Search

CN-122021023-A - Fault modeling comprehensive evaluation system and evaluation method for test failure scene

CN122021023ACN 122021023 ACN122021023 ACN 122021023ACN-122021023-A

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a fault modeling comprehensive evaluation system of a test failure scene, which comprises a product information acquisition module, a test reliability weight calculation module, a fault propagation relation modeling module, a diagnosis data matching library construction module and a fault detection capability evaluation module which are sequentially connected in a communication mode, wherein the system is further provided with a data synchronization unit and a model iteration unit, and the data synchronization unit realizes real-time data interaction among the modules. According to the invention, through multidimensional acquisition and standardized processing of product related data and combination of a dynamic weight calculation mechanism and accurate fault propagation network modeling, efficient matching of fault characteristics and test data is realized, a multidimensional quantitative evaluation system is constructed, a closed-loop optimization mechanism is formed, the accuracy, stability and adaptability of fault diagnosis under a test failure scene are effectively improved, and the problems of isolation and insufficient adaptability of a traditional system module are solved.

Inventors

  • DENG WEI
  • LUO KANG

Assignees

  • 湖南遥光科技有限公司

Dates

Publication Date
20260512
Application Date
20260130

Claims (10)

  1. 1. The system is characterized by comprising a product information acquisition module, a test reliability weight calculation module, a fault propagation relation modeling module, a diagnosis data matching library construction module and a fault detection capability assessment module which are sequentially connected in a communication mode, and further comprises a data synchronization unit and a model iteration unit, wherein the data synchronization unit realizes real-time data interaction among the modules, the model iteration unit dynamically optimizes module operation parameters based on assessment results, solves the problem of insufficient diagnosis precision caused by module isolation in the traditional system, and realizes full-flow closed-loop accurate modeling and assessment of product faults under the test failure scene.
  2. 2. The system of claim 1, wherein the product information acquisition module integrates a sensor data acquisition unit, a document analysis unit and a semantic processing unit, wherein the sensor data acquisition unit synchronously acquires real-time physical parameters such as vibration, temperature, voltage and the like when a product runs, the sampling frequency is not lower than 100Hz, the document analysis unit supports structural extraction of a PDF (portable document format), a CAD (computer aided design) product design drawing and a part BOM (binary object model) table, the semantic processing unit adopts a BERT (binary object model) model to carry out semantic analysis on unstructured fault description text, natural language is converted into a standardized fault feature label, and the label matching accuracy is not lower than 95%.
  3. 3. The system of claim 1, wherein the test credibility weight calculation module adopts a combined weighting mechanism combining a hierarchical analysis method and an entropy weight method, introduces three core influence parameters of an aging coefficient of test equipment, an ambient temperature and humidity interference factor and a signal to noise ratio of test signals, determines subjective weights of all test items through AHP, calculates objective weights based on historical test data through the entropy weight method, finally fuses according to a ratio of 7:3 to obtain the comprehensive credibility weight, establishes a test failure probability dynamic correction model, acquires operation state data of the test equipment once every 5 minutes, updates weight calculation results in real time, and solves the defect that the traditional static weighting cannot adapt to dynamic change of test working conditions.
  4. 4. The system of claim 1, wherein the fault propagation relationship modeling module constructs a fault propagation network by using a directed weighted graph, wherein nodes in the graph comprise primary fault nodes, secondary fault nodes and test failure nodes, the edge weights are calculated by the product of fault propagation probability and influence intensity, a test failure correlation matrix is introduced, matrix elements represent interference coefficients (value ranges 0-1) of a certain test failure type on a specific fault propagation path, influence of the test failure on the propagation relationship is quantized by matrix operation, dynamic update and visual display of the fault propagation path are realized, and quick positioning of a key propagation link is supported.
  5. 5. The system of claim 1, wherein the diagnostic data matching library construction module comprises a feature vector extraction unit, a sample library updating unit and a matching model training unit, wherein the feature vector extraction unit performs feature extraction on fault data by adopting a CNN neural network to generate a standardized feature vector with 256 dimensions, the sample library integrates historical fault data, simulation test failure samples and industry standard fault cases, supports classified storage according to product types and fault grades, the matching model training unit performs similarity calculation on the feature vector by adopting a Siamese network, and improves the distinguishing capability of a model on similar faults by adopting an online hard negative case mining strategy, and the matching response time of a novel test failure scene is not more than 3 seconds.
  6. 6. A fault assessment method based on the system of any one of claims 1-5, comprising the steps of: s1, a product information acquisition module completes acquisition, cleaning and standardized pretreatment of multidimensional product data, and removes abnormal data (eliminates data which deviate from a mean value by 3 times of standard deviation); s2, a test credibility weight calculation module combines real-time working condition parameters to calculate dynamic comprehensive weight values of all test items; s3, the fault propagation relation modeling module constructs a dynamic fault propagation network based on the dynamic weight value and the test failure incidence matrix; s4, the diagnostic data matching library realizes the accurate matching of fault characteristics and test data through the feature vector similarity calculation; And S5, the fault detection capability evaluation module outputs four core evaluation indexes of fault diagnosis accuracy, omission rate, false detection rate and fault positioning accuracy, and the quantitative evaluation of the fault diagnosis capability is completed.
  7. 7. The method of claim 6, wherein in step S2, a Weibull distribution model is adopted for calculating the test failure probability, the accumulated running time of the test equipment is taken as an independent variable, the factory life parameter and the historical failure data of the equipment are combined, a failure probability function is obtained through fitting, when the test equipment has abnormal alarm, an emergency weight correction mechanism is triggered, the credibility weight of the test item is reduced by 30% -50%, and the weight calculation result is ensured to be consistent with the actual test working condition.
  8. 8. The method of claim 6, wherein in step S3, when a fault propagation network is constructed, a greedy algorithm is adopted to optimize a network topology structure, redundant paths with propagation probability lower than 5% are removed, a test failure tracing mechanism is established, test failure nodes causing fault diagnosis deviation are located by traversing the propagation network reversely, data support is provided for maintenance of test equipment, and fine modeling and tracing of fault propagation relationship are realized.
  9. 9. The method of claim 6, wherein in the step S5, the fault detection capability assessment module adopts a fuzzy comprehensive assessment method to establish a multi-dimensional assessment system comprising four primary indexes of diagnosis precision, response speed, stability and adaptability and 12 secondary indexes, adopts an entropy weight-TOPSIS method to determine the weights of the indexes and comprehensively score, and divides the fault diagnosis capability into four grades of excellent grade (more than or equal to 90 grades), good grade (80-89 grades), qualified grade (60-79 grades) and unqualified grade (less than 60 grades) according to the scoring result, thereby realizing comprehensive quantitative assessment and grade division of the fault diagnosis capability.
  10. 10. The method of claim 6, further comprising a fault diagnosis optimization step S6, wherein based on the evaluation result and the grading of the step S5, if the diagnosis capability is qualified or less, the model iteration unit automatically adjusts the subjective-objective weight fusion proportion of the test reliability weight calculation, and simultaneously optimizes the topology structure of the fault propagation network and the training parameters of the diagnosis data matching model, and after each optimization, the steps S2-S5 are repeated for verification until the diagnosis capability reaches good or more, a closed loop mechanism of evaluation-optimization-verification is formed, and the fault diagnosis precision and stability under the test failure scene are continuously improved.

Description

Fault modeling comprehensive evaluation system and evaluation method for test failure scene Technical Field The invention relates to the technical field of fault diagnosis, in particular to a fault modeling comprehensive evaluation system and an evaluation method for a test failure scene. Background In the technical field of fault diagnosis, the traditional fault modeling and evaluation system has obvious limitations that each functional module is isolated from each other, a real-time data interaction and cooperative operation mechanism is lacked, so that diagnosis precision is insufficient, a static weighting mode is adopted in weight calculation, dynamic change of a test working condition cannot be adapted, accuracy of an evaluation result is affected, fault propagation relation modeling is extensive, interference of a test failure on a transmission path is difficult to quantify, key fault links are difficult to position, a perfect closed loop optimization mechanism is lacked, and when a novel test failure scene is faced, adaptability, stability and response efficiency of diagnosis are poor, and requirements of accurate modeling and efficient evaluation of product faults in a complex scene cannot be met. Therefore, we propose a comprehensive evaluation system and an evaluation method for fault modeling of a test failure scene to solve the problem. Disclosure of Invention (One) solving the technical problems Aiming at the defects of the prior art, the invention provides a comprehensive fault modeling evaluation system and an evaluation method for a test failure scene, which solve the problems in the background art. (II) technical scheme The system is further provided with a data synchronization unit and a model iteration unit, wherein the data synchronization unit realizes real-time data interaction among the modules, the model iteration unit dynamically optimizes the operation parameters of the modules based on the evaluation result, solves the problem of insufficient diagnosis precision caused by module isolation in the traditional system, and realizes full-flow closed-loop accurate modeling and evaluation of product faults under the test failure scene. Preferably, the product information acquisition module integrates a sensor data acquisition unit, a document analysis unit and a semantic processing unit, the sensor data acquisition unit synchronously acquires real-time physical parameters such as vibration, temperature, voltage and the like when a product runs, the sampling frequency is not lower than 100Hz, the document analysis unit supports structured extraction of a product design drawing and a part BOM table in PDF and CAD formats, the semantic processing unit adopts a BERT model to carry out semantic analysis on unstructured fault description text, natural language is converted into a standardized fault feature label, and the label matching accuracy is not lower than 95%. The test reliability weight calculation module adopts a combined weighting mechanism combining a hierarchical analysis method and an entropy weight method, introduces three core influence parameters of an aging coefficient of test equipment, an ambient temperature and humidity interference factor and a signal to noise ratio of a test signal, determines subjective weights of all test items through AHP, calculates objective weights based on historical test data through the entropy weight method, finally fuses according to the proportion of 7:3 to obtain comprehensive reliability weights, establishes a test failure probability dynamic correction model, acquires running state data of the test equipment once every 5 minutes, updates weight calculation results in real time, and solves the defect that the traditional static weighting cannot adapt to dynamic change of test working conditions. Preferably, the fault propagation relation modeling module adopts a directed weighted graph to construct a fault propagation network, wherein nodes in the graph comprise primary fault nodes, secondary fault nodes and test failure nodes, the edge weight is calculated through the product of the fault propagation probability and the influence intensity, a test failure incidence matrix is introduced, matrix elements represent the interference coefficient (value range 0-1) of a certain test failure type on a specific fault propagation path, the influence of test failure on the propagation relation is quantified through matrix operation, the dynamic update and visual display of the fault propagation path are realized, and the rapid positioning of a key propagation link is supported. The diagnosis data matching library construction module comprises a feature vector extraction unit, a sample library updating unit and a matching model training unit, wherein the feature vector extraction unit performs feature extraction on fault data by adopting a CNN neural network to generate a standardized feature vector with 256 dimensions, the sample library integrates h