Complex system FMECA-oriented multi-signal flow graph automatic construction technology

Abstract

The invention discloses a multi-signal flow graph automatic construction technology for a complex system FMECA. According to the FMECA-oriented multi-signal flow graph automatic construction technology, non-standard FMECA files with different expressions are automatically constructed into a multi-signal flow graph, systematic and intuitive fault propagation logic is provided for maintenance and troubleshooting personnel, and the fault diagnosis efficiency is improved. According to the FMECA-oriented multi-signal flow graph automatic construction technology, input of an FMECA file is constrained by standardizing a header and designing an ontology, and the FMECA-oriented multi-signal flow graph automatic construction technology is divided into three steps: firstly, for a fault event entity, word segmentation is performed through a dictionary, and a TF-IDF value of each segmented word is calculated to obtain a vector expression of the fault event entity; combining fault event entity pairs greater than a certain threshold value through weighted similarity analysis of cosine of fault event entity vectors and dynamic time warping; and finally, acquiring corresponding nodes of the multi-signal flow graph and a relationship thereof according to the normalized and simplified FMECA file under the constraint of the ontology, and constructing a complete multi-signal flow graph.

Description

Technical Field

[0001] The invention relates to a knowledge-driven fault diagnosis method, in particular to an automatic construction method of a multi-signal flow graph based on an FMECA file in a system where the writing of fault diagnosis knowledge lacks certain specifications. Background Art

[0002] In complex, difficult to model or data-scarce systems, knowledge-driven fault diagnosis methods detect, isolate and identify faults by building a knowledge model of the system. In particular, in the design phase, knowledge-driven fault diagnosis methods for system structure include FTA (fault tree analysis), FMECA (failure mode, effects and criticality analysis), multi-signal flow graphs and other methods.

[0003] (1) FTA uses a logic diagram to show the relationship between faults and underlying causes. As a top-down analysis method, FTA starts from system failures or unsafe states (top events), gradually decomposes them into more basic fault causes (basic events), and uses logic gates to represent the causal relationship between events. FTA can clearly identify critical fault paths and locate single or multiple basic events, but its construction process is complex and cannot exhaust all possible basic events.

[0004] (2) FMECA takes all possible failure modes of the system as the research object, starting from the components, analyzing the failure information of each component, and expanding from the bottom to the system level step by step, and finally determining the impact of each possible failure mode on the system and summarizing it into an FMECA table. Therefore, FMECA can clarify the impact of each failure mode on the system or equipment, but the FMECA tables in complex systems are numerous and huge, and the global correlation between each component and each failure mode is easily obscured. In addition, the manually written FMECA data has strong differences and poor standardization, which increases the difficulty for maintenance and troubleshooting personnel to diagnose faults.

[0005] (3) As a graphical fault diagnosis model, the multi-signal flow graph describes the relationship between system components and test logic through a directed graph. In actual modeling, the model defines each fault mode as a "signal", and the positive and abnormal states of the "signal" are monitored by the "measurement point". The signal transmission model of the system is constructed by using the dependency relationship between signals and measurement points, signals and components, and components and components. The multi-signal flow graph improves the knowledge diagnosis efficiency of complex systems by focusing on the modeling of the signal transmission path involved in the fault subject and the failure mode. However, since the modeling of multi-signal flow graphs requires the troubleshooters to have a full understanding of the system principles, manual modeling is difficult and its application is difficult.

[0006] In the above-mentioned knowledge-driven fault diagnosis methods for complex system structures, FTA can clearly identify key fault paths, but the construction process is complicated and cannot cover all basic events; FMECA can analyze the impact of each fault mode on the system in detail, but there are many tables, global correlation is easily ignored, and the standardization of manually written data is poor; multi-signal flow graphs improve the efficiency of fault diagnosis through graphical methods, but manual modeling is difficult, which limits its application scope. In view of this, the present invention proposes a multi-signal flow graph automatic construction technology based on FMECA, which combines the detailedness of FMECA analysis with the intuitiveness of multi-signal flow graphs, reduces the difficulty of multi-signal flow graph construction, and assists maintenance and troubleshooting personnel to understand the global correlation between each component and fault mode, and improves the comprehensiveness and accuracy of complex system fault diagnosis. Summary of the invention

[0007] The present invention provides a multi-signal flow graph automatic construction technology for complex system FMECA to solve the problem that the numerous and complex FMECA tables in complex systems are not standardized, which aggravates the difficulty of fault diagnosis and the difficulty of modeling multi-signal flow graphs for engineering diagnosis scenarios. This method can automatically construct non-standard and differently expressed FMECA files into multi-signal flow graphs, provide systematic and structured fault propagation logic for maintenance and troubleshooting personnel, reduce the difficulty of using FMECA files, and improve fault diagnosis efficiency.

