Artificial Intelligence-Supported Predictive Maintenance in Industrial Systems: Methods Applications and Limitations

Abstract

Maintaining equipment reliability, reducing unplanned downtime, and controlling maintenance costs are essential for production continuity in industrial systems. Predictive maintenance uses equipment condition data to identify maintenance requirements before failures occur. Advances in sensor technologies, the Internet of Things, and industrial data acquisition systems have enabled continuous monitoring and the application of machine learning and deep learning methods. This review examines artificial intelligence-supported predictive maintenance in terms of methods, industrial applications, and limitations. Support vector machines, Random Forest, decision trees, k-nearest neighbors, XGBoost, and artificial neural networks are widely used for fault detection, condition classification, and remaining useful life estimation. CNN and LSTM models are prominent in analyzing high-dimensional and time-dependent sensor data. Applications in manufacturing, automotive, mining, and facility management have reported favorable outcomes. Major limitations include data quality, insufficient labeled failure data, explainability, computational requirements, model generalizability, and system integration. Long-term industrial validation and adaptable models remain research priorities.