Abstract
This paper presents a deterministic and computationally efficient framework for joint temperature and conductance estimation in spacecraft thermal networks, suitable for real-time CubeSat applications. We introduce a bootstrapped Gauss-Seidel (BGS) estimator that combines sliding-window least-squares updates with statistical bootstrapping to provide uncertainty quantification at a fraction of the cost of stochastic filters. An extended variant incorporates a graph Laplacian regularizer to enforce spatial smoothness across thermally connected nodes, improving stability under sparse sensing. The framework is benchmarked against a Least Squares baseline, a standard Gauss-Seidel solver, and an Ensemble Kalman Filter (EnKF) on a 16-node CubeSat thermal network with realistic orbital heating profiles. Results show that BGS achieves near-EnKF accuracy in conductance and temperature estimation while being much faster. The Laplacian-regularized variant further reduces temperature prediction errors in low-observability regions, with negligible runtime overhead. Both deterministic solvers demonstrate predictable complexity, numerical stability, and interpretability suitable for embedded deployment.
Funding Information
  • JSPS KAKENHI (JP24K01110)