Privacy-Preserving Explainable Federated Digital Twins for Predictive Maintenance in AIoT-Enabled Smart Manufacturing

Industrial artificial intelligence increasingly depends on data collected from heterogeneous sensors, enterprise systems, and maintenance logs; however, centralizing such data is often limited by privacy, governance, latency, and plant-specific operating conditions. This paper proposes a privacy-preserving explainable federated digital twin framework for predictive maintenance in AIoT-enabled manufacturing. The framework integrates edge sensor preprocessing, local temporal degradation models, federated aggregation, uncertainty-aware remaining-useful-life estimation, and explainable artificial intelligence to support maintenance decisions without transferring raw operational data across plants. A controlled simulation with five virtual manufacturing sites was designed to represent non-identically distributed vibration, temperature, current, pressure, and acoustic signals under normal, degraded, and pre-failure states. The proposed federated digital twin was compared with local-only and centralized baselines using F۱-score, area under the ROC curve, mean absolute error for remaining useful life, communication cost, and interpretability consistency. Results indicate that the proposed approach improves cross-site generalization compared with isolated local models while reducing raw-data exposure compared with centralized learning. The paper concludes that federated explainable digital twins can provide a practical interdisciplinary bridge between artificial intelligence, industrial engineering, cybersecurity, and managerial decision support in smart factories.

کلیدواژه ها:

maximum Artificial intelligence ، Federated learning ، Digital twin ، Predictive maintenance ، AIoT ، Explainable AI ، Smart manufacturing ، Privacy preservation