[0008] The multi-signal flow graph automatic construction technology for FMECA is mainly aimed at knowledge-driven complex system fault diagnosis scenarios. Specifically, the multi-signal flow graph automatic construction technology for complex system FMECA constrains the input form of FMECA files by standardizing the header and designing the ontology. The main process includes three core steps: word vector generation based on TF-IDF (term frequency-inverse document frequency), entity de-redundancy based on comprehensive similarity analysis of word vectors, and automatic construction of multi-signal flow graphs based on standardized streamlined FMECA files.

[0009] The input of the standard header and design ontology constraint FMECA file normalizes the header of the FMECA file into seven main entities, and further constrains the seven entities to three ontologies of "equipment components", "fault events" and "measurement points" and their relationships; the word vector generation based on TF-IDF is performed by segmenting the fault event entity, and then calculating the TF-IDF value of each segmentation, and further vectorizing each entity according to the TF-IDF value of each segmentation; the entity de-redundancy based on word vector comprehensive similarity analysis uses the TF-IDF word vectors of two entities to weightedly calculate the cosine similarity and dynamic time warping similarity, and de-redundant entities greater than the threshold are merged; the multi-signal flow graph based on the standard simplified FMECA file is automatically constructed, based on the standard simplified FMECA file, a complete multi-signal flow graph is constructed through the constraints of the ontologies "equipment components", "fault events", "measurement points" and their relationships.

[0010] The characteristics of the present invention are:

[0011] (1) The present invention solves the problem of the difficulty of fault diagnosis knowledge management and application due to the strong difference in FMECA file data by designing a standardized FMECA file header and body constraints;

[0012] (2) The present invention proposes a word vector generation method based on TF-IDF and a comprehensive entity similarity analysis algorithm based on cosine similarity and dynamic time warping, thereby realizing the unified integration of fault event entities and solving the problem of increased ambiguity in fault diagnosis caused by differences in manual expressions in FMECA files.

[0013] (3) The present invention is based on standardized FMECA files, clarifies the causal relationship between fault events and measurement points; the propagation relationship between fault events; the attribution relationship between fault events and equipment components, and between equipment components and equipment components, and automatically constructs a multi-signal flow graph based on this, solving the problem of difficulty in manual modeling of multi-signal flow graphs. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Overall steps of the technology for automatically building multi-signal flow graphs for FMECA of complex systems

[0015] Figure 2 Standardize the header ontology constraints for FMECA

[0016] Figure 3 Generate entity word vectors for fault events based on IF-IDF

[0017] Figure 4 De-duplication of fault event entities based on comprehensive similarity analysis of word vectors DETAILED DESCRIPTION

[0018] The present invention provides a method for automatically constructing a multi-signal flow graph for a complex system FMECA file. The specific implementation of the present invention is described below in conjunction with the accompanying drawings. The multi-signal flow graph automatic construction technology for complex system FMECA constrains the input form of FMECA by standardizing the file header and the design ontology. The main process includes three core steps: word vector generation based on TF-IDF, entity deduplication based on comprehensive similarity analysis of word vectors, and automatic construction of multi-signal flow graphs based on standardized streamlined FMECA files. Figure 1 shown.

[0019] 1. Standardization of headers and ontology constraints of FMECA documents

[0020] The header standardization and ontology constraint of the FMECA file include standardizing the headers of the non-standardized FMECA table files written manually from different sources; and constraining each header in the ontology of "equipment components", "fault events", "measurement points" and the "attribution", "propagation" and "causality" relationships between them according to the standardized FMECA file headers, so as to serve the subsequent automatic construction of multi-signal flow graphs. The ontology constraints are as shown in the attached Figure 2 shown.

