不均衡下分類器評(píng)價(jià)輔助GAN的軸承故障診斷方法

打開(kāi)文本圖片集

中圖分類號(hào)：TB9;TH165.3 文獻(xiàn)標(biāo)志碼：A文章編號(hào)：1674-5124（2025）06-0170-09

Bearing fault diagnosis method based on classifier evaluation-assisted GAN under imbalanced samples

ZHANG Yuehong'，YUAN Zhaocheng1，HUANG Fengfei2， ZHANG Kai23， ZHENG Qing2，3

（1.Chengdu Institute of SpecialEquipment Inspectionand Testing， Chengdu 610299，China; 2.School ofMechanical

Engineering，Southwest Jiaotong University，Chengdu 610o31，China;3.TechnologyandEquipmentofRailTransit Operation and Maintenance Key Laboratory of Sichuan Province， Southwest Jiaotong University，

Chengdu 610031， China）

Abstract： The growth of big dataand IoT technology makes rolling bearing data monitoring possble，but most of the datacollcted is information about the normal status.There is less information accessible for different kinds of faults.The ensuing imbalanced normal and fault samples will impact the accuracy of rolling bearing defect identification. A classifier evaluation-assisted generative adversarial networks （CEAGAN） method for roling bearing failure identification under imbalance is proposed to solve this issue. First，the processuses the short-time Fourier transform to extract the time-frequency characteristics of one-dimensional signals. Second， it builds an auxiliary clasification module in the generative adversarial network to constrain the class of the generated samples.Third， the auxiliary classifier and the discriminator jointly score the generated samples to ensure that the generator produces the desired type of samples.Finally，the generated are mixed with the original imbalanced samples to form a new balanced dataset，and the efectiveness of the proposed method is verified by training and testing the constructed convolutional neural network. The experimental results show that， under the Case Western Reserve University bearing public dataset and the real data of wind turbine gearboxes，the proposed method improves the diagnostic accuracy of the imbalanced case by 15.20% and 13.93% ， respectively， which proves that CEAGAN can effectively improve the fault diagnostic accuracy of the imbalanced samples after the data augmentation.

Keywords： roling bearing; fault diagnosis; imbalanced samples; generative adversarial networks

0 引言

滾動(dòng)軸承作為機(jī)械設(shè)備中的關(guān)鍵零部件，通常用于支撐旋轉(zhuǎn)軸與降低摩擦。（剩余10284字）