging process, heat treatment process, machining process and surface treatment process, among which the machining process involves a large number of machining equipment. The machining processes of products 1 and 2 are shown in Table 2 and Table 3.

[0135] Table 2

[0136]

[0137] Table 3

[0138]

[0139] After determining the available equipment for each process, the relevant information is obtained through the designed production line layer matching method based on the graph database. The production line layer matching process is as follows: Figure 6 As shown. In this embodiment, the "milling shape" process is first matched for faults, and the team responsible for the process "milling shape" is the first team. In the production line fault operation and maintenance knowledge graph, the adjacent entities with the "Has" relationship with the process "milling shape" are queried, or the adjacent entities with the "Belong_to" relationship of the team "first team" are queried to obtain all the available equipment for the process, including CNC milling machine XK718, vertical machining center VMC1000, vertical machining center VMC1000B, such as Figure 7 Similarly, query the available equipment for the remaining processes in turn to obtain the available equipment query subgraph for the production line. Then determine the entire system structure, as shown in Figure 8 shown.

[0140] In this embodiment, step S3 is specifically as follows:

[0141] The Fault Tree Analysis (FTA) is an important method for analyzing the reliability and safety of complex electronic equipment. It takes an undesirable event in the system as the analysis target, and traces all possible causes layer by layer, and traces the direct cause of each layer, so as to find possible faults in the system. It is usually represented by an inverted tree diagram.

[0142] Assume that Z independent events constitute the fault tree, and the faults between events are independent of each other. Let φ(X) represent the structure function of the fault tree, and the variable x i The state variable representing the bottom event of the fault tree, φ(X) takes the value of 0 or 1, indicating that the top event fails or is normal, x i The value of is 0 or 1, indicating that the subcomponent fails or is normal. Then the structural function expression of the fault tree is as follows:

[0143]

[0144] The logical symbols of the fault tree include: "AND gate", "OR gate", and "NOT gate". The structural function expression of the fault tree "AND gate" is as follows:

[0145]

[0146] When x i When only 0 or 1 is taken, the expression is as follows:

[0147]

[0148] When the parallel system corresponds to the "AND gate" structure of the fault tree, that is, when all subcomponents fail, the system fails, that is, when all available equipment in parallel with the production line fails, the production line fails.

[0149] The structure function expression of the fault tree "OR gate" is as follows:

[0150]

[0151] When x i When only 0 or 1 is taken, the expression is as follows:

[0152]

[0153] When the parallel system uses the "AND gate" structure of the fault tree, that is, when all subcomponents fail, the system fails, that is, when all available equipment in parallel with the production line fails, the production line fails.

[0154] Sort out the logical relationships in the fault tree, connect them with the corresponding "OR gate" and "AND gate" respectively, and finally get a complete production line fault tree.

[0155] After determining the entire production line structure based on step S2, this embodiment constructs a production line fault tree by analyzing the logical relationship between the production line failure and the various factors that cause the failure. The processing area failure state is defined as all available equipment being in a failure state, that is, the failure of each processing area is connected to the top-level event through an "OR gate"; the production line failure state is defined as any processing area being in a failure state, that is, the failure of available equipment in each process is connected to the failure of the processing area through an "AND gate". Finally, a complete production line fault tree is obtained, such as Fig. 9 As shown, A, B, and C in the figure represent the faults in the first shift area, the second shift area, and the third shift area respectively; 1-11 represent the available equipment in each process obtained by querying the production line fault operation and maintenance knowledge graph.

[0156] In this embodiment, step S4 is specifically as follows:

[0157] A production line refers to a complex system consisting of a series of mechanical equipment, tools and workstations, which are combined in a specific order and manner to complete the manufacturing or processing process of a specific product or workpiece.

