Towards hydrogen-powered electric aircraft: Physics-informed machine learning based multi-domain modeling and real-time digital twin emulation on FPGA

In response to environmental concerns related to carbon and nitrogen emissions, hydrogen-powered aircraft (HPA) are poised for significant development over the coming decades, driven by advances in power electronics technology. However, HPA systems present complex multi-domain challenges encompassing electrical, hydraulic, mechanical, and chemical disciplines, necessitating efficient modeling and robust validation platforms. This paper introduces a physics-informed machine learning (PIML) approach for multi-domain HPA system modeling, enhanced by hardware accelerated parallel hardware emulation to construct a real-time digital twin. It delves into the physical analysis of various HPA subsystems, whose equations form the basis for both traditional numerical solution methods like Euler's and Runge-Kutta methods (RKM), as well as the physics-informed neural networks (PINN) components developed herein. By comparing physics-feature neural networks (PFNN) and PINN with conventional neural network strategies, this paper elucidates their advantages and limitations in practical applications. The final implementation on the Xilinx (R) UltraScale+TM VCU128 FPGA platform showcases the PIML method's high efficiency, accuracy, data independence, and adherence to established physical laws, demonstrating its potential for advancing real-time multi-domain HPA emulation.