Design Optimization with Variable Screening by Interval-Based Sensitivity Analysis

Abstract

Design optimization problems are very common in engineering practice. Determining their solution may be challenging when many design variables are involved. A means to cope with such large number of design variables consists of first screening influential variables which drive the objective function the most. Then the optimization is carried out with respect to the influential variables while the other noninfluential variables are fixed at specific values. There is no doubt that an accurate identification of influential variables is crucial for high-dimensional optimization problems. In this paper, an interval-based sensitivity index is introduced to identify the influential variables. It was compared theoretically with two types of existing indices. The performance of these indices for dimensionality reduction in optimization was examined using a test function. The proposed procedure for high-dimensional design optimization with variable screening was analyzed considering two illustrative examples. Then the proposed strategy was applied to a practical engineering problem involving an aeronautical hydraulic pipeline. The results show that the interval sensitivity index is an effective tool and is superior to the other two existing sensitivity indices for variable screening in design optimization.