[0021] 1) The standardized FMECA document header and its corresponding description are shown in the following table:

[0022] Table 1 Standardized FMECA header and its meaning

[0023] FMECA file specification header Table header meaning Device Code The coding describes the relationship between components and the entire device, making it easier to analyze the hierarchy and location. Device Name The device name or model corresponding to the device code. Device Function Record the functions of products and equipment. Failure Mode The state in which the product fails to work properly is divided into potential failure and functional failure, covering various failure modes. Cause of failure Factors that lead to failure include environmental, human, physical and chemical changes, etc. Failure Effect The impact of the failure on itself, the superior product and the overall equipment. Failure Detection Methods for discovering failure sites, including pre- and post-failure detection.

[0024] 2) The specific description of constraining the standardized FMECA header to the three ontologies of "equipment components", "fault events" and "measurement points" and the three ontological relationships of "attribution", "propagation" and "cause and effect" between them is shown in the following table.

[0025] Table 2 Ontology constraints and their specific descriptions

[0026]

[0027] 2. Word vector generation based on TF-IDF

[0028] The word vector generation based on TF-IDF includes dictionary segmentation, calculation of TF-IDF value and generation of word vector. The specific process is as shown in the attached Figure 3 shown.

[0029] The dictionary segmentation is performed on the "fault event entity" extracted from the FMECA file through the jieba Chinese segmentation database, and a segmentation database of the current FMECA file is obtained.

[0030] The word vector generation based on TF-IDF includes the calculation of TF and IDF values. The TF value represents the frequency of a word in all fault event entities in the word segmentation database. The normalized calculation formula for the TF value of word i is as follows:

[0031]

[0032] The IDF value represents the frequency of the reverse document and represents the rarity of the word. If a word appears in more fault event entities, the IDF value of the word is close to 0, which means that it cannot distinguish different fault event entities well; on the contrary, the IDF value of the word is high and can be used as a keyword to distinguish different fault event entities. The specific calculation formula is shown as follows:

[0033]

[0034] The word vector is generated by using the product of the TF and IDF values ​​of each word in the word segmentation database to represent the TF-IDF value of each word segmentation, expressing the fault event entity through different word segmentation combinations, and obtaining the vector expression of each fault event entity according to the IF-IDF value of each word segmentation.

[0035] 3. Entity redundancy removal based on comprehensive similarity analysis of word vectors

[0036] The entity deduplication based on word vector comprehensive similarity analysis includes calculation of cosine similarity of fault event entity vectors, calculation based on dynamic time warping similarity and fault entity deduplication based on comprehensive similarity.

[0037] The calculation of the cosine similarity of the fault event entity vectors is performed by calculating the cosine value between the two fault event entity vectors to characterize the similarity between the two vectors. The calculation formula is as follows:

[0038]

[0039] Where: S1 is the TF-IDF vector expression of fault event entity 1, S2 is the TF-IDF vector expression of fault event entity 2, and cosθ is the cosine similarity between S1 and S2.

[0040] The similarity calculation based on dynamic time warping includes the distance matrix calculation of the two fault event entity vectors, the calculation of the dynamic time warping distance and the normalized similarity calculation. The distance matrix calculation is obtained by calculating the Euclidean distance between the values ​​of each element in the two fault event entity vectors; the dynamic time warping distance calculation finds a path with the first element in the upper left corner of the distance matrix as the starting point and the last element in the lower right corner as the end point. It is stipulated that when finding the path, it moves in the three directions of "right, bottom, and bottom right", and the sum of the distance matrix elements passed by the path is minimized, and the minimum element sum is the dynamic time warping distance; the normalized similarity calculation is calculated by the following formula:

[0041]

[0042] Where: DTW(S1,S2) is the dynamic time warping distance between the fault event entity vector S1 and the fault event entity vector S2, len(·) represents the dimension of the vector, and dtw_sim is the word vector similarity based on dynamic time warping.

