TWINE: ISO 23247-compliant digital twin and explainable AI framework for anomaly detection

Modern manufacturing industries face critical challenges in detecting operational anomalies within increasingly interconnected systems, where even minor deviations, such as unexpected vibrations or misalignments, can lead to production downtime and reduced product quality. Traditional rule-based monitoring systems are unable to handle the growing volume and complexity of sensor data and lack the adaptability required for evolving machinery conditions. Meanwhile, existing anomaly detection approaches often operate reactively, treating anomalies as isolated events without considering the broader context of the cyber-physical system (CPS). This paper introduces TWINE, an ISO 23247-compliant framework that integrates digital twin (DT) technology with explainable AI (XAI) for real-time anomaly detection in industrial robotic arms used across manufacturing industries. The TWINE framework addresses gaps in current literature by providing a standardized, modular, and interoperable approach that enables accurate physical-digital synchronization while maintaining transparency in AI-driven decisions. The TWINE framework combines supervised machine learning (ML) with the Shapley Additive Explanation (SHAP) mechanism, enabling operators and engineers to detect anomalies and understand the underlying factors contributing to their identification. A proof-of-concept implementation demonstrates the TWINE framework’s practical feasibility in detecting anomalies in robotic arm operations with interpretable, context-aware insights.