In the newer vision of Industry 5.0, the industrial transport industry is evolving towards paradigms that prioritise the adoption of intelligent predictive maintenance systems. These systems enable the prediction of assets at risk of imminent failure, thus allowing maintenance to be planned before failure. This study utilises a real-world dataset provided by Scania to predict the imminent failure of an anonymised engine component in heavy-duty trucks. The dataset presents common real-world challenges, including anonymised values, class imbalance, and missing values. A multi-stage feature selection and preprocessing framework is proposed to address these challenges. A comparative analysis of several machine learning models is conducted to evaluate predictive performance. The findings indicate that the framework improves model robustness and transparency. Moreover, by enabling more accurate failure prediction, the approach contributes to a tangible reduction in maintenance and/or replacement costs associated with the monitored component.
Ferrisi et al. (Thu,) studied this question.