Search

CN-122020042-A - Crimping quality diagnosis method and system based on feature guidance and domain countermeasure migration

CN122020042ACN 122020042 ACN122020042 ACN 122020042ACN-122020042-A

Abstract

The invention discloses a crimping quality diagnosis method and a crimping quality diagnosis system based on feature guidance and domain countermeasure migration, and belongs to the technical field of quality detection of electrical systems. The invention realizes high-precision classification by constructing an integrated diagnosis system comprising multi-domain feature extraction and preprocessing, adopts a single-domain mixed guide network, adopts a cross-domain double-alignment domain adaptation network to realize discrimination alignment and statistics alignment of inter-domain features, reduces domain difference and adapts to various input modes, and finally generates a high-precision interpretable comprehensive quality diagnosis result. Compared with the traditional crimping quality detection method, the technical scheme of the invention remarkably improves the diagnosis precision, the cross-domain generalization capability and the robustness to category missing scenes, thereby providing an intelligent and automatic innovative solution for the crimping process quality control of a modern electrical system.

Inventors

  • QIAN QUAN
  • Cai ze

Assignees

  • 上海大学

Dates

Publication Date
20260512
Application Date
20260129

Claims (10)

  1. 1. A crimp quality diagnosis method based on feature guidance and domain countermeasure migration is characterized by comprising the following steps: acquiring an original crimping force curve, then carrying out multi-domain feature extraction on the original crimping force curve, carrying out feature correlation analysis and complementarity verification on the extracted features, and reserving features capable of comprehensively describing the physical characteristics and quality association of the crimping process; aiming at the crimping data characteristics of the same specification and the same equipment, an intelligent mixed guiding network is adopted for quality diagnosis; aiming at the crimping data characteristics among different specifications and different devices, a double-alignment hybrid domain adaptation network is adopted to carry out cross-domain quality diagnosis; the method comprises the steps of constructing a crimping quality diagnosis system comprising four modules of data domain analysis, single domain diagnosis, cross-domain migration and result analysis, automatically identifying data domain attributes and adaptively selecting a diagnosis strategy by using the system, and generating a comprehensive quality diagnosis result.
  2. 2. The method for diagnosing crimp quality according to claim 1, wherein the multi-domain feature extraction includes extracting time domain features, frequency domain features and time-frequency domain features from the original crimp force curve, respectively, using a feature engineering module, and the extracted features are visualized through feature correlation analysis and thermodynamic diagram to verify complementarity of the multi-domain features.
  3. 3. The method for diagnosing crimp quality according to claim 2, wherein the time domain feature extraction method is to extract basic statistical features, ordinal statistical features, linear fitting features, polynomial fitting features, variation features and peak related features from an original crimp force curve.
  4. 4. The method for diagnosing crimping quality according to claim 2, wherein the frequency domain feature extraction method is that the original crimping force curve is subjected to fast fourier transform, and the spectrum center of gravity, the spectrum bandwidth, the spectrum roll-off point and the spectrum entropy are extracted.
  5. 5. The crimp quality diagnostic method according to claim 2, wherein the time-frequency domain feature extraction method is to extract wavelet energies A4, D3, D2, D1 by performing 4-layer decomposition on an original crimp force curve using wavelet transform.
  6. 6. The crimp quality diagnostic method according to any one of claims 1 to 5, wherein the quality diagnostic using the intelligent hybrid guidance network specifically comprises the steps of: Designing an original curve encoder and a multi-domain feature encoder, and extracting advanced feature representation based on a multi-scale gating residual block and a depth feedforward network respectively; Constructing three auxiliary supervision and guide modules of curve guidance, characteristic guidance and integrated guidance, and jointly optimizing a main classification task and an auxiliary task through a comprehensive loss function, wherein the curve guidance module strengthens the distinguishing property of curve characteristics, the characteristic guidance module improves the manual characteristic classification capability, and the integrated guidance module introduces soft label supervision of a pre-training XGBoost model to strengthen model generalization and boundary learning capability; designing a depth classifier containing a gate control residual block as a main classification module, and realizing end-to-end training by jointly minimizing total loss; only the main classifier module is retained in the reasoning phase, all the guide modules are discarded, and the prediction of the new sample is given based on the output of the main classifier.
