Fault diagnosis method of cross-modal bridging graph convolutional neural network

By building a cross-modal bridged graph convolution neural network, using graph convolution and cross-modal bridge mechanism to extract and fusion fault features, combined with dynamic collaboration mechanisms and multi-channel fusion Transformer module, the shortcomings of traditional fault diagnosis methods in multi-modal data processing are solved, and efficient and stable fault diagnosis effect is achieved.

CN119719878AActive Publication Date: 2025-03-28ANHUI UNIV OF SCI & TECH
8 Cites 0 Cited by

Patent Information

Application Number
CN202411366157.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-29
Publication Date
2025-03-28
Estimated Expiration
2044-09-29

AI Technical Summary

Technical Problem

Traditional fault diagnosis methods are difficult to extract comprehensive fault characteristics, and they perform poorly when processing multimodal data, resulting in unstable diagnostic effects.

Method used

Build a cross-modal bridged graph convolution neural network, generate multi-modal graph structure data through KNN graphs, extract key features using graph convolution, and realize feature fusion through cross-modal bridge mechanism, introduce dynamic collaboration mechanism to improve comprehensive analysis capabilities, and combine multi-channel fusion Transformer module to achieve comprehensive information integration and fault identification.

Benefits of technology

The proprietary features of multiple fault modes are effectively extracted, and the feature complementation is enhanced through the cross-modal bridging mechanism, which improves the reliability and generalization of diagnostic results and achieves more accurate fault diagnosis.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a fault diagnosis method for a cross-modal bridge graph convolutional neural network, and the method specifically comprises the steps: enabling each sample to represent a node for collected original multi-modal data, and constructing an adjacent matrix through the first k fault nodes; extracting proprietary features of a plurality of modals by using graph convolution, and introducing a cross-modal bridging module to complementarily enhance feature information of different modals; in consideration of common fault features hidden in multiple modes, a dynamic collaborative sharing mechanism is introduced to capture similar features; on the basis of the above processing, information of different channels is fused and aggregated by using multiple channels, and then the information is input into Transform to carry out fault classification on fused features. Experimental results show that the method is an effective fault diagnosis method.
Need to check novelty before this filing date? Find Prior Art