Refined Advanced Surrogate Assisted Multi-Objective Optimization Algorithm- RASAMO

Abstract

Multi- objective optimization in structural applications is generally performed with the help of complex computer codes such as Finite Element Analysis (FEA) which are computationally very expensive. Surrogate models or meta- models are comparatively economical and very useful to optimize design solutions. In the earlier studies, authors have developed Advanced Surrogate Assisted Multi- objective Optimization Algorithm (ASAMO) by creating and selecting best single and mixture surrogate models for each offspring solution by Dempster- Shafer theory (DST). For this purpose MATSuMoTo, the MATLAB based tool box is modified for multi- objective optimization problems. In the present study, a Refined Advanced Surrogate Assisted Multi- objective Optimization Algorithm (RASAMO) is presented in which the quality of Pareto- Front of ASAMO algorithm is improved by adopting a Target Value Strategy.
The effectiveness of Target Value Strategy is improved: (i) by adding multiple points per optimization iteration, and (ii) by developing most efficient surrogate models. RASAMO is applied to multi- objective machine tool spindle design problem. RASAMO resulted into 1.5% improvement in NHV value and 8.5% for the spread value for less number of function evaluations as compared to ASAMO. RASAMO is very easy to apply on benchmark and engineering applications.