LEVERAGING AI FOR PREDICTIVE MAINTENANCE IN INDUSTRIAL IOT: A COMPARATIVE STUDY OF MACHINE LEARNING ALGORITHMS

Abstract

Introduction/Importance of Study: The incorporation of Industrial IoT (IIoT) has dramatically impacted maintenance plans in place, and likely, there are more approaches for decreasing unanticipated outages for maintenance tasks, such as a prominent type called prescriptive maintenance. Previous studies have highlighted the efficiency of using AI-based machine learning algorithms for predictive maintenance. Still, an in-depth analysis of such an approach in the context of an industrial setting has not yet been conducted.

Novelty Statement: This paper offers a comparative evaluation of different machine learning models, such as Random Forest, Support Vector Machines, Neural Networks, and Gradient Boosting, for aims at predictive maintenance in IIoT systems.

Material and Methods: The study used historical data on maintenance for industrial sensors and IoT devices. Several classical and advanced machine learning techniques were deployed and assessed in terms of accuracy, F1-score, precision, recall rates, and computational time and space complexity other than scalability in real-time industrial applications.

Results and Discussion: The comparison showed that there is a rather notable difference in the efficiency of the used machine learning algorithms. Neural Networks provided the most incredible accuracy, and Random Forests provided a good compromise between accuracy and computational time. Accuracy SVMs performed very well but were not very efficient in terms of Space Complexity. Gradient Boosting was highly accurate, but it took a lot of time to implement; hence, it was not feasible for real-time problems.

Concluding Remarks: Remarkably, the results highlight that, although Neural Networks and Gradient Boosting have a higher accuracy, Random Forests may be more effective to be used in the Big Data and real-time industrial environments in terms of efficiency and demand. Something that this study offers to industrial practitioners is how to decide on the best machine learning algorithm for predictive maintenance in IIoT systems.