基于多模态堆叠自动编码器的感应电机故障诊断

Induction motor fault diagnosis based on multimodal stacked auto-encoder

针对感应电机多源监测数据利用率不高,难以有效融合多传感器信息进行电机故障的准确识别等问题,提出了一种多模态堆叠自动编码器模型(MSAE)。该模型直接从原始信号中获取其最为显著的特征向量,有效减少了手动提取特征指标造成的故障信息遗漏,并能学习到多源信号的共享表示实现多源融合的故障诊断,为融合多传感器信息的设备故障诊断提供了新思路。实验证明,与使用单一传感器信息的堆叠自动编码器模型、具有同样隐藏层结构的多层感知机以及使用手动提取特征的支持向量机相比,提出模型具有最高的诊断准确率(94.84%),并在振动信号被噪声损坏的情况下展现了良好的适应性。因此该方法可用于多传感器融合的感应电机故障诊断。

In view of the low utilization of multi-sensor monitoring data collected from induction motor and the difficulties of effectively fusing multi-sensor information to accurately identify motor faults,a multimodal stacked auto-encoder( MSAE) is proposed in this paper.The proposed model directly obtains the most expressive feature vector from the original signal,which could effectively reduce the omission of the fault information caused by manual feature extraction. Then the proposed model could learn the sharing representation from multi-sensor signals thus fusing the multi-sensor data,which provides a new idea for machine fault diagnosis according to multisensor information. Comparing with stacked auto-encoder models which take single sensor information as input,multi-layer perceptron with the same hidden layer structure and support vector machine using manual extracted features,experimental results show that the proposed model achieves the highest diagnostic accuracy( 94. 84%) and exhibits favorable adaptability when the vibration signal is corrupted with noise. It is proved that the proposed method can be used for fault diagnosis of induction motor with multi-sensor information.