Journal of Scientific Engineering Advances

Abstract

This paper discusses a framework for an AI-driven predictive maintenance system to detect belt misalignments in conveyor belt systems in the food and beverage industry. Conveyor belts in the FMCG sector are critical for efficient production, but they often fail due to belt drift and misalignment. This can lead to downtime and safety risks. This approach uses sensor data, preprocessing, and AI modelling to predict misalignments. A Kaggle dataset was analysed to study parameters such as vibrations, motor current, load, and belt tension. A random forest classifier was used as a proof of concept, predicting misalignment events with 88% accuracy, 85% precision, and 86% recall. The framework has five main layers; data collection, data preprocessing, AI model interface, performance evaluation, and decision support. So even without using real-time industrial data, the results show that AI- based predictive maintenance is feasible in resource-limited settings like Papua New Guinea. This framework offers a practical basis for local industries to reduce unplanned downtime, improve overall production efficiency and take the first step to Industry 4.0.