[0043] The fault entity deduplication based on comprehensive similarity is to assign a weight value w1 to the cosine similarity cosθ and a weight value w2 to the dynamic time warping distance similarity dtw_sim, and w1+w2=1, and calculate the comprehensive similarity sim=w1cosθ+w2dtw_sim. Then, based on the comprehensive similarity sim, the fault event entity pairs greater than a certain threshold are merged to complete the deduplication. 4. Automatic construction of multi-signal flow graph based on standardized simplified FMECA files

[0044] The multi-signal flow graph based on the standardized and streamlined FMECA file is automatically constructed. Under the constraint of the ontology, the corresponding nodes of the multi-signal flow graph such as "equipment components", "fault events", and "measurement points" are obtained through the standardized FMECA file. The FMECA file is further constructed through the comprehensive similarity analysis between the fault event entities, and a complete multi-signal flow graph model is constructed based on the "attribution", "propagation", and "causality" relationships between the nodes. The specific steps are as follows:

[0045] 1) According to the equipment code, failure detection, failure mode, failure cause and failure impact of the standardized FMECA file, under the constraints of the "equipment components" and "fault events" entities, determine the "attribution" relationship between the equipment components that constitute the system, and the "attribution" relationship of each "fault event" in each equipment component, and on this basis, summarize the fault event nodes.

[0046] 2) According to the failure mode, failure cause and failure impact information in the standardized FMECA file, under the ontology constraint of the "failure event", determine the "propagation" relationship between each "failure event" to determine the "failure event"

[0047] The direction of transmission between nodes.

[0048] 3) According to the failure detection, failure mode, failure cause and failure impact in the standardized FMECA document, the failure detection, failure mode, failure cause and failure impact in the “measurement point”

[0049] Under the ontology constraints of the "fault event", the test nodes are generated according to their causal relationship, and the "fault event" is determined.

[0050] The transfer direction between the "measurement point".

[0051] 4) Merge all the above nodes to build a complete multi-signal flow graph.

Claims

1. A multi-signal flow graph automatic construction technology for complex system FMECA, characterized by: The FMECA-oriented multi-signal flow graph automatic construction technology includes the input form of the standard header and the design ontology constraint FMECA file, and the main steps include the word vector generation step based on TF-IDF (term frequency-inverse document frequency), the entity redundancy step based on the comprehensive similarity analysis of the word vector, and the multi-signal flow graph automatic construction step based on the standard simplified FMECA file.

2. The multi-signal flow graph automatic construction technology of FMECA according to claim 1 is characterized in that: The standard header and design ontology constraint FMECA file input form in the FMECA-oriented multi-signal flow graph automatic construction technology standardizes the header of the non-standard FMECA table file written manually from different sources into equipment code, equipment name, equipment function, failure mode, failure cause, failure impact and failure detection; and constrains the header in the ontology of "equipment component", "fault event", "measurement point" and the "attribution", "propagation" and "causality" relationship between them.

3. The multi-signal flow graph automatic construction technology of FMECA according to claim 1 is characterized in that: The TF-IDF-based word vector generation step in the FMECA-oriented multi-signal flow graph automatic construction technology is performed by performing dictionary segmentation on the fault event entities in the FMECA file, calculating the TF-IDF values ​​of each segmentation, expressing the fault event entities through different segmentation combinations, and combining the TF-IDF values ​​of each segmentation to obtain the TF-IDF vector expression of each fault event entity.

4. The multi-signal flow graph automatic construction technology of FMECA according to claim 1 is characterized in that: The entity deduplication step based on word vector comprehensive similarity analysis in the FMECA-oriented multi-signal flow graph automatic construction technology uses TF-IDF word vector weighted calculation of cosine similarity and dynamic time warping similarity, and merges fault event entity pairs with similarity greater than a certain threshold to complete deduplication.

5. The multi-signal flow graph automatic construction technology of FMECA according to claim 1 is characterized in that: The multi-signal flow graph automatic construction step based on the standardized and simplified FMECA file in the FMECA-oriented multi-signal flow graph automatic construction technology obtains the corresponding nodes of the multi-signal flow graph such as "equipment components", "fault events", "measurement points", etc. through the standardized and de-redundant simplified FMECA file under the ontology constraint, and constructs a complete multi-signal flow graph model according to the "attribution", "propagation" and "causality" relationships between the nodes.

Images