Efficient response surface method for fatigue life prediction

Abstract

An efficient response surface construction method for fatigue life prediction is investigated in this study, based on the Gaussian process (GP) modeling approach. The model is applied to fatigue crack growth analysis, in order to save the computational effort in calculating the stress intensity factor through finite element analysis. The advantage of GP modeling is shown through comparison with traditional methods such as polynomial approximation developed by previous researchers. Then the GP method is applied to railroad wheel crack growth analysis. The GP method is also applied to develop an analytical model of fatigue crack growth data. An adaptive sample selection procedure is implemented to improve the accuracy of the GP model. The use of