Automatic problem identifier and parameter tuning in local optimization

点击翻译按钮获取中文摘要

Many engineering problems require optimization of highly complex simulations in order to achieve the best product or process design. In practice, high dimensionality, computational cost and lack of mathematical expressions are three key issues in modern optimization, and their combination makes the design optimization very challenging. Empower Operations, based in BC, is a company which offers design exploration and optimization software suitable for High dimensional, Expensive and Black-box (HEB) problems, to be used by manufacturers. Manufacturers need to use optimization in the design of their product, but not all companies have optimization specialists who can work with optimization codes. The current available optimization methods all require the user to choose the suitable approach for a given problem. Moreover, each method has special parameters which should be tuned. These became the hurdle for practical application of the otherwise powerful optimization. The product of Empower Operations, OASIS (Optimization Assisted Simulation Integration Software), aims to provide a framework in which the user can use global optimization algorithms, with minimum knowledge about the algorithm details. The OASIS software contains a module for local optimization but little work is done to find the best method for different problems. In this project, the objective is to find a strategy to identify the type of problem and tune algorithm parameters automatically, with the purpose of solving a wider range of problems. The proposed research will significantly enhance the performance of the OASIS optimization algorithm, and make OASIS accessible to a wider scope of design engineers. The developed techniques can bridge the gap between optimization theory and practical demand from engineers, and ultimately increase the global competitiveness of Canadian manufacturers.