Idea of Motor Fault Diagnosis and Classification Using the Latent Vector of Autoencoder

This paper proposes a diagnosis and classification method for various motor faults using vibration signals. We suggest a diagnosis and classification method of motor conditions using a stacked autoencoder (SAE) with a PCA (principle component analysis) to solve this problem. The PCA is used for extracting important components of the given vibration data and the autoencoder is used for fault diagnosis. In the first stage, we used the Fourier transform and PCA to preprocess and extract the key components from the vibration signal in the frequency domain. The autoencoder was trained with the results of PCA. Each fault could be distinguished by visualizing the latent vector. Experimental results demonstrate that the proposed method works well though there are some areas to be supplemented. The proposed simple method can be applied to industrial applications.