Efficient Design Under Uncertainty of Spacecraft and Trajectory Systems Using Gaussian Process Surrogates

Design under uncertainty for coupled spacecraft propulsion and interplanetary trajectory systems is prohibitive due to the large number of high-fidelity multidisciplinary analysis (MDA) evaluations required for Monte Carlo-based reliability assessment. This paper presents a surrogate-enabled design under uncertainty framework for a coupled spacecraft and trajectory Earth--Mars transfer problem inspired by MAVEN, using both reliability-based design optimization (RBDO) and robust design optimization (RDO). The high-fidelity MDA integrates the Rocket Propulsion Analysis (RPA) tool and NASA’s General Mission Analysis Tool (GMAT) within OpenMDAO to model a single-stage chemical propulsion spacecraft executing a trajectory correction maneuver (TCM) and Mars orbit insertion (MOI). Gaussian process (GP) surrogates, trained on MDA-generated samples and used with Monte Carlo simulation (MCS), provide low-cost predictions of time of flight and burn-wise fuel consumption. These surrogates enable large-scale MCS evaluations within a differential evolution optimizer, where RBDO enforces a reserve fuel reliability constraint requiring at least 200 kg remaining with reliability of 0.99%, and RDO minimizes a weighted mean and variance objective to reduce sensitivity to uncertainty. Three uncertainty cases are examined: launch date uncertainty, engine specific impulse uncertainty, and combined uncertainty. Across all cases, the deterministic optimum exhibits poor post-processed reliability (22-51%), while the GP-enabled RBDO solutions achieve the target reliability with modest increases in initial fuel mass (20-39 kg) and small changes in mission-level performance. In addition, the RDO results demonstrate that the GP-enabled framework can strongly reduce variability in the fuel consumption outputs under uncertainty. The entire RBDO methodology for the combined case completes in 223 s, improving computational cost by several orders of magnitude and making uncertainty-aware co-design tractable for multidisciplinary mission studies.