Online Uncertainty Evaluation on the Launch Vehicle Mission Re-planning Under Thrust Faults

Abstract

The propulsion system failure is statistically acknowledged as the most fatal factor of launch vehicles, which has received extensive attention. In this paper, mission re-planning is conducted to address typical thrust faults when they exceed the adaptability of the guidance system. This is achieved by generating degraded orbits according to the residual capacity of the launch vehicle using successive convex optimization. Since the uncertainties of some critical parameters provided by the fault diagnosis system are not considered in the re-planning process, their influence is then analyzed by the polynomial chaos expansion (PCE) method. Considering the non-coincident engine cut-off phenomenon, an additional coasting phase is introduced to enable the evaluation of the stochastic distribution of the final orbit by PCE. Moreover, a deep neural network (DNN) is trained to reduce the time consumption of the uncertainty evaluation process. By implementing the DNN, the terminal states can be predicted directly from the fault information, enabling the online application. Simulation results verify the effectiveness and the accuracy of the PCE-based uncertainty evaluation. Besides, the DNN-assisted PCE is confirmed to greatly improve the computational efficiency while maintaining comparable accuracy to Monte-Carlo simulation and conventional PCE. Since the computational time of the DNN-assisted PCE is on the order of hundreds of milliseconds, it can be applied in real-time to support making appropriate and reliable mission re-planning decisions based on probability.