Useful Life Prediction of Bearings in Abrasive Environments through FEM Modeling, IoT Validation, and Visual Training Using Machine Learning

Abstract

The reliability of rolling bearings in mining machinery is essential for maintaining operational continuity in extreme environments such as Cerro de Pasco, Peru, where abrasive particles, humidity, and cyclic loads accelerate mechanical degradation. To address the limitations of conventional maintenance approaches, this study proposes a hybrid methodology for predicting the Remaining Useful Life (RUL) of industrial bearings by integrating Finite Element Modeling (FEM), IoT-based monitoring, and machine learning techniques. The analysis focuses on the SKF 22244 CC/W33 spherical roller bearing, widely used in rotating equipment such as crushers and grinding mills. FEM simulations were performed under realistic loads to identify stress concentration zones, while real-time data from vibration, temperature, and pressure sensors were used to train XGBoost and Random Forest models for RUL estimation. Additionally, a Convolutional Neural Network (CNN) classified wear severity from bearing images with accuracy above 93%. Results show that actual bearing lifespan in abrasive environments may decrease by up to 15% compared with theoretical ISO 281 predictions, highlighting the need for predictive models adapted to real operating conditions. Cross-validation between simulations, sensed data, and intelligent algorithms confirms the robustness of the proposed approach for early fault detection, maintenance optimization, and reliability improvement in mining systems. This methodology also demonstrates scalability to other industrial sectors aligned with Industry 4.0 and Industry 5.0 frameworks.