Evidence-Based Multi-disciplinary Robust Optimization for Mars Microentry Probe Design

摘要

Atmospheric pressure on Mars is approximately 1 % of that on Earth and varies about 15 % during the year due to condensation and sublimation of its primarily CO_2 atmosphere. Impacts of the uncertainties during the entry are difficult to be modeled. The situation becomes more complex when uncertainties are from different disciplines. In this work, a robust multi-disciplinary optimization method for Mars microentry probe design under epistemic uncertainties is presented. Objectives of the evidence-based robust design are set to minimize the interior temperature of thermal protection systems (TPS) and maximize its belief value under uncertainties. A population-based multi-objective estimation of distribution algorithm (MOEDA) is designed for searching the robust Pareto set. Candidate solutions are adaptively clustered into groups. In each group, principal component analysis (PCA) technique is performed to estimate population distribution, sample and reproduce individuals. Non-dominated individuals are sorted and selected through the NSGA-II-like selection procedure. Adaptive sampling and binary branching techniques are employed for computing the evidence belief functions. PCA dimensionality reduction technique is implemented for identifying and removing uncertain boxes with little contribution of the beliefs. With variable fidelity model management, analytical aerodynamic model is used first to initialize the optimization searching direction. Artificial neural network (ANN) surrogate model is used for reducing the computational cost. When the optimization goes close to the optima, more data from the high accuracy model are put into the aerodynamic database, making the optimization procedure converge on optima quickly while keeping high-level accuracy.