A Combined Enhanced Comprehensive Learning Particle Swarm Optimization with Gaussian Process Regression Model for Size and Shape Optimization of Space Trusses

Abstract

The paper proposes the combined machine learning-based, called Gaussian process regression (GPR), method with enhanced comprehensive learning particle swarm optimization (ECLPSO) algorithm to perform the simultaneous size and shape optimization of space trusses under applied forces. At variance with standard meta-heuristic design techniques, the approach advantageously by-passes the need to iteratively call the time-consuming finite element analyses for structural responses through the construction of the GPR predictive model. The model maps out the accurate structural behaviors from the sufficient input (i.e., nodal coordinates and member sizes) and output (member forces and nodal displacements) dataset generated by a series of structural analyses. The ECLPSO algorithm is then performed solely on the computed GPR model presenting the sufficiently accurate response predictions. The accuracy and robustness of the proposed method are illustrated through the designs of space trusses successfully solved, where the minimum total weight can be achieved.