DIGITAL TWIN–BASED INTELLIGENT MONITORING OF INDUSTRIAL SYSTEMS USING EXPLAINABLE AI

Industrial systems increasingly rely on Industrial Internet of Things (IIoT) sensors for real-time monitoring and predictive maintenance. However, most existing digital twin–based monitoring solutions depend on static or black-box machine learning models, limiting interpretability, operator trust, and safe deployment in safety-critical environments. In response to these challenges, the author develops the Adaptive Hybrid Digital Twin with Causality-Aware Explainable Artificial Intelligence (HADT-C-XAI) framework to offer transparency and intelligence in industrial monitoring. The framework describes three integrated layers: (i) acquisition of real-time sensors, (ii) continually synchronized hybrid digital twin modeling, which is the integration of physics and data hybrid modeling and (iii) an intelligent analysis layer where LSTM-based anomaly detection is ungraded with explainable feature attribution. A closed-loop learning mechanism updates the model dynamically to adapt to operational drift while generating interpretable fault causes for operator decision support. Experiments were conducted on a multi-sensor industrial testbed containing 120 hours of vibration, temperature, acoustic, and rotational data. The implemented system shows a 94.8% detection accuracy, 95.4% recall, and a 4.1% low false alarm rate, which surpasses standard LSTM (88.5%) and threshold-based monitoring (82.9%). With edge-level inference, detection latency has been reduced to 26-30 ms, which allows for real-time deployment. Results demonstrate that integrating adaptive digital twins with explainable AI improves reliability, transparency, and fault diagnosis while maintaining computational efficiency. The proposed framework provides a scalable and trustworthy solution for predictive maintenance, Industry 4.0 applications, and cyber–physical system monitoring.