A Federated Learning Framework for Predictive Maintenance in IoT-Enabled Smart Environments

Predictive maintenance is a critical component in the management of smart environments, aiming to reduce unplanned downtimes and enhance operational efficiency. Conventional centralized machine learning approaches often pose challenges regarding data privacy, latency, and scalability, particularly in IoT-enabled systems with distributed sensor networks. In response to these limitations, this paper proposes a novel federated learning framework designed for predictive maintenance tasks in smart environments. The framework allows decentralized model training directly on edge devices without transmitting raw data, thereby preserving user privacy and minimizing communication overhead. The system integrates an LSTM-based autoencoder for time-series anomaly detection, tailored for resource-constrained IoT nodes. Additionally, the model is optimized using TinyML techniques for real-time inference at the edge. Experiments are conducted on benchmark datasets relevant to industrial equipment and smart buildings, demonstrating that the proposed approach achieves high accuracy and low latency compared to traditional cloud-based solutions. The results confirm the framework’s effectiveness in maintaining privacy, reducing computational load, and enabling timely fault prediction in real-world scenarios. This work contributes to the advancement of intelligent, secure, and scalable maintenance strategies for next-generation IoT infrastructures.

کلیدواژه ها:

Federated Learning ، Edge AI ، Predictive Maintenance ، Smart Environments ، IoT ، LSTM Autoencoder ، TinyML ، Anomaly Detection