A Co-kriging Multi-fidelity Surrogate Model Assisted Robust Optimization Approach

The purpose of robust optimization design is to minimize the impact of uncertainty on the output value of the system while ensuring the best performance of the system output.In the field of modern product optimization design,simulation has become an indispensable means in the field of design optimization.With the development of science and technology,the accuracy of simulation is improved.Along with that came the increasing of time cost and resource consumption.The application of robust optimization design is confined to these practical problems.By replacing the expensive simulations with the surrogate model,the accuracy of the initial problem is guaranteed with a lower cost.The Co-Kriging multi-fidelity surrogate model is constructed to integrate the sample data from both low-fidelity(LF)and high-fidelity(HF)models.In this way,the multi-fidelity surrogate model could efficiently obtain the balance between prediction accuracy and modeling cost,thus multi-fidelity surrogate model is brought into focus in robust optimization design field.However,most designers treat the surrogate model as a deterministic model and ignore the objective existence of interpolation uncertainty.While the existing works are either lack of a comprehensive quantification of all existing uncertainties including the design variable uncertainty,noise parameters uncertainty and the surrogate model uncertainty,nor limited to the selection of surrogate models.In this thesis,the whole situation has been expanded to the multi-fidelity model from the single fidelity,which made the application scope of the comprehensive uncertainty quantization method greatly expanded too.It also increased the efficiency of the robustness optimization assisted with multi-fidelity surrogate model.