模糊投影孪生宽度学习及其在滚动轴承故障诊断中的应用  

Fuzzy projection twin broad learning and its application to rolling bearing fault diagnosis

作　　者：郑凤雷 潘海洋[1] 郑近德[1] 童靳于[1] 程健 ZHENG Fenglei;PAN Haiyang;ZHENG Jinde;TONG Jinyu;CHENG Jian(School of Mechanical Engineering,Anhui University of Technology,Ma’anshan 243032,China)

机构地区：[1]安徽工业大学机械工程学院,安徽马鞍山243032

出　　处：《振动与冲击》2025年第8期192-198,共7页Journal of Vibration and Shock

基　　金：国家自然科学基金(51975004);安徽省高校杰出青年项目(2022AH020032);安徽省自然科学基金(2408085ME113)。

摘　　要：由于通过传感器获得的振动信号往往包含大量的高噪声信息和冗余信息,其严重影响滚动轴承故障诊断的准确性。基于此,提出了一种基于模糊投影孪生宽度学习(fuzzy projection twin broad learning,FPTBL)的滚动轴承故障诊断方法。在FPTBL建模过程中,首先会在特征空间中为每一类数据找到一个投影方向,并利用模糊隶属度函数计算隶属度,评估每个样本属于各类别的可能性,并以此为约束调整投影方向让同类样本相互靠近,异类样本分散开来,以解决模型对噪声信息敏感性的问题。同时,FPTBL不仅关注经验风险最小化,还关注结构风险最小化,为防止模型过拟合,通过构造非平行的投影空间并将正则项纳入目标函数,有效地弱化冗余信息对模型的影响。最后,利用两种滚动轴承数据集进行试验验证,结果表明,FPTBL在准确率、Kappa系数、F_(1)-分数、精确率和召回率等指标下,相较于现有的方法,具有更好的效果。Vibration signals obtained from sensors often contain substantial noise and redundant information,which can significantly impact the accuracy of rolling bearing fault diagnosis.To address this challenge,a rolling bearing fault diagnosis method was proposed in this paper based on fuzzy projection twin broad learning(FPTBL).In the FPTBL modeling process,a projection direction was firstly found for each type of data in the feature space,and a fuzzy membership function was used to calculate the membership degree.To address the sensitivity of model to noise information,the likelihood of each sample belonging to each category was evaluated,and the projection direction was adjusted to make similar samples closer to each other and disperse dissimilar samples.Meanwhile,FPTBL not only focuses on minimizing empirical risk,but also on minimizing structural risk.To prevent model overfitting,non-parallel projection spaces were constructed and regularization terms were incorporated into the objective function to effectively weaken the impact of redundant information on the model.Experimental validation on two rolling bearing datasets demonstrates that FPTBL achieves superior performance in terms of accuracy,Kappa coefficient,F_(1)-score,precision and recall rate.

关 键 词：模糊投影孪生宽度学习(FPTBL) 投影孪生 故障诊断 滚动轴承 

分 类 号：TH165.3[机械工程—机械制造及自动化]