SPACECRAFT FEM UPDATE USING MULTI-OBJECTIVE OPTIMISATION AND SURROGATE MODELS

Abstract eng:
The vibrations that a spacecraft experiences during launch are critical for the safety and integrity of its payload. To simulate the launch, a Coupled Load Analysis (CLA) must be carried out, in which the spacecraft (SC) finite element model (FEM) is coupled to a model of the launch vehicle (LV) to predict the loads during launch. It is extremely important to have accurate predictions of the coupled SC/LV system to verify its dynamic behaviour, and for that reason to have accurate models of all the subsystems. FEM update is the process by which the correlation between the analytical FEM and the experimental test results is improved by modifying certain parameters in the model. Discrepancies may be due to uncertainty in some parameters or perhaps incorrect modelling assumptions or unrealistic boundary conditions. FEM update, which can be very time consuming, has extensively been used in the literature in a range of civil engineering structures. Some studies have also applied FEM updating techniques to satellite structures [1, 2], either using a simple optimisation approach or using stochastic methods with neural networks. The proposed presentation investigates the use of different optimisation techniques to improve the correlation between the FEM and the experimental results for the BepiColombo SC. Traditional optimisation methods will be severely limited by the cost of a single function evaluation, and for that reason surrogate models will be considered. This study uses Isight, a digital platform that provides simulation automation and optimisation solutions, with MSC NASTRAN for the FEM of the spacecraft, to compare the results and the suitability of different methods, from gradient descent to genetic algorithms (GA) and surrogate models. Preliminary results with a simpler space structure have been successful in setting up the methodology and validating the approach. Different sets of variables were used (thicknesses, stiffnesses, etc.) and a range of responses were considered (frequencies, mode shapes, MAC). Rather than validating against test results, the solution from the baseline FEM was taken as the "nominal" or "test" solution, and certain parameter values were altered to generate a "modified" FEM, treated as the uncorrelated FEM to be updated. This served both to test the quality of the workflow as well as to explore the optimiser's capabilities. The approach was then applied to the BepiColombo FEM. The Isight workflow was built and validated against its own nominal solution. Early results show that the correlation can be improved with this method, although the cost of computation is high. The presentation will include a range of solution techniques, as well as validation against actual experimental models. The Isight workflow will be extended to include surrogates.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket