| The primary function of a thermal protection system (TPS) is to protect the space vehicle from extreme aerodynamic heating and to maintain the underlying structure within acceptable temperature limits. Currently used TPS are not load bearing members. One potential method of saving weight is to have a load-bearing TPS that performs both thermal and structural functions. One such concept called the Integrated Thermal Protection System (ITPS) uses a corrugated-core sandwich structure. Optimization of an ITPS requires thousands of three-dimensional (high-fidelity model) simulations, which is very expensive. Hence, a finite element (FE) based homogenization procedure is developed in which the ITPS is modeled as an equivalent orthotropic plate. The results for deflection and stresses obtained using the plate model (low fidelity model) are not very accurate. The high-fidelity model is analyzed at only few designs in order to reduce the cost, and these results in errors in the response surfaces based only on the high-fidelity model. To resolve this difficulty, the low fidelity two-dimensional (2D) plate models is fitted with a high quality surrogate, which is then corrected by using a small number of high fidelity three-dimensional (3D) finite element analyses. Fitting the difference or the ratio between the high fidelity analyses and the low fidelity surrogate with a response surface approximation allows construction of the so called correction response surface. This multi-fidelity or variable-complexity modeling requires significantly fewer high fidelity analyses for a given accuracy.;A MATLABRTM and a JAVA code has been developed in conjunction with the Surrogate Toolbox in order to carry out these FE analyses automatically using ABAQUSRTM. The multi-fidelity response surface approximation (RSA) is used to optimize the mass of the ITPS for a given material combination and loading conditions. For the same given accuracy, multi-fidelity response surface took 30 percent less time as compared to the full 3D response surface. Further, one can choose the best correction model based on the data without the test points, and the test points also confirmed the choice. |
