High-to-Low Initial Sample Ratio of Hierarchical Kriging for Film Hole Array Optimization

摘要

This study aims at proving the existence of the optimal number of high initial sample for time-efficient optimization and demonstrating that the reliable optimization results can be guaranteed when the hierarchical kriging model is used as a surrogate model. The hierarchical kriging model is one sort of variable-fidelity modeling, which mimics the general tendency of a model based on the low-fidelity information and calibrates the accuracy of the model with the high-fidelity information. There, however, have been still several ambiguities to be clarified for practical utilization of hierarchical kriging model. For clarifying these, the film-cooling hole arrangement is selected as an optimization target The optimization is progressed with the expected-improvement-based efficient global optimization algorithm coupled with the hierarchical kriging model. The fidelity in this paper is classified by mesh density: 2 million and 8 million grid system. According to this overall optimization process, case studies are conducted with respect to the number of high initial samples. As a result, the reliable optimization results are obtained regardless of different numbers of high initial samples, and approximately 40% time reduction can be obtained.