A maintenance plan based on Monitoring to detect equipment failures in a mining company using machine learning

The project aims to design a condition monitoring plan for equipment in the water resources division of a private mining company in Chile, located in the Tarapacá and Antofagasta regions. This division plays a critical role in the extraction of water for production processes related to potassium nitrate, specialized fertilizers, iodine, potassium, and lithium. The initiative was triggered by previous failures in the extraction system, leading to partial production interruptions and high maintenance costs. The proposed approach includes a software architecture that supports the development of predictive models using pre-trained algorithms tailored to the operational context. Condition-based maintenance is implemented through techniques such as vibration analysis and infrared thermography, applied to submersible and centrifugal pumps, low and medium voltage electrical panels, substations, reservoirs and more than 150 km of pipelines. A hybrid model combining LSTM architectures and XGBoost decision trees was implemented to enhance the accuracy of failure detection and characterization. Offline validation of the system yielded positive results eight anomalies were detected, five electrical and three mechanical, four of which were classified as critical, demonstrating the effectiveness of the proposed solution.