Multi-disciplinary shape optimization of an entry capsule integrated with custom neural network approximation and multi-delity approach

This paper describes a new integrated approach for the multi-disciplinary optimization of a entry capsule’s shape. Aerothermodynamics, Flight Mechanics and Thermal Protection System behaviour of a reference spaceship when crossing Martian atmosphere are considered, and several analytical, semi-empirical and numerical models are used. The multi-objective and multi-disciplinary optimization process implemented in Isight software environment allows finding a Pareto front of best shapes. The optimization process is integrated with a set of artificial neural networks, trained and updated by a multi-fidelity evolution control approach, to approximate the objective and constraint functions. Results obtained by means of the integrated approach with neural networks approximators are described and compared to the results obtained by a different optimization process, not using the approximators. The comparison highlights advantages and possible drawbacks of the proposed method, mainly in terms of calls to the true model and precision of the obtained Pareto front.