The document summarizes a proposed methodology for bearing fault detection using a convolutional autoencoder based feature extraction approach. Key points:
- It proposes using a convolutional autoencoder to extract features from vibration signal data that can better capture the patterns and time-series characteristics compared to traditional statistical approaches.
- The extracted latent variables from the autoencoder aim to concisely represent the essential characteristics of the input signals.
- A Hotelling T2 control chart would then be used to monitor the latent variables and detect abnormalities, with bootstrap resampling used to determine non-parametric control limits.
- The overall goal is to develop an improved bearing fault detection model for condition-based maintenance by leveraging deep learning feature