[0158] To analyze the reliability of a production line, we first need to analyze the mechanical equipment inside it, and use the mean time between failures and failure rate as two indicators to evaluate the reliability of production equipment. The mean time between failures (MTBF) refers to the average time that mechanical equipment experiences during continuous normal operation, that is, the average time interval from one failure to the next. MTBF is one of the important indicators for measuring system reliability. The larger its value, the lower the frequency of production line failures and the higher the reliability. The MTBF calculation expression is as follows:

[0159]

[0160] Among them, r i represents the failure frequency of the i-th equipment in the evaluation period, t i It represents the actual working time of the i-th equipment during the evaluation period, and N represents the number of equipment sampled.

[0161] First, we conduct a qualitative analysis of the minimum cut set, and the expression is as follows:

[0162]

[0163] Among them, c j represents the jth minimum cut set, x i Represents the i-th bottom event in the j-th minimum cut set. The fault tree structure function expression is as follows, which means that when a minimum cut set occurs in the fault tree, the top event occurs.

[0164]

[0165] The minimum cut set solution of the fault tree in this embodiment is shown in Table 4.

[0166] Table 4

[0167]

[0168] Then, this embodiment estimates the failure rate of each device based on the equipment failure record table and equipment operation history record obtained from the enterprise MDC system (manufacturing information system) in the past 10 months. Some equipment failure record tables and historical operation data are shown in Tables 5 and 6.

[0169] Table 5

[0170]

[0171] Table 6

[0172]

[0173] Then, the equipment failure frequency and total operating time are calculated at the same time to obtain the average failure interval of the above 11 equipment, and finally the equipment failure rate is calculated. The equipment failure frequency and operating time are shown in Table 7.

[0174] Table 7

[0175]

[0176]

[0177] Each station of the production line is composed of multiple devices in parallel. Each device can complete the corresponding processing steps. At the same time, each station is connected in series to form the entire production line, forming a complex series-parallel system for production and manufacturing. Assume that the system has Z minimum cut sets, namely E1, E2, ..., E Z , then the probability P(T1) of the top-level event T1 of system failure is expressed as follows:

[0178]

[0179] in, It means the operation of "disjoint union" ("logical disjoint sum"), which means that the input variables are first disjointly processed and then the Boolean algebra "and" operation is performed, that is, The superscript ' indicates that the event does not occur.

[0180] Let ak represent the failure of the above equipment 1-11, then the failure probability P(P1) of the production line is expressed as follows:

[0181]

[0182] After evaluating the failure rate of the production line, it is also necessary to calculate the importance of each equipment unit. The key importance is the key indicator of importance, which is used to measure the importance of the impact of the bottom event on the top event and identify the weak links of the system. The calculation expression is as follows:

[0183]

[0184] Among them, I cr (j) represents the critical importance of the jth equipment unit, Q1 represents the system failure probability, q j represents the failure probability of equipment unit j. The key importance of each equipment unit in this embodiment is shown in Table 8.

[0185] Table 8

[0186]

[0187] It can be seen that equipment 5, 6, and 7 have the highest critical importance, that is, they are most likely to cause failure of the production line. Fault diagnosis and risk control can be focused on the above equipment.

[0188] Based on the above process, all possible failure modes that may lead to production line failure can be qualitatively found, that is, the minimum cut set (MCS) of the fault tree can be obtained; at the same time, based on the obtained minimum cut set, the weak links of the system can be determined, and the probability of production line failure and the critical importance of each equipment unit can be quantitatively calculated.

[0189] In summary, the method of the present invention obtains production line-product-process-available equipment related information from the knowledge graph in sequence through the designed production line layer query matching method based on graph data search, and constructs a fault tree model from production line failure-processing operation area failure-equipment according to its logical relationship, and then uses the model to qualitatively analyze and quantitatively calculate the critical importance of equipment units, so as to identify the weak links and weak degrees of the production line system in real time. The method of the present invention is simple and easy to implement, and can upgrade the reliability analysis of the current production system from the equipment level to the entire production line level considering the coupling between equipment. By using the characteristics of the production line and the reverse standardization and verification of the model layer by using the characteristics of the production line and the fault knowledge, it helps production management personnel understand the laws of equipment or system failures and formulate reasonable maintenance strategies; compared with the existing knowledge graph, it has more production line layer information and a multi-level structure. In addition, the knowledge graph is used to obtain the fault tree model of the production line for reliability analysis and calculation. Compared with the existing equipment-level and single-system fault tree analysis, it can more accurately reveal the equipment coupling effect and key node bottlenecks in the production process, thereby providing a more targeted operation and maintenance strategy.

