Machine Learning Models Application for Maintenance Operations Enhancement in a Petrochemical Industry

Abstract

This study investigates the use of Machine Learning (ML) techniques in predictive maintenance for industrial five-stage compressors, emphasizing cost efficiency, model performance, and real-world applicability. Four ML models—Random Forest, Decision Trees, Logistic Regression, and Gradient Boosting were assessed using actual data. The Random Forest model achieved the highest accuracy at 94%, with Decision Trees closely behind. While Logistic Regression was computationally efficient, it falls short in predictive accuracy. Cross-validation and hyper-parameter tuning reinforced the Random Forest model’s strong generalization. A cost analysis revealed notable financial gains from adopting ML-based predictive maintenance, with reduced downtime and optimized maintenance schedules offsetting initial setup costs. The study also discussed the integration of ML within the broader Industry 4.0 landscape, highlighting its potential to improve equipment reliability, lower operational costs, and also enabling intelligent data-driven maintenance practices. This research offers a robust framework for industries that wishes to shift from traditional maintenance to ML-driven methods, which advances operational efficiency and sustainability in industrial settings.