基于决策融合方法和迁移学习的齿轮箱故障诊断

打开文本图片集

中图分类号：TP206 文献标志码：A

Gearbox Fault Diagnosis Based on Decision Fusion Methods and Transfer Learning

LIU Tingting1，2， WANG Zheming 1 ， YU Wenying' ， LU Wu1 ， YU Wei 2 ， LIU Yongsheng 1 （1.Institute of Solar Energy，Shanghai Universityof Electric Power，Shanghai 2O13O6，China; 2.Huadian Electric Power Research Institute，Hangzhou 31OO3O，Zhejiang，China）

Abstract：Toadress the problemsof frequent gearboxfailures inindustrial scenarios，large datarequirements，and low interpretability indeep learning diagnostics，a gearbox fault diagnosis model was proposedbasedondecisionfusionmethods andtransfer learning.Vibration signal features were extracted from CWRU Bearing Dataset using a knowledge-driven approach.Theoptimal feature subsets，including complex envelope spectra，time-domain statistical features，and wavelet packetnalysisfeatures，were then selected using arecursive feature elimination algorithm.Sixclasification models were employed，andthedecision fusion methodsofvoting，stacking，and blending werecombined toestablish thefaultdiagnosis model.The model trainedon CWRU Bearing Dataset was applied to Southeast University Planetary Gearbox Dataset using transfer learning todiagnosegearbox faults.The experimentalresultsshow that the stacking-based ensemble model， integrating sixclassfiers，performs excelentlyonboth the CWRU Bearing Dataset and the gearbox fault diagnosis task， withminimal accuracy differences between thetwo tasks.Theaccuracyrateof the model inthefaultdiagnosis task of gears and bearings in gearbox reaches 100% ，which has better diagnostic ability compared with other fault diagnosis models.

Keywords： gearbox； fault diagnosis； feature extraction； decision fusion； transfer learning

随着工业系统复杂性的增加，机械设备故障诊断技术的发展对保证工业生产和正常生活起着至关重要的作用[1-2]。（剩余13683字）