基于深度小波自动编码器和极限学习机的轴承故障诊断

Fault Diagnosis of Bearing Based on Deep Wavelet Automatic Encoder and Extreme Learning Machine

针对原始振动数据无监督特征学习问题,提出了一种深度小波自动编码器(deep wavelet automatic encoder,DWAE)与鲁棒极限学习机(extreme learning machine,ELM)相结合的滚动轴承的智能故障诊断方法。首先,利用小波函数作为非线性激活函数设计小波自动编码器从而有效地捕获信号特征。其次,利用多个小波自动编码器构造一个深度小波自动编码器来增强无监督特征学习能力。最后,采用鲁棒极限学习机作为分类器,对不同的轴承故障进行分类识别。对实验所得的轴承振动信号进行对比分析,结果验证了研究结果能够在原始振动数据无监督特征学习的条件下该方法优于传统方法和标准深度学习方法。

Aiming at unsupervised feature learning of original vibration data,an intelligent fault diagnosis method for rolling bearings was proposed,which combined deep wavelet automatic encoder with robust extreme learning machine.Firstly,the wavelet function was used as a non-linear activation function to design an automatic wavelet encoder to capture the signal features effectively.Secondly,a deep wavelet auto-encoder was constructed by using multiple wavelet auto-encoders to enhance unsupervised feature learning ability.Finally,robust limit learning machine was used as classifier to classify and identify different bearing faults.The experimental results show that the proposed method is superior to the traditional method and the standard deep learning method under the condition of unsupervised feature learning of the original vibration data.