Supervised autoencoder for fault detection and diagnosis in predictive maintenance of bearing ring grinding machine

Modern manufacturing industry requires advanced maintenance strategies to ensure process reliability and efficiency. This study develops a deep learning-based algorithm using an autoencoder for fault detection and diagnosis in a bearing ring grinding machine. The proposed methodology follows a two-step classification framework: a binary classifier detects the presence of faults, while a multi-class classifier identifies the specific failure type. The dataset, sourced from the Swedish National Data Service, is analyzed using deep learning techniques to optimize predictive maintenance. Experimental results demonstrate high classification accuracy, enabling real-time monitoring and early identification of machine degradation. This approach enhances maintenance planning, reduces unexpected failures, and improves overall operational performance. Future research will focus on refining the estimation of Remaining Useful Life (RUL) to further optimize maintenance strategies.