Integration of Deep Learning Models and Explanatory Models to Detect Anomalous Periods in Vital Signs of Mining Truck Engines

Lima I.K.H.; Gladys Cutipa Arapa E. · Proceedings of the 2024 IEEE 31st International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2024 (2024) · DOI: 10.1109/intercon63140.2024.10833458

Resumen

'Latex The Peruvian mining industry is beginning to adopt IoT technologies. The increasing complexity of modern engines makes it crucial to implement predictive maintenance systems to ensure reliability and performance optimization. This study is based on data obtained from an open-pit copper mine in southern Peru. The use of advanced machine learning models is proposed, evaluating between Long Short-Term Memory with Autoencoders (LSTM-AE) and Artificial Neural Network (ANN) for anomaly detection in multivariate time series for diesel engine health monitoring. This project addresses three issues of the current analysis methodology: first, point anomalies are used to monitor potential failures, generating a high rate of false alarms, while the proposed method identifies abnormal time periods; second, the threshold for alerts is manually calculated and requires expert intervention periodically, whereas the proposed methodology calculates the threshold in an unsupervised manner. Third, deep learning models are often considered 'black box' models since it is not easy to interpret how the model reached a conclusion, which is why model explainability is important for its adoption within the industry. The objective is to develop a predictive model that identifies potential failures early, allowing for the planning of preventive maintenance and reducing unplanned downtime. The results show that the combination of these models significantly improves the accuracy in failure prediction, facilitating timely and precise interventions.

Instituciones: Universidad Nacional de San Antonio Abad del Cusco