Publication Details
Abstract
This study presents the design of an artificial intelligence-based fault forecasting system that will be used to improve predictive maintenance in three important industrial sectors, which include the following; electrical systems, cooling systems, and elevator mechanisms. With the application of real datasets equipped with sensors, the protocols consisted of machine learning models with configurations towards multiclass, binary and multi-label classification. A lot of pre-processing that included noise control, class balancing, and manual sampling in enhancing data quality and model robustness was done. The results show that the current and vibration signals feature considerable domain-specific characteristics and greatly contribute to better fault detection performance with the ROC-AUC measuring the effectiveness of the model yielding an adequate score and the confusion matrix also confirming the model accuracy. Results visualization and features significance, including the drawback of labels, proved the implement ability and explain ability of the models. The system is distinguished as the only one that provides a unified, empirically validated solution that has the potential to fault detect across domains thus serving as a prelude to scalable, real-time implementation in smart maintenance ecosystems.