Design of a digital twin of an electrical drive using Machine Learning

Physics-Informed Neural Networks (PINNs) are a powerful and emerging tool for solving problems related to dynamical systems governed by Ordinary Differential Equations (ODEs). Unlike traditional data-driven models, PINNs incorporate the residuals of differential operators directly into the cost function, ensuring that the learned solution is not only data-consistent but also physically plausible. This enables the network to approximate system dynamics while minimizing the discrepancy between observed and theoretically expected behaviors. This study focuses on the development of a PINN that employs the inverse problem approach to estimate the passive parameters of an AC/DC power converter, specifically its inductances and capacitance. These parameters play a crucial role in the converter’s operation: inductances influence the dynamic response of the system by regulating current variations, while the capacitance C stabilizes the DC-link voltage, ensuring proper energy storage and distribution. Accurate estimation of these components is essential for modeling and optimizing power electronic devices. The inverse problem framework treats the physical parameters of the ODEs as trainable variables, allowing them to be inferred directly during the learning process. Instead of being predefined, inductances and capacitance are dynamically updated during training by embedding them in the cost function through the residuals of the governing equations. This approach enables the PINN to efficiently extract hidden physical information from observed signals, even in scenarios where direct measurements are challenging or impractical. The development of AI-driven methodologies for power electronics presents a significant opportunity for advancing real-time digital models of power converters. These AI-enhanced digital twins provide substantial benefits in performance optimization, efficiency improvement, and predictive maintenance. By combining physics-based modeling with deep learning techniques, PINNs offer a data-efficient and interpretable solution for monitoring and controlling power electronic systems. Such models not only enhance reliability but also contribute to the development of more sustainable and adaptive energy management strategies, paving the way for next-generation intelligent power systems.