自适应小波分析和多层卷积极限学习自编码器的轴承故障识别研究

Fault Identification of Rolling Bearing Based on Adaptive Wavelet Analysis and Multiple Layers Convolution Extreme Learning Auto-encoder

针对滚动轴承振动信号由于强时变和强噪声等特性导致其故障难以辨识的问题,提出一种基于自适应小波分析(AWA)和多层卷积极限学习自编码器(MLCELAE)的滚动轴承故障识别模型。首先,提出一种新的轴承振动信号频谱边界检测方法,对信号频谱进行自适应分割,进而将信号分解为若干本征模态分量;然后选择较能反映轴承故障特征的模态分量并重构;最后构造卷积极限学习自编码器,并逐层堆叠建立深层网络MLCELAE,将信号样本输入MLCELAE进行自动特征学习与故障识别。试验结果表明:提出方法的平均故障识别准确率达到了98.48%,标准差仅为0.17,相比于其他方法在轴承故障识别准确率方面更具优势,适用于滚动轴承故障的自动识别。

Aiming at the problems of rolling bearing vibration signals were difficult to identify due to strong time-varying and strong noisy characteristics,a method based on adaptive wavelet analysis(AWA)and multiple layers convolution extreme learning auto-encoder(MLCELAE)was proposed.Firstly,a new method was proposed to detect the vibration signals spectrum boundary of rolling bearing,which could divide the signals spectrum adaptively,and then decompose the vibration signals into several intrinsic modal components.Secondly,the components which could best reflect the fault characteristics of the vibration signals were screened and reconstructed.Finally,the convolution extreme learning auto-encoder was constructed and multiple layers convolution extreme learning auto-encoder was built by stacking layer by layer,then the vibration signals samples of rolling bearing were fed into MLCELAE for automatic feature learning and fault identification.The experimental results show that the average fault identification accuracy of the proposed method reaches 98.48%and the standard deviation is only 0.17.Compared with other methods,it has more advantages in fault identification accuracy of rolling bearing,which is suitable for automatic identification of rolling bearing faults.