Efficient Global Optimization of Actuator Based on a Surrogate Model Assisted Hybrid Algorithm

Computationally expensive numerical techniques are often involved in the actuator design optimization process, where efficiency is a major issue. Although surrogate-based optimization is a promising solution, the challenge to the optimization efficiency is still considerable. Aiming to address this challenge, a new method, called the parallel adjoint sensitivity and Gaussian process assisted hybrid optimization technique (PAGHO), is presented. The central concept is a new optimization framework employing computationally cheap partial derivatives obtained by the adjoint sensitivity method to tackle computationally expensive infill sampling for surrogate-based optimization. A silicon microactuator and a mathematical benchmark problem with different kinds of challenges are selected as the test cases. Comparison results show that PAGHO can obtain comparable results with popular global optimization methods, while at the same time having significant advantages in efficiency compared to standard global optimization methods and state-of-the-art surrogate-based optimization methods.