[0190] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A complex production line reliability analysis method that combines a fault operation and maintenance knowledge graph with a fault tree. The specific steps are as follows: S1. Build a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method; S11. Build a knowledge graph model layer for production line fault operation and maintenance; First, we sorted out the relevant concepts of production line fault operation and maintenance knowledge. We obtained fault text from the data source and summarized and refined it, identifying the entities and attributes of the fault operation and maintenance knowledge. Using Protégé ontology modeling, we defined each entity and attribute in detail to obtain the production line fault knowledge ontology. Finally, we used this ontology to export it to RDF / XML in Protégé and then import it into a graph database to build the production line fault operation and maintenance knowledge graph model layer. in, The data sources include: manufacturing information systems, fault logs, and equipment manuals; the fault operation and maintenance knowledge entities in the production line fault operation and maintenance knowledge graph model layer include: equipment, components, processes, fault phenomena, fault causes, and solutions; the fault operation and maintenance knowledge attributes in the production line fault operation and maintenance knowledge graph model layer include: product, team, name, specification model, manufacturer, equipment code, purchase date, start time, end time, alarm number, and alarm text; S12. Based on the guidance of the model layer constructed in step S1, knowledge is extracted from the fault text to construct the data layer of the production line fault operation and maintenance knowledge graph; First, the preprocessed raw fault operation and maintenance data is labeled. After labeling, the data is divided into training, validation, and test sets. This data is then input into the established BERT-BILSTM-CRF entity extraction model to automatically identify and extract entities related to production line fault operation and maintenance knowledge. A template-based approach is then used to extract relationships, resulting in the data layer of the production line fault operation and maintenance knowledge graph. S13. Based on step S12, fault knowledge triples are obtained through entity extraction and relationship extraction. The triples are fused and stored and visualized using Neo4j, completing the construction of a multi-level production line fault operation and maintenance knowledge graph based on production line-equipment-fault-method. The graph database in the production line fault operation and maintenance knowledge graph data layer stores data in a graph structure. The Neo4j graph database is used to store and visualize the extracted triples. In the knowledge graph, entities are represented by nodes, the relationships between entities are represented by edges, and attributes are used to describe the characteristics of nodes or edges. S2. Based on the production line fault operation and maintenance knowledge graph constructed in step S1, input the fault information to perform a production line fault matching query and construct an available equipment subgraph. That is, query the current product process data to obtain the available equipment subgraph for each process. S3. Based on step S2, a production line fault tree is constructed by analyzing the logical relationship between the production line failure and the various factors that cause the failure; S4. Based on step S3, qualitative analysis is performed on the production line fault tree and quantitative calculation of the production line failure rate is performed to achieve reliability analysis of the production line.

