Alarm management and fault prediction using machine learning applied to intelligent systems embedded in mining equipment.
Prediction; Regression; electrical motors; insulation resistance; machine learning.
The identification of the operational condition of electric motors in industrial plants is essential for ensuring the continuity of the production process. The degradation of insulation resistance in three-phase induction motors (IMs) represents one of the most critical challenges in asset management, as it leads to unplanned downtime, increased corrective maintenance costs, higher energy consumption, and, consequently, a premature reduction in equipment lifespan. This study proposes a methodology to infer the corrected insulation resistance in industrial three-phase induction motors using Machine Learning. Through the construction of a dataset based on real process data from a plant operating six motors, the entire development was carried out, including techniques related to feature engineering, regression analysis, and classification, applying both linear and non-linear models. Preliminary results indicate that non-linear models performed better in estimating insulation resistance; however, classification approaches demonstrated superior performance overall. In both cases, feature selection had a significant impact. The proposed solution proves to be effective in optimizing predictive maintenance and preventing unplanned downtime.