Electric Motor Health Monitoring Using a Deep Variational Autoencoder (D-VAE)

Abstract The integration of the Industrial Internet of Things (IoT) and machine learning techniques in condition-based maintenance (CBM) systems represents a significant advancement in the monitoring and maintenance of electromotors, particularly those used in elevator systems. This study explores the effectiveness of a Condition Based Maintenance (CBM) framework utilizing Arduino microcontrollers and Microelectromechanical Systems (MEMS) sensors for real-time vibration data collection, coupled with the analysis through a Deep Variational Auto-Encoder (D-VAE) model for anomaly detection in electromotor operation particularly in the elevator system. Traditional methods such as Isolation Forest and K-means clustering, enhanced by Principal Component Analysis (PCA), are also implemented for comparative analysis. The performance of these models is evaluated using unsupervised learning metrics, including the silhouette score, Davies-Bouldin Index, and Calinski-Harabasz score, to assess their efficacy in detecting anomalies within vibration data indicative of potential electromotor malfunctions. Our findings suggest the superior capability of the D-VAE model in discerning between normal and faulty operational states through detailed clustering, despite some limitations under problematic conditions. This study underscores the potential of data-driven approaches and IoT technologies in enhancing operational efficiency, reliability, and safety in elevator systems through predictive maintenance strategies.