基于度量学习的收割机滚动轴承故障诊断

打开文本图片集

中图分类号：S23；TP23 文献标识码：A 文章编号：2095-5553（2025）10-0112-08

Abstract：Toachievethediagnosisofroling bearing faultsinagriculturalmachinery，reducemaintenancecostsand ensure production eficiency，this study designs amatching network based on the idea of metric learning for bearing fault diagnosisinscenarioswith limitedactual samples．Thematching networkconsistsoftwostructurally identical sub-networks.Thesesub-networksextractand fuse the time-frequency featuresof thebearing vibration signals，and determine thefault type of the sample bycalculating the similarity between the output feature vectorsof the two sub-networks.Theresults show thatthe proposed model canaccurately extract the time-domainand frequency-domain featuresof bearing faults，clearlydistinguish thedistributionof diferent typesoffault samplesinthefeature space，and its performance is the best when the number of learnable samples is the same.The diagnostic accuracy reaches up to 99.33% .Whenthe sample size issmall，theaccuracyrateof the metric learning methodis higher than that of the conventional training method.Comparedwiththemethodof extracting featuresfromasingledomain，themethod that integrates time-frequency featuresperformsbetter.

Keywords：harvester；rolling bearing；fault diagnosis；metric learning；matching network；time-frequency feature：

0 引言

农用收割机的作业效率与轴承的运行状态密切相关，旋转部位的轴承起支撑轴上零件及控制旋转精度的作用。（剩余11976字）