An integrated activity tracking based anomaly detection framework for mining dumpers in surface coal mines

Mining dumpers operate under highly variable loading and environmental conditions, making reliable condition monitoring challenging in surface coal mines. Conventional vibration-based monitoring approaches generally treat activity tracking and anomaly detection as independent tasks, ignoring the operational context in which vibration signals are generated. Consequently, normal variations associated with different operating activities can obscure early mechanical faults and reduce the reliability of anomaly detection. To address this limitation, this study proposes an integrated activity tracking-based anomaly detection framework for intelligent monitoring of mining dumpers using vibration signals. The proposed framework combines multi-criteria feature selection, operational activity tracking, and unsupervised anomaly detection within a unified pipeline suitable for real-time deployment. Twenty statistical features are evaluated against selection criteria, and discriminative features are used to train lightweight neural networks for activity tracking. Then, tracked idling activity is analyzed using an autoencoder to detect anomalies. The framework achieved 96.20–98.47% F1-scores at 0.2 ms inference time and successfully identified an anomaly as an incipient bent engine pushrod, as confirmed by maintenance records, thereby supporting predictive maintenance.