| Digital twins have become a central paradigm for modeling, monitoring, and optimizing complex engineering systems across manufacturing, energy, transportation, aerospace, and civil infrastructure [1,4,6]. When paired with artificial intelligence, the worth of digital twins increases dramatically due to the fact that the digital representation of a physical asset is able to go beyond descriptive simulation into prediction, diagnosis, optimization, and adaptive control [9, 13]. This article provides a comprehensive scientific survey of artificial intelligence in digital twins for engineering, concentrating on methodological underpinnings, algorithmic families, implementation strategies, assessment metrics, and real-world design tradeoffs. The paper begins by describing the architecture of AI-driven digital twins and separating physics-based, data-driven, and hybrid modeling approaches. Thereinafter, the manuscript analyzes the digital twins’ application in relation to machine learning (supervised learning, unsupervised learning, reinforcement learning), graph neural networks, and physics-informed neural networks. One of the most critical components of this research effort is the thorough comparison of the algorithms in terms of data needs, interpretability, computational cost, resilience, and real-time applicability, as well as advantages and disadvantages. A case study on predictive maintenance for rotating machinery is created to provide a basis for the conversation. The case study includes data flow, feature engineering, potential models, evaluation metrics, and an example comparison of findings. Finally, open challenges and future research directions are discussed, including self-learning twins, edge intelligence, trustworthy AI, and federated digital twin ecosystems. The resulting manuscript is intended as a structured reference for researchers and practitioners designing intelligent digital twins in engineering settings. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.