The traditional approaches in material research and hardware design are insufficient to address the evolving Operation and Maintenance (O&M) demands in contemporary power electronics. Overengineering and data acquisition practices lead to unsustainable costs and reduced profit margins. Digital Twins (DTs), defined as real-time simulation models of physical systems, emerge as promising solutions to meet stringent O&M requirements. In power electronics, DTs offer significant potential in thermal management, crucial for control performance, safety, and system lifespan. This paper aims to analyze the development of computationally efficient and high-fidelity DTs tailored for power electronics applications, emphasizing their predictive reliability. To achieve this goal, the proposed physics-based approach is enhanced by integrating Data-Driven Artificial Intelligence (AI)based techniques. The predictive reliability of the DTs produced through this workflow is then experimentally validated against a power electronic converter designed for induction heating applications. Additionally, the feasibility of real-time execution is demonstrated, affirming the practical applicability of the developed DTs.