Towards Spacecraft Real-Time Thermal Simulation with Artificial Neural Networks

José Daniel Reis Junior¹; Ana Maria Ambrosio¹; Fabiano Luis de Sousa¹

¹ National Institute for Space Research - INPE

doi:10.20906/CPS/COB-2015-0316

Resumo

One of the most complex software developed to support the spacecraft operation phase is the Operational Spacecraft Simulator (OSS). It must provide data in real-time to the spacecraft operators, as if they were transmitted from the real spacecraft, with the best possible fidelity. All spacecraft subsystems are presented in the OSS and their computer models need not only be representative of the real subsystem performance behavior, but also computationally inexpensive, such that real or near real telemetry response simulation is guaranteed. The spacecraft simulation model used for its thermal design is usually computationally expensive for application in OSS, and simplified models must be developed for this purpose. In this work, the use of Artificial Neural Networks (ANN) is investigated as a potential estimator of the spacecraft thermal behavior in orbit. The trained ANNs are used to provide the transient thermal behavior of a simplified model of a hypothetical CubeSat in Earth orbit. Results show that the ANN can reproduce with good fidelity the orbit temperatures generated by a thermal simulation software including points not used for training.

Palavras-chave: thermal design; spacecraft; real-time simulation; artificial neural networks