2. The complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree according to claim 1 is characterized in that: The step S12 is specifically as follows: S121. Label the pre-processed fault operation and maintenance raw data; The fault operation and maintenance original data is the fault text in step S11; The characteristics of the fault operation and maintenance raw data include: structured, semi-structured, and unstructured; and the sources of the fault operation and maintenance raw data include: manufacturing information system, PDF documents; the fault operation and maintenance raw data is pre-processed as follows: (1) Preprocessing of raw data of fault operation and maintenance obtained by the manufacturing information system; The raw data from fault operation and maintenance acquired by the manufacturing information system includes equipment fault logs and equipment records. This data is structured and stored in a relational database. The data has clear meaning and naming conventions. Preprocessing includes missing value processing, outlier processing, and date and time formatting. Among them, for the missing values ​​and abnormal values ​​in the equipment fault log and equipment ledger, the method of deleting missing values ​​is adopted to deal with them; for the date and time, the format is unified into the ISO8601 standard format; (2) Preprocessing of the original fault operation and maintenance data obtained from PDF documents; The original fault operation and maintenance data obtained from PDF documents are semi-structured and unstructured data in the equipment manual. The data is displayed in two forms: tables and plain text. The preprocessing is as follows: For data in tabular form, important data is integrated row by row and converted into a one-dimensional text sequence, and then named entity recognition is performed for knowledge extraction. For data in plain text form, the acquired fault-related text is cleaned and annotated, and data including line breaks, spaces, useless characters, and ambiguous records is deleted. After performing the corresponding data preprocessing operations, combined with the production line fault operation and maintenance knowledge graph model layer constructed in step S11, the important fault-related information is annotated using the BIO method. The BIO method expresses the category and position of the named entity of the character sequence as follows: The important fault-related information includes: fault phenomenon, fault cause, and solution; item i Represents the sequence of named entity pairs in the i-th production line fault text, which fully records the position and type of each named entity character in the production line fault text; Indicates the position of the jth character in the nth named entity in the i-th production line fault text sequence in the production line fault text; Represent the named entity category of the jth character in the nth named entity in the i-th production line fault text sequence; use annotation tools to annotate the collected text to obtain a dataset of record number, entity category, starting character number, ending character number, and entity content; S122: Based on step S121, the labeled data is divided into a training set, a validation set, and a test set, and input into the established BERT-BILSTM-CRF entity extraction model. The training set and the validation set are used to train the model, and then the experimental parameters are adjusted. The optimal model during training is used to automatically identify and extract entities from the test set data for production line fault operation and maintenance knowledge. The data division ratio is set according to the actual situation. When training the model with the training set and the validation set, the three indicators of precision P, recall R, and the harmonic mean F1 of the two are selected to evaluate the performance of the model. The calculation formulas of the three indicators are as follows: Among them, TP represents the number of production line fault operation and maintenance knowledge entities correctly identified by the entity extraction model; FP represents the number of production line fault operation and maintenance knowledge entities incorrectly identified by the knowledge extraction model; FN represents the number of production line fault operation and maintenance knowledge entities that are not identified; Entity extraction is achieved through the BERT-BILSTM-CRF algorithm model. That is, based on the BILSTM-CRF model, the BERT distributed input representation method is adopted. First, the input text sequence is converted into the corresponding word vector representation through the pre-trained BERT model. Based on the BERT output, a bidirectional LSTM layer is introduced. The hidden state from the bidirectional LSTM is then used as the feature of the subsequent CRF layer; finally, the CRF layer is used to infer the category of each word; Among them, the input vector E of the BERT model is formed by the sum of word vector, position vector and segment vector embedding , the expression is as follows: AND embedding =And wordpiece +E segment +E position Among them, E wordpiece Represents word vector, E segment represents the segment vector, E position Represents the position vector; in the BERT structure, the self-attention mechanism is introduced, and the calculation expression is as follows: Among them, Q, K, V represent word vector matrices, T represents the transpose operation, and d k Represents the dimension of the input vector. A single-layer bidirectional long short-term memory network (BILSTM) is composed of two long short-term memory networks, one of which processes the forward information of the input sequence, and the other processes the reverse information of the input sequence. After both the forward and reverse sequences are processed, the output results of the two LSTMs are concatenated to obtain the final BILSTM output result. S123. Based on step S122, use a template-based method to extract relationships and construct a production line fault knowledge graph data layer; After performing named entity recognition on the fault text, the node part of the knowledge graph is constructed. The entities obtained in step S122 are then subjected to row relationship extraction using a template-based method. This involves sequentially querying whether the matches contain a given entity pair. If a match is successful, the corresponding triples are output, ultimately resulting in knowledge triples, i.e., edges of the knowledge graph. Among them, the relationship between the entity pair "Parts-Device" is "Belong_to", the relationship between the entity pair "Fault_phenomenon-Parts" is "Located", the relationship between the entity pair "Fault_cause-Fault_phenomenon" is "Cause", and the relationship between the entity pair "Solution-Fault_cause" is "Solve"; the knowledge triple records the subject and object and the relationship between them, in the form of <entity 1, relationship, entity 2>.

