GAS-AU: an average uncertainty-based general adaptive sampling approach

Currently, surrogate models have been used in various fields due to their ability to save high computational cost of simulation. However, in practical engineering applications, the surrogate model constructed from the initial sample set may suffer from insufficient accuracy. Therefore, building a usable surrogate model usually requires further infilling with some new samples. The adaptive sampling can be produced new samples to gradually expand the dataset, thereby improving the accuracy of the initial model. Thus, this work develops a general adaptive sampling approach based on the average uncertainty. The new samples are generated at the point with the maximum value of the average uncertainty. Then, the initial model is updated until the updated model achieves acceptable accuracy. Six test functions and an engineering problem are employed to test the performance of the proposed approach. The results show that the proposed approach has higher priority than other approaches under the same number of added samples. Furthermore, the performance of the proposed approach is tested again by setting a stopping criterion. The proposed approach can satisfy the stopping criterion with the least number of iterations, meaning that this approach can save a lot of computational cost compared to other approaches. This work provides a reference for the design and optimization of engineering problems.