Multifidelity Framework for Small Satellite Thermal Analysis

Anomalies, unexpected events, and model inaccuracies have detrimental effects on satellite operations. High-fidelity models are required, but these models quickly become large and expensive. Cheap or low-fidelity models speed up computation but lack accuracy. To compromise these requirements, this study proposes a multifidelity framework based on cokriging. The proposed multifidelity framework is compared against three other standard methods often used in satellite simulations: a standalone gated recurrent unit, Gaussian process regression, and the autoregressive integrated moving average with explanatory variables model. The robustness of high-fidelity data point placement is also examined. Moreover, the real-time aspect of the simulation is considered by applying the sliding window technique. This multifidelity framework is demonstrated using temperature data obtained from thermal vacuum testing of Small Demonstration Satellite 4: a 50-kg-class satellite. The multifidelity framework provided higher accuracy and robustness than the other methods, however, having a higher computational cost as compared to a purely low-fidelity model. Up to 92% reduction of the error was achieved by the proposed framework.