基于深度自编码网络的刚性罐道故障诊断

为解决提升机刚性罐道故障诊断中故障特征提取困难的问题,结合深度自编码网络的特征提取能力优势,提出了一种基于深度自编码网络的刚性罐道故障诊断方法。以重构误差作为深度自编码网络的评价准则,在各层自编码网络之间采用反向传播的方式,逐层对网络的权值和偏置进行优化。利用得到的最优权值和偏置组成特征提取网络模型,基于该网络模型提取刚性罐道的故障特征。以SVM作为分类器实现刚性罐道的故障分类。实验结果表明,该方法提取的故障特征可识别性较好,识别率较高。

Rigid Cage Guide Diagnosis based on Deep Auto- encoder Network

In order to solve the problem of difficulty for fault diagnosis of rigid cage guide of the hoist in fault feature extraction. Combining the advantages of feature extraction, a new fault diagnosis method of rigid cage guide was proposed based on deep auto-encoder. The reconstruction error was used as evaluation criterion for deep auto-encoder network. The weight and offset of auto-encoder network was optimized layer by layer with back propagation. The network model for feature extraction was constructed by optimal weight and offset. Then fault features of rigid cage guide were extracted based on this network. The rigid cage guide fault classification is realized by SVM used as classifier. The experimental results show that the fault feature extracted by the method is recognizable, and hasShigher recognition rate.