Joseph Nyangon
The integration of machine learning with domain-specific physics has revolutionized the design, monitoring, and control of electrical machines and drives. This state-of-the-art review of physics-informed machine learning (PIML) is being leveraged to address data scarcity, improve model interpretability, and enforce physical laws, thus achieving computationally cost efficient and accurate solutions. The review also discusses challenges related to parameter sensitivity, dynamic behavior, and robustness, thereby underlining the potential of PIML to deliver computationally efficient, accurate, and scalable solutions for Industry 4.0 applications. Special emphasis is placed on hybrid physics-informed neural network (PINN) models that combine machine learning with physics-based principles for real-time diagnostics, digital twins, fault detection, and high-fidelity modeling, control and optimization of electrical machines and drives. Notable examples of these hybrid models include Deep Operator Networks (DeepONets), Fourier Neural Operators, Extreme Learning Machines (ELM)-enhanced PINNs, Graph-Based PINNs (PIGNNs), and domain decomposition PINNs, all of which illustrate the shift from traditional black-box approaches to transparent, physics-informed strategies. Through a state-of-the-art review of AI methodological contributions of PIML case studies and simulations, the potential of PIML to provide robust, scalable solutions is demonstrated, fostering a paradigm shift from black-box data-driven approaches to white-box, physics-informed strategies.