The integration of Artificial Intelligence (AI) into Industrial Internet of Things (IIoT) systems has enhanced predictive maintenance strategies by enabling early detection of faults in machinery. However, centralized AI models often face challenges related to data privacy, latency, and communication overhead in industrial environments. This study aims to develop a decentralized AI framework utilizing Federated Learning (FL) on edge devices to enhance predictive maintenance in a medium-scale manufacturing plant. The proposed system enables local edge nodes to collaboratively train machine learning models without sharing raw data, thereby preserving data privacy and reducing network load. A prototype was developed using embedded edge devices integrated with vibration and temperature sensors to detect machine anomalies. Federated averaging was used to aggregate local models into a global model. Experimental results show that the federated model achieved 91.4% accuracy in anomaly detection, comparable to centralized approaches, while significantly reducing data transmission volume by 68%. This research demonstrates the feasibility of deploying federated learning on resource-constrained edge devices for predictive maintenance in IIoT environments. The findings suggest that decentralized AI at the edge can offer efficient, privacy-preserving, and scalable solutions for industrial applications
Supriadi et al. (Fri,) studied this question.