  7. 7. The method for diagnosing crimp quality according to any one of claims 1 to 5, wherein the step of performing cross-domain quality diagnosis using the dual-aligned hybrid domain adaptive network comprises the steps of: The feature extraction and fusion network of the inherited intelligent hybrid guide network is used as a shared backbone, three modes of original curve, multi-domain feature and hybrid input are supported, and a unified feature representation basis is provided for a source domain and a target domain; designing a domain classifier and optimizing a training strategy thereof, and prompting a feature extractor to generate domain invariant features through domain countermeasures so as to inhibit domain specific information and discriminant alignment; Based on the maximum mean difference MMD design distribution alignment loss, directly measuring and reducing the characteristic distribution difference of a source domain and a target domain in a regeneration kernel Hilbert space, and improving alignment robustness by matching with a multi-core function; And realizing the collaborative optimization of discriminant alignment and statistical alignment through the weighted combination of domain antagonism loss, MMD loss, classification loss and auxiliary loss.
  8. 8. The method for diagnosing crimp quality according to any one of claims 1 to 5, wherein the specific steps of generating the comprehensive quality diagnostic report using the crimp quality diagnostic system are as follows: Carrying out feature extraction and domain characteristic analysis on input data by using a data domain analysis module, and judging whether the data belongs to same domain data or cross-domain data; the intelligent mixed guiding network IHGNet is called by using the single-domain diagnosis module to carry out quality detection on the crimping data characteristics of the same specification and the same equipment; Using a cross-domain migration module to call a double-alignment hybrid domain adaptation network DAHDANet to carry out cross-domain quality detection on crimping data characteristics among different devices with different specifications; and carrying out deep analysis, credibility evaluation and visual display on the prediction result by using a diagnosis result analysis and monitoring module.
  9. 9. A crimp quality diagnostic system based on feature guidance and domain countermeasure migration, wherein the system is configured to perform the diagnostic method according to any one of claims 1 to 8, and the system includes a feature engineering module, a data domain analysis module, a single domain diagnostic module, a cross domain migration module, and a diagnostic result analysis and monitoring module, each module cooperatively implementing crimp quality diagnosis, and the functions of each module are as follows: The characteristic engineering module is used for carrying out multi-domain characteristic extraction and data preprocessing on an original crimping force curve, extracting time domain characteristics, frequency domain characteristics and time-frequency domain characteristics, and providing quality characterization information after characteristic correlation analysis and complementarity verification; The data domain analysis module is used for carrying out feature extraction and domain characteristic analysis on input data, automatically identifying whether the data belongs to the same domain data or cross-domain data, and providing basis for diagnosis strategy selection; The single-domain diagnosis module is internally provided with an intelligent hybrid guidance network IHGNet, which is provided with a dual-path input structure, a curve guidance, a characteristic guidance and an integrated guidance triple supervision mechanism and is used for classifying quality grades of crimping data of the same specification and the same equipment; The cross-domain migration module is internally provided with a double-alignment mixed domain adaptation network DAHDANet, the network inherits the characteristic extraction and fusion backbone network of IHGNet, an integrated domain counter training and maximum mean difference MMD measurement double-alignment mechanism is adopted, three modes of original curve, multi-domain characteristics and mixed input are supported, and the cross-domain quality diagnosis is carried out on crimping data between different specifications and different devices; the diagnosis result analysis and monitoring module is used for carrying out deep analysis, credibility evaluation and visual display on the diagnosis result and outputting decision support information with engineering value.
  10. 10. The crimp quality diagnostic system of claim 9 wherein the intelligent hybrid guidance network IHGNet comprises a raw curve encoder, a feature encoder, a triple auxiliary guidance module, and a depth classifier, and wherein the dual-aligned hybrid domain adaptation network comprises a shared feature extractor, a domain classifier, and a loss optimization unit.

