Spiral Fault Diagnosis Method Based on Data-Driven Incremental Fusion

Abstract

The invention discloses a spiral fault diagnosis method based on data-driven incremental fusion, comprising the following steps: collecting data points and dividing them into normal samples and fault samples; random sampling to obtain unbalanced samples with different slope rates and dividing them into four group; use the resampling method based on the division of neighbors to obtain relatively balanced samples; input it into DAE to extract fault features, incrementally merge feature patterns when there are new data, and then input SVM for fault diagnosis; select both informative and representative samples The dynamic and comprehensive evaluation of effective features and examples is carried out; the effective example set is merged with the new data, and the incremental learning process is repeated. With full consideration of sample noise and distribution characteristics, this method obtains balanced data that is conducive to accurate identification of fault types. By selecting features and instances for dynamic evaluation and incremental merging, effective information is retained and passed on, thereby realizing fast and efficient equipment failures. Incremental learning and classification diagnosis.

Description

technical field [0001] The invention relates to the technical field of fault diagnosis of bearing equipment, in particular to a spiral fault diagnosis method based on data-driven incremental fusion. Background technique [0002] Smart devices are mostly used in important fields such as industry, aviation, and national defense, and the consequences of their failures are relatively serious. In recent years, intelligent manufacturing, as the core content of Industry 4.0, has gradually developed into an important research field. At the same time, with the development of the Industrial Internet of Things, a large amount of operating data continues to emerge in the production process of large-scale equipment. Quickly and efficiently analyzing and extracting fault information through operating data, and effectively diagnosing and predicting fault types, has become a research field in the field of intelligent manufacturing. hotspot. [0003] With the deep integration of informatio...

AI Technical Summary

Problems solved by technology

Compared with the traditional incremental learning method, this method makes the fault data reach a relative balance by adding the proposed resampling technology based on the division of neighbors, and selects and transfers both informative and representative instances to retain all effective information to the maximum extent. And the dynamic forgetting weight is used to comprehensively evaluate the extracted features and selected examples, forming a new spiral fault diagnosis method dynamically connected with incremental information, which effectively solves the problem of massive, unbalanced, high noise, and unstable equipment fault data. , strong causal correlation and other characteristics, realize the unbalanced incremental learning of fault data and the dynamic evaluation and transmission of effective features and instances, and achieve the effect of accurate and efficient identification of incremental fault information in equipment operation data

Images