Surrogate Modeling: Review and Opportunities for Expert Knowledge Integration

Digital Twins (DTs), essential for digitalization and Industry 4.0, demand high-fidelity yet computationally efficient models for real-time replication of physical systems. However, simulating continuous dynamical systems, often stiff and highly granular, poses significant computational challenges. Driven by industry's need for more efficient simulations, this study explores Surrogate Modeling (SMing) as a solution to reduce computational costs while preserving accuracy, enabling real-time performance, iterative design optimization and enhancing feasibility of DT simulations. We review existing traditional and machine learning SMing approaches, analyzing their limitations, accuracy and efficiency. We, furthermore, emphasize the critical role of domain expertise in SMing workflows and explore systematic strategies for incorporating expert knowledge to improve model reliability and applicability. Finally, we identify and discuss challenges and opportunities that emerge by the fusion of expert knowledge with machine learning techniques, highlighting their potential to advance next-generation SMing techniques.