基于小波卷积自编码器和LSTM网络的轴承故障诊断研究

Fault diagnosis of bearing based on wavelet convolutional auto-encoder and LSTM network

针对传统滚动轴承故障诊断方法过度依赖专家经验和故障特征提取困难的问题,结合深层神经网络处理高维、非线性数据的优势,提出了一种基于深层小波卷积自编码器(DWCAE)和长短时记忆网络(LSTM)的轴承故障诊断方法。首先构造了小波卷积自编码器(WCAE),改进了其损失函数,并加入了收缩项限制防止网络过拟合;其次将多个WCAE堆叠构成DWCAE,利用大量无标签样本对DWCAE进行了无监督预训练,挖掘出更有利于故障诊断的深层特征;最后利用深层特征训练LSTM网络,从而建立了诊断模型。仿真信号和实验数据分析结果表明:该方法能有效地对轴承进行多种故障类型和多种故障程度的识别,特征提取能力和识别能力优于人工神经网络、支持向量机等传统方法及深度信念网络、深层自编码器等深度学习方法。

Aiming at the problems that traditional fault diagnosis algorithms of rolling bearings have such shortcomings as largely dependent on expert prior knowledge and difficulty in fault feature extraction, combined with the merits of deep learning in dealing with high-dimensional and nonlinear data, a method based on deep wavelet convolutional auto-encoder (DWCAE) and long short term memory ( LSTM) network was proposed. Firstly, wavelet convolutional auto-encoder (WCAE) was designed, and improved loss function and contraction term restriction were introduced to alleviate the over-fitting of the network. Secondly, several WCAEs were stacked to construct DWCAE. A large number of unlabeled samples were used for unsupervised pre-training of DWCAE, and the deeper features that were more favorable to fault diagnosis were mined. Finally, LSTM network was trained with deeper features, and the diagnosis model was established. The results of simulation signal and engineering application analysis indicate that the proposed method can effectively identify the beaiing faults under multiple working conditions and multiple fault severities. The proposed method has better aWlity of feature extraction and recognition than traditional methods such as artificial neural network,support vector machine and deep learning methods such as deep belief network,deep auto-encoder and so on.