Joint Trajectory Replanning for Mars Ascent Vehicle Under Propulsion System Faults: A Suboptimal Learning-Based Warm-Start Approach

This article presents a suboptimal joint trajectory replanning (SJTR) method for Mars ascent vehicle (MAV) launch missions under propulsion system faults. Conventional step-by-step trajectory replanning may fail to make timely decisions, risking mission failure. The SJTR method formulates a joint convex optimization problem of target orbit and flight trajectory after a fault. By applying penalty coefficients for terminal constraints, it adheres to the orbit redecision principles, enabling a concise and rapid solution. To further enhance the convergence and the accuracy of orbit-type determination, a learning-based warm-start scheme is proposed. Offline, a deep neural network (DNN) is trained with data generated by various trajectory replanning methods following the redecision principles. Online, the DNN provides initial guesses for the time optimization variables based on the fault scenario. Numerical simulations on mass flow rate and specific impulse drops validate the reliability of the proposed method, demonstrating at least 49.5% higher computational efficiency compared with the upgrading and downgrading replanning methods.