A motor bearing diagnosis model training method, device and equipment and storage medium

By enhancing the signal using a microphone array and delayed summation beamforming, and combining Mel spectrum and image temporal feature extraction, the problem of insufficient generalization ability of the motor bearing diagnostic model in complex environments is solved, and high-precision diagnosis under strong interference conditions is achieved.

CN122087653APending Publication Date: 2026-05-26HUBEI ENERGY GRP EZHOU POWER GENERATION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN ยท China
Patent Type
Applications(China)
Current Assignee / Owner
HUBEI ENERGY GRP EZHOU POWER GENERATION CO LTD
Filing Date
2026-02-12
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing technologies lack the generalization ability of motor bearing diagnostic models under complex interference environments, making it difficult to achieve accurate diagnosis.

Method used

A microphone array and delayed summation beamforming method are used to perform directional signal enhancement on noisy signals. Mel spectrum feature extraction and image temporal dual-channel feature extraction are combined. Fault classification is performed using CNN and Transformer models. The model is iteratively optimized using cross-entropy loss and Adam optimizer.

Benefits of technology

This improved the model's adaptability and generalization ability in complex environments, enhanced the weight of the target device signals, and improved the accuracy and robustness of the diagnosis.

โœฆ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122087653A_ABST
    Figure CN122087653A_ABST
Patent Text Reader

Abstract

A method, apparatus, device, and storage medium for training a motor bearing diagnostic model include: acquiring noisy signal samples based on a microphone array; performing directional signal enhancement on the noisy signal samples in the direction of the target device using delayed summation beamforming to obtain enhanced signal samples; performing Mel-spectrum signal conversion on the enhanced signal samples to obtain Mel-spectrum features; extracting bearing fault features from the Mel-spectrum features using image temporal dual-channel features; performing fault classification prediction on the bearing fault features to obtain predicted diagnostic results; determining the prediction loss based on the diagnostic prediction results and the actual diagnostic results; and iteratively optimizing the initial motor bearing diagnostic model based on the prediction loss to obtain a fully trained motor bearing diagnostic model. This application uses delayed summation beamforming to perform directional enhancement on the samples, reducing and preserving noise interference, improving the model's adaptability to environmental noise, and meeting the diagnostic needs of production environments.
Need to check novelty before this filing date? Find Prior Art