A built-in prime benefit of Kriging surrogate model resides in its unbiased prediction and the associated confidence intervals

comparisons have been done between the classical DoE methods and Kriging model based sequential design. This latter considers both the design space global exploration and optimum-neighborhood exploitation. A parallel point-adding training strategy and its corresponding convergence criterion were introduced to improve the model precision. The approach had been illustrated with two classical optimization test functions. Results show not only a more accurate model but also a possibility to reduce the number of sampling points. Finally

the training strategy was experimented for the optimal design of a sliding bearing. Friction power loss per unit load capacity was taken as the objective function to be modeled. Two surrogate model based optimizations had been per-formed based on classical DoE such as Orthogonal Latin Hypercube and Kriging sequential strategy respectively. These two optimization results are compared with a previously optimization using Complex-optimization method. Within a limited number of iterations

the Kriging model based training strategy showed the best convergence to the global optimum among those 3 methods.