3. The complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree according to claim 1 is characterized in that: The step S2 is specifically as follows: Based on the knowledge graph constructed in step S1, first connect to the graph database and input known information, namely product information and faulty equipment information. Then, query the processes of the product and the team to which the faulty equipment belongs. Then, query the processing area required for the product process and the processing area to which the team belongs. Finally, query the available equipment in the processing area and obtain the sub-graph of available equipment for each process.

4. The complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree according to claim 1 is characterized in that: The step S3 is specifically as follows: Assume that Z independent events constitute a fault tree, and the faults between events are independent of each other. Let φ(X) represent the structure function of the fault tree, and the variable x i The state variable representing the bottom event of the fault tree, φ(X) takes the value of 0 or 1, indicating that the top event is failure or normal, x i The value of is 0 or 1, indicating that the subcomponent fails or is normal; the structural function expression of the fault tree is as follows: The logical symbols of the fault tree include: "AND gate", "OR gate", and "NOT gate". The structural function expression of the fault tree "AND gate" is as follows: When x i When only 0 or 1 is taken, the expression is as follows: When the parallel system corresponds to the "AND gate" structure of the fault tree, that is, when all subcomponents fail, the system fails. That is, when all available equipment in parallel on the production line fails, the production line fails. The structure function expression of the fault tree "OR gate" is as follows: When x i When only 0 or 1 is taken, the expression is as follows: When the parallel system uses the "AND gate" structure of the fault tree, that is, when all subcomponents fail, the system fails. That is, when all available equipment in parallel with the production line fails, the production line fails. Sort out the logical relationships in the fault tree, connect them with the corresponding "OR gate" and "AND gate" respectively, and finally obtain a complete production line fault tree.

5. The complex production line reliability analysis method combining a fault operation and maintenance knowledge graph with a fault tree according to claim 1 is characterized in that: The step S4 is specifically as follows: To analyze the reliability of a production line, we first analyze the mechanical equipment inside it and use the mean time between failures (MTBF) and failure rate as indicators to evaluate the reliability of the production equipment. Mean Time Between Failures (MTBF) refers to the average time a mechanical device experiences during continuous normal operation, that is, the average time interval from one failure to the next. The MTBF calculation expression is as follows: Among them, r i represents the failure frequency of the i-th device during the evaluation period, t i represents the actual working time of the i-th device during the evaluation period, and N represents the number of sampled devices; First, we conduct a qualitative analysis of the minimum cut set, and the expression is as follows: Among them, c j represents the jth minimum cut set, x i Represents the i-th bottom event in the j-th minimum cut set. The fault tree structure function expression is as follows, which means that if a minimum cut set occurs in the fault tree, the top event will occur; Each station of the production line is composed of multiple devices connected in parallel. Each device can complete the corresponding processing steps. At the same time, each station is connected in series to form the entire production line, forming a complex series-parallel system for production and manufacturing. The system is assumed to have Z minimum cut sets, namely E1, E2, ..., E Z , then the probability P(T1) of the top-level event T1 of system failure is expressed as follows: in, It means "disjoint union" operation, which means that the input variables are first disjointly processed and then the Boolean algebra "union" operation is performed, that is, The superscript ' indicates that the event does not occur; After evaluating the failure rate of the production line, the importance of each equipment unit is calculated. The key importance is the key indicator of importance, which is used to measure the importance of the impact of the bottom event on the top event and identify the weak links in the system. The calculation expression is as follows: Among them, I cr (j) represents the critical importance of the jth equipment unit, Q1 represents the probability of system failure, q j represents the failure probability of equipment unit j; In summary, we can qualitatively identify all possible failure modes that may lead to production line failures, that is, calculate the minimum cut set (MCS) of the fault tree. At the same time, based on the calculated minimum cut set, we can determine the weak links in the system and quantitatively calculate the probability of production line failure and the critical importance of each equipment unit.

Images