A Survey of AI Techniques based on Predictive Maintenance in Lean Manufacturing

Predictive maintenance plays a pivotal role in lean manufacturing by helping manufacturers identify and resolve maintenance issues, thereby avoiding expensive downtime or equipment failures. Diverse algorithms have been more prevalent in artificial intelligence and machine learning in recent years, significantly impacting various kinds of industries. This paper surveys the current landscape, focusing on the popularity and effectiveness of different algorithms. Our methodological approach includes a systematic literature review, utilizing a variety of keywords and their synonyms, mainly the Google Scholar Database, for pertinent information. Three tiers of inclusion criteria are further subdivided into the data processing, serving as filters to improve the search results. The information is analyzed, comprising data reduction, descriptive bibliometric analysis, and journal identification that is influential and productive. Our findings highlight that the most widely used algorithm for predictive maintenance in lean manufacturing is a decision tree, which is valuable for both classification and regression tasks across multiple industries. Additionally, convolutional neural networks (CNNs) are noted for their efficacy in pattern recognition within sensor data, aiding in anomaly detection and maintenance forecasting. Another prominent model is artificial neural networks (ANNs), known for their courage in coping with complex problems beyond the scope of conventional methods. However, the popularity of these models raises important questions about their limitations, including overfitting risks and interpretability issues. Examining these details indicates that while certain algorithms are obviously more accurate than others, they can also be perplexing. Therefore, it is suggested that future efforts shift to hybrid models in an effort to achieve a compromise between algorithmic clarity and robust performance. The study concludes by emphasizing the importance of domain-specific evaluations and iterative refinement in models, providing a thorough lens through which experts in machine learning, professionals, and researchers can assess the constantly changing field of artificial intelligence.