Generative AI in AI-Based Digital Twins for Fault Diagnosis for Predictive Maintenance in Industry 4.0/5.0

Generative AI (GenAI) is revolutionizing digital twins (DTs) for fault diagnosis and predictive maintenance in Industry 4.0 and 5.0 by enabling real-time simulation, data augmentation, and improved anomaly detection. DTs, virtual replicas of physical systems, already use generative models to simulate various failure scenarios and rare events, improving system resilience and failure prediction accuracy. They create synthetic datasets that improve training quality while addressing data scarcity and data imbalance. The aim of this paper was to present the current state of the art and perspectives for using AI-based generative DTs for fault diagnosis for predictive maintenance in Industry 4.0/5.0. With GenAI, DTs enable proactive maintenance and minimize downtime, and their latest implementations combine multimodal sensor data to generate more realistic and actionable insights into system performance. This provides realistic operational profiles, identifying potential failure scenarios that traditional methods may miss. New perspectives in this area include the incorporation of Explainable AI (XAI) to increase transparency in decision-making and improve reliability in key industries such as manufacturing, energy, and healthcare. As Industry 5.0 emphasizes a human-centric approach, AI-based generative DT can seamlessly integrate with human operators to support collaboration and decision-making. The implementation of edge computing increases the scalability and real-time capabilities of DTs in smart factories and industrial Internet of Things (IoT) systems. Future advances may include federated learning to ensure data privacy while enabling data exchange between enterprises for fault diagnostics, and the evolution of GenAI alongside industrial systems, ensuring their long-term validity. However, challenges remain in managing computational complexity, ensuring data security, and addressing ethical issues during implementation.

---

Source: Semantic Scholar - Applied Sciences (58 citations) PDF: N/A Original Link: https://www.semanticscholar.org/paper/5dbb0776662824df97bf31db96abfea0c230068b