稀疏自编码深度神经网络及其在滚动轴承故障诊断中的应用
Application of Deep Neural Network with Sparse Auto-encoder in Rolling Bearing Fault Diagnosis

针对目前滚动轴承故障诊断主要采用监督式学习提取故障特征的现状，提出了一种基于稀疏自编码的深度神经网络，实现非监督学习自动提取滚动轴承振动信号的内在特征用于滚动轴承故障诊断。首先，将轴承故障振动信号的频谱训练稀疏自编码获得参数；然后用稀疏自编码获得的参数和轴承振动信号频谱的频谱训练深度神经网络，并结合反向传播算法对深度神经网络进行整体微调提高分类准确度；最后用训练好的深度神经网络来识别滚动轴承故障。对正常轴承、外圈点蚀故障、内圈点蚀故障和滚动体裂纹故障振动信号的分析结果表明：相比反向传播神经网络，提出的深度神经网络更能准确的识别滚动轴承故障类型。
To overcome the problem of using supervised learning to extract fault features for most current rolling bearing fault diagnosis methods, a deep neural network algorithm is proposed, which is realized sparse auto-encoder, to achieve unsupervised feature learning by automatic extracting the inherent characteristics of the rolling bearing vibration signal for fault diagnosis of rolling bearing fault diagnosis. Firstly, the spectrum of the bearing vibration signal is used to train sparse auto-encoder in order to obtain parameters; secondly, the parameters from sparse auto-encoder and spectrum of the rolling bearing vibration signal are used to train the deep neural network, and the back-propagation algorithm is used for fine-tuning the deep neural network with the purpose of improving classification accuracy. Finally, the deep neural network has been trained to identify faults of rolling bearings. The analysis results from vibration signals with roller normal condition of the rolling bearing,pitting fault of bearing outer ring, pitting fault of bearing inner ring and crack fault of bearing rolling element show that, compared with back propagation neural network, the proposed deep neural network can accurately identify fault type of rolling bearing faults