Description

Crimping quality diagnosis method and system based on feature guidance and domain countermeasure migration Technical Field The invention belongs to the technical field of quality detection of electrical systems, and particularly relates to a crimping quality diagnosis method and system based on characteristic guidance and domain countermeasure migration. Background With the rapid development of high-end manufacturing fields such as new energy automobiles and smart grids, the crimping connection of wires and terminals has become a key foundation for guaranteeing safe and stable operation of an electrical system. The crimping technology realizes non-solder permanent connection through mechanical extrusion, has excellent conductivity and mechanical stability, and the quality directly determines the contact resistance, mechanical strength and service life of the connecting part. Even small crimp defects (such as wire strand missing, insulation crimp, etc.) may cause serious consequences such as poor contact, local overheating, even system disconnection, etc., so real-time accurate detection of crimp quality is of paramount importance. With the deep advancement of industry 4.0, the production line is transformed to flexible manufacturing of multiple varieties and small batches, and the crimping data presents obvious domain distribution differences due to factors such as conductor terminal pairs of different specifications, batch raw material supply, equipment wear aging and the like. The traditional crimping quality detection method faces a plurality of limitations: 1. The characteristic utilization is single, is difficult to comprehensively characterize crimping quality characteristics: Most of the existing methods only depend on time sequence characteristics of an original crimping force curve or independent manual extraction characteristics, and complementary values of the two types of characteristics cannot be fully exerted. The original time sequence features contain complete dynamic information, the redundancy is high, the physical meaning of the manual features is clear, the information dimension is limited, and the complex physical characteristics and quality association relation of the crimping process are difficult to comprehensively describe by a single feature utilization mode. 2. The cross-domain generalization capability is insufficient, and the effect of adapting to a multi-specification scene is poor: the traditional threshold method relies on expert experience to set a fixed threshold, has poor robustness and is difficult to adapt to parameter changes of conductor terminal pairs with different specifications. Although the data driving method is good in a single equipment scene, when facing to cross-domain scenes of different specifications and different equipment, the model generalization capability is obviously reduced due to domain distribution difference, and the requirements of industrial field multi-variety and small-batch flexible manufacturing cannot be met. 3. Class-missing scene is not well-adapted, and migration performance is severely attenuated: sample scarcity or even absence of some fault classes in industrial sites is a common phenomenon, and existing cross-domain migration methods are not well adapted to such scenarios. When the source domain lacks the key fault class of the target domain, migration performance can be greatly attenuated, and the challenges of unbalanced class distribution in the actual industrial environment cannot be effectively met. The problem of insufficient model generalization caused by domain distribution difference can be relieved in the theory of cross-domain migration learning, but the current cross-domain migration research aiming at crimping quality diagnosis scenes is still in a starting stage, and the special requirements of crimping quality diagnosis are difficult to meet by the existing general cross-domain migration method. Therefore, there is a need for an intelligent, automated cross-domain quality diagnostic system and method that can accommodate the variety and complexity of crimp data to overcome the above-mentioned dilemma and provide a comprehensive guarantee for the reliability of crimp quality, thereby meeting the urgent need for high quality crimp connections for modern electrical systems. Disclosure of Invention The invention aims to overcome the limitations of the traditional crimping quality detection method in the aspects of single algorithm application, insufficient cross-domain adaptation capability, imperfect feature utilization and the like, and provides a crimping quality diagnosis method and system based on feature guidance and domain countermeasure migration. According to the invention, by integrating multi-domain feature extraction, a single-domain mixed guiding network and a cross-domain double-alignment domain adaptation network, various diagnosis results are unified and integrated to generate a high-precision and interpre