AI-Based Predictive Maintenance for Smart Manufacturing Systems: A Comparative Machine Learning Approach to Equipment Failure Prediction

Abstract

Few problems in modern plants run up higher bills than unexpected downtime, which explains why predictive maintenance built on artificial intelligence (AI) has been getting so much attention. Using operational sensor readings gathered from factory machines, this study set up a quantitative comparison among several AI-driven predictive maintenance models. To forecast how likely a machine was to break down, the models drew on five predictors: vibration, temperature, pressure, total run time, and mechanical load. We trained three algorithms—Random Forest, Support Vector Machine, and Extreme Gradient Boosting (XGBoost)—and scored them on a held-out test split. We gauged model performance through accuracy, precision, recall, and F1-score, adding Root Mean Square Error where failure probabilities were estimated. XGBoost came out on top across every single metric, Random Forest trailed by only a small margin, and the Support Vector Machine suffered from distinctly lower recall even though its precision held up well. When we ran feature-importance checks on the XGBoost model, vibration amplitude and load stood out as the main drivers of breakdown, with run hours, temperature, and pressure playing lesser roles. What these results demonstrate is that feeding multivariate sensor inputs into ensemble learners produces dependable, practical failure warnings, giving plants a way off reactive repairs and fixed schedules and onto condition-based, predictive footing. The paper finishes by outlining what these results mean for plant managers, maintenance teams, and facilities working through Industry 4.0 upgrades, before pointing out avenues for follow-up work.