A Reinforcement Learning-Based Multiphysics Optimization of Composite Material Component With Slot-Structure in UAV System

In unmanned aerial vehicle (UAV) systems, the shielding effectiveness (SE) of composite material components with slot structures is analyzed and optimized under thermal and fluid constraints via reinforcement learning in this article. To improve the SE of UAV systems when satisfying the balance requirements of the electromagnetic-thermal-fluid (ETF) multiphysics coupling, the twin delayed deep deterministic policy gradient (TD3) algorithm for ETF with a carefully designed reward is proposed, which is termed ETF-TD3. To accelerate the multiphysics optimization process, an ETF predictive convolutional neural network (ETF-PCNN) is proposed to predict multiphysics results and integrated into the ETF-TD3 algorithm. Three typical composite material components in UAV systems, including UAV airframe and electric equipment boxes, are used to apply the proposed ETF-TD3 algorithm with the ETF-PCNN method. Optimization results and computational cost demonstrate the benefits and advantages of the ETF-TD3 algorithm in the multiphysics design of slotted composite material components. When compared with some commonly used optimization algorithms, the optimal design obtained by the ETF-TD3 algorithm achieves better SE while the thermal management and aerodynamic characteristics are satisfied. In addition, the proposed algorithm provides design guidelines while revealing the underlying physical principles.