Reinforcement learning-based inverse design of composite films for spacecraft smart thermal control

Thermal control is crucial for the normal operation of spacecraft, as it directly affects the performance and lifespan of the payload. The thermochromic properties of VO₂ give it a natural advantage in smart thermal control of spacecraft, but traditional design methods still struggle to meet the design expectations for thermal control composite films. Based on this, this paper focuses on the thermal control requirements for spacecraft and proposes the design of composite films that integrate the coordinated control of dynamic thermal emitters and solar reflectors using machine learning algorithms. Firstly, a reinforcement learning optimization framework was constructed using a transfer matrix method combined with deep Q-learning. Multiple Fabry-Pérot resonator stacked structures have been optimized for the design of dynamic thermal emitters. The results indicated that a three-resonator stacked structure achieved a wide adjustable emissivity range of 0.939 near the 10 μm wavelength band. Based on this structure, a solar reflector was further designed to achieve low absorption in the solar wavelength band. The resulting smart thermal control composite film achieved a low absorption of 0.180 in the solar wavelength band, while maintaining a high emissivity adjustable range of up to 0.806. The simulated thermal control performance in a space environment indicates that, accounting for solar absorptivity, the cooling power of the film can shift from -183.42 W m^-2 to 83.13 W m^-2 before and after the phase transition. Furthermore, the composite film can maintain good thermal control performance within a large range of incident angles, promoting the practical application research of smart thermal control composite films.