KRIGING-BASED INFILL CRITERIA FOR EXPENSIVE OPTIMIZATION IN AEROSPACE PROBLEMS

• Main Author: Daffa Robani, Muhammad
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/39634
• Institution: Institut Teknologi Bandung

Kriging-based Optimization has recently been developed continuosly due to its rising fame. The procedure is attractive because Kriging model can give good prediction and error, hence, it is possible to make tradeoffs between sampling at the current optimal or at the highest error. One important aspect of Kriging-Based Optimization is the Infill Criteria, which is a criterion to select which design point to evaluate next by using the surrogate model of the samples before as the model to get the optimal value of its function. The selection of this infill criteria becomes important when the characteristics of the problem that is optimized are different, one of them is the appearance of noise in the response, as often found in many real-world aerospace problems. Approximate Knowledge Gradient (AKG), Reinterpolation Procedure (RI) and Expected Improvement (EI) infill criteria are developed here and used in optimizing some analytical test functions with added noise and a real-world case of minimizing the transonic airfoil’s coefficient of drag. Although the computation of AKG needs a more computational power, the infill criteria perform well at optimizing expensive-to-evaluate test functions with added noise and the optimization case of transonic airfoil compared to RI and EI infill criteria. Hence, it is interesting to continue developed and used in some real-world noisy aerospace problems optimization that uses relatively higher cost and time of evaluation.