基于加权概率神经网络的齿轮箱抗噪故障诊断  

Noise robustness research in gearbox fault diagnosis based on weighted probabilistic neural network

作　　者：崔逊波[1] 邹俊[1] 阮晓东[1] 傅新[1] 

机构地区：[1]浙江大学流体传动及控制国家重点实验室,浙江杭州310027

出　　处：《机电工程》2010年第2期54-56,82,共4页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(50705082)

摘　　要：针对齿轮箱现场故障诊断易受噪声干扰、诊断精度低的问题,提出了一种基于区分性权重概率神经网络的故障诊断方法。该方法考虑了不同子带特征受噪声的污染程度不同,提高噪声影响小的特征在诊断中的权重,降低噪声影响大的特征在诊断中的权重,以提高诊断的噪声鲁棒性,最终实现了齿轮箱故障的诊断。试验研究结果表明,与BP神经网络和概率神经网络诊断相比,该方法具有较高的诊断正确率和较强的诊断鲁棒性;并且该方法中平滑度参数对故障诊断精度影响不大,可以避免该参数选择困难的问题,具有良好的工程应用前景。In order to solve the problem that the fault sigal of gearbox is difficult to detect and the diagnosis is easily disturbed by noise,a fault diagnosis method based on weighted probabilistic neural network（WPNN） was presented.The different levels of noise pollution on different characteristics were taken in to consideraion,the noise robustness was improved,eventually the gearbox fault diagnosis was realized.Comparing with BP neural network（BPNN） and probabilistic neural network（PNN） in the experiments.The method has higher diagnostic accuracy and noise robustness,meanwhile,it can reduce the difficulty of choosing correct smothing parameter,thus has a good industrial application.

关 键 词：故障诊断 加权概率神经网络 抗噪 齿轮箱 

分 类 号：TH137[机械工程—机械制造及自动化]