Physics-Informed Optimization of a Printed Circuit Board Stator Axial-Flux Motor for Robotic Actuators

This paper proposes a physics-informed neural network (PINN)-based design optimization framework for a compact smart actuator motor in physical artificial intelligence (Physical AI) robotic systems, where artificial intelligence is integrated with sensing, actuation, and physical interaction to perceive and act in the real world. To improve integration density and joint compactness, a printed circuit board (PCB) stator axial-flux permanent magnet (AFPM) motor is adopted as a thin and highly integrated topology. Because its performance is strongly affected by coupled design variables, including outer diameter, PCB count, turns per slot, trace width, and magnet thickness, an efficient and physically consistent optimization method is required. The proposed framework constructs a finite element analysis (FEA)-based design database and trains a physics-reconstructed PINN surrogate model to predict torque and loss components. Unlike purely data-driven models, the proposed PINN reconstructs output power and efficiency using physical power-balance relations, thereby improving consistency among torque, loss, output power, and efficiency. The trained surrogate is coupled with the nondominated sorting genetic algorithm II (NSGA-II) to maximize torque and efficiency while minimizing AC loss under dimensional and performance constraints. The optimized candidates are further verified by high-fidelity FEA. The results demonstrate that the proposed framework provides an effective Physical AI-oriented design methodology for compact robotic smart actuators by integrating PCB stator AFPM motor topology, physics-informed learning, and multi-objective optimization.