• PROF. SUYOG SANGHARATNA DHOKE
• MR. MOHAMMED SHAADURRAHIM SHEIKH

Modern industrial operations rely heavily on electrical equipment that must run reliably under various operating conditions. Sudden failures can lead to production loss, safety issues, and high maintenance costs. While predictive maintenance techniques have improved the ability to forecast failures, they often do not go far enough in helping technicians decide what actions should be taken in real time. In many cases, maintenance decisions are either delayed or overly cautious, resulting in avoidable costs and downtime. In this work, we present a practical framework that combines federated learning, transformer-based models, and deep reinforcement learning (DRL) to enable explainable prescriptive maintenance for electrical equipment. The system is designed to operate in distributed environments where sensor data from machines cannot be centrally shared due to privacy or regulatory concerns. Using a transformer architecture trained via federated learning, each machine estimates its own Remaining Useful Life (RUL) without exposing raw data. These local models contribute to a shared global understanding of equipment health while preserving data ownership. Building on these predictions, a DRL agent is trained to recommend optimal maintenance actions—such as inspection, repair, or continued operation—based on real-time equipment states and predicted lifetimes. To ensure the model's decisions are understandable and trustworthy to engineers and plant operators, we integrate an explain ability layer. Attention weights from the transformer and SHAP-based interpretations of the DRL output help highlight which sensor features or conditions influenced a specific recommendation. We validate our approach using publicly available industrial equipment datasets, simulating real-world conditions. Results show improvements in prediction accuracy, maintenance decision quality, and overall system transparency. Importantly, the use of federated learning ensures that sensitive operational data remains secure, making this solution practical for deployment in industries where data sharing is limited. Our framework brings together privacy, action ability, and explain ability in a single, efficient pipeline for industrial maintenance.

Federated Learning, Transformer Model, Deep Reinforcement Learning, Explainable AI, Prescriptive Maintenance, Remaining Useful Life, Electrical Equipment, Data Privacy, Fault Detection, Industry.

""Federated Transformer DRL Framework for Explainable Prescriptive Maintenance of Electrical Equipment"", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.12, Issue 7, page no.b603-b610, July-2025, Available :http://www.jetir.org/papers/JETIR2507169.pdf

""Federated Transformer DRL Framework for Explainable Prescriptive Maintenance of Electrical Equipment"", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.12, Issue 7, page no. ppb603-b610, July-2025, Available at : http://www.jetir.org/papers/JETIR2507169.pdf