Multi-disciplinary Optimization of Wing Sandwich Structure using Proper Orthogonal Decomposition and Automatic Machine Learning

Abstract

The coupling between different disciplines for multi-disciplinary optimization greatly increases the complexity of a computational framework, while at the same time increasing CPU time and memory usage. To overcome these difficulties, first, proper orthogonal decomposition and radial basis function were used to generate a reduced-order model from the initial experimental points. Second, analysis results for additional experimental points were predicted using the reduced-order model. Third, using automated machine learning, surrogate models for the objective and constraint functions were obtained from the analysis results at the initial and additional experimental points. Last, optimization was performed using the surrogate models for the objective and constraint functions. As an example, the multi-disciplinary optimization problem of determining the thicknesses of the composite lamina and sandwich core when the composite sandwich structure was used as an aircraft wing skin material was analyzed.