基于深度神经网络的滚动轴承故障诊断

摘要

Shallow neural network method which rely on expert experience and signal processing technology for artificial extraction the characteristics of big rolling bearing data is becoming more and more difficult, what' s more, shallow structure limits that the neural network learn the function of the complex nonlinear re-lationship. Based on the advantages of feature extraction and big data processing, a new method of rolling bearing fault diagnosis based on deep neural network is studied. The method directly extract the useful char-acteristics from the original data, the extracted features can be as the BP neural network input to identify rolling bearing fault categories. Through the analysis of the four kinds of state of the normal state , the inner race failure, the outer race fault and the ball failure, and the severity of the different faults of the rolling bearings in each state, experimental results show that the research of the method not only can dig out useful fault feature from original signal, also can diagnose the fault severity, compared with BPNN has higher di-agnostic accuracy.%针对浅层神经网络方法依靠专家经验和信号处理技术进行人工提取大量滚动轴承数据的特征变得越来越困难,而且神经网络浅层结构限制了神经网络学习复杂非线性关系的功能.结合深度学习在特征提取和处理大数据等优势,研究一种基于深度神经网络的滚动轴承故障诊断方法.该方法直接从原始数据中提取出有用特征,所提取的特征作为BP神经网络(BPNN)的输入识别滚动轴承的故障类别.通过对滚动轴承正常状态,内圈故障,外圈故障和滚珠故障四种状态以及各个状态的不同故障严重程度的分析,实验结果表明所研究的的方法不仅仅能够从原始信号中挖掘出有用的故障特征,还可以诊断出故障的严重程度,和BPNN相比具有更高的诊断准确率.