A predictive and explanatory model for remaining useful life of crushers using deep learning

Current maintenance models lack the technological capabilities to generate key performance indicators that optimize both critical equipment behavior and the surrounding processes. Artificial intelligence offers powerful tools for predicting and interpreting sensor data collected from such equipment, enabling continuous improvement. This paper proposes a tool that leverages deep learning to predict the remaining useful life (RUL) of a large-scale mining crusher. Additionally, the model incorporates result interpretation algorithms to analyze both training cycles and subsequent production cycles. This analysis not only identifies a process "fingerprint" but also recommends adjustments to the crusher system within the ongoing maintenance plan. By employing a dense neural network and interpretation algorithms, the proposed tool predicts the current crusher cycle’s RUL and compares its interpretation graphs to the process fingerprint. This comparison identifies discrepancies, which in turn inform maintenance recommendations tailored to specific crusher components.