For rotating critical parts of aero engines, such as turbine disks, it is essential to perform reliable life predictions. Probabilistic methods are ideal to investigate these life predictions. Beside other system parameters, the distribution of geometrical parameters has a strong effect on the system behavior. However, as there are always so many geometry parameters, it always takes lots of time to complete such a probabilistic analysis considering geometry.
Within this paper, a probabilistic method based on two surrogate models is proposed and applied to an analysis of a turbine disk. In order to save the computation time as well as to get accurate results, this process is divided into two cycles. The purpose of the first cycle is to filter the parameters which have little influence on the life of the disk. In this cycle, DOE method is used and a normal response surface is created as a surrogate model to calculate the sensitivities of all the input parameters. With the sensitivities some key parameters can be selected as the inputs for the second cycle. In the second cycle DACE method is used and a more accurate Kriging model is created as the surrogate model. By conducting MCS on the calculated Kriging model, the reliability of the turbine disk can be get. In this way a huge number of computations can be avoided, thus much time can be saved and the computational efficiency can be improved.