Enhancing Industrial Anomaly Detection with Auto Encoder-Based Temporal Convolutional Networks for Motor Fault Classification

The creation of binary and multi-classification models with the goal of accurately detecting and categorizing motor defects is important to study. This work explores how autoencoders can be used to apply self-supervised learning in industrial settings. A generic temporal convolutional network with an autoencoder was built and trained using the milling data. The successful identification and classification of motor faults underscore the industrial anomaly detection using advanced machine learning techniques. Also, paper highlights the importance of data processing and domain expertise in achieving significant model performance improvements. The performance used a fraction of the data that was available, highlighting the need of future research to include a more extensive range of fault states and the intensities of those fault states in order to enhance the diagnostic capabilities of the model.