MATHEMATICAL MODELLING AND ARTIFICIAL INTELLIGENCE TECHNIQUES FOR PREDICTIVE MAINTENANCE IN MECHANICAL SYSTEMS

Abstract

Predictive maintenance is an important aspect of smart system management, motivated by the need for reliable predictions and optimised resource allocation. While artificial intelligence techniques are rapidly evolving, many of these methods do not incorporate mathematically-based models, creating a gap in interpretability and theoretical foundations. This research is an attempt to bridge this gap by proposing a hybrid system that integrates mathematical models and artificial intelligence for predictive maintenance in mechanical systems. The research uses the AI4I 2020 dataset, which includes structured operational features like temperature, rotation speed, torque and tool wear, to develop physically meaningful features and build a mathematical model of failure-probability. This model is compared with machine learning algorithms such as logistic regression, support vector machines and ensemble techniques. The findings show that features derived from mathematical models improve prediction accuracy and explainability of machine learning models. Moreover, the hybrid methods outperform methods in isolation, showing the potential benefits of integrating mechanistic and data-driven approaches. The results also highlight the need to incorporate domain expertise in predictive models and offer a scalable solution for mechanically governed systems. This research advances predictive maintenance approaches by bridging mathematical and computational efficiency, providing theoretical and practical value.