A review of computational modeling, artificial intelligence, and digital twins in metal casting and foundry operations

Metal casting is an essential industrial process that consistently encounters issues with energy efficiency, fault minimization, and environmental sustainability. This review presents a cohesive two-dimensional framework that systematically amalgamates physics-based numerical modeling, data-driven Artificial Intelligence (AI), and digital twin methodologies throughout critical phases of the metal casting value chain, encompassing design, mold preparation, pouring, solidification, and finishing. The paradigm facilitates a systematic integration of computational methods, applications, and deployment issues, rather than the fragmented reporting of individual investigations. The paper emphasizes the pivotal role of computational modeling and AI in transforming foundry operations into intelligent and adaptable manufacturing systems. Essential numerical methods, including Finite Element, Finite Volume, Finite Difference, and Boundary Element Methods, are analyzed for their efficacy in modeling mold filling, solidification, and the progression of thermo-mechanical stresses. Microstructure modeling methodologies, including Phase Field, Cellular Automata, and Kinetic Monte Carlo, are evaluated for their efficacy in forecasting grain evolution and phase transitions during solidification. AI-driven techniques, encompassing machine learning algorithms and expert systems, facilitate fault forecasting, process enhancement, and predictive upkeep. Unlike traditional offline simulation models, digital twins are characterized as cyber-physical systems that provide real-time synchronization between physical casting operations and virtual models via bidirectional data sharing and feedback control methods. In the context of Industry 4.0, these technologies utilize Internet of Things (IoT) connection and data-driven analytics to enhance process efficiency and resource usage. Notwithstanding considerable advancements, obstacles persist in data accessibility, the integration of hybrid physics-AI models, the deployment of real-time digital twins, and extensive industrial validation, underscoring critical avenues for future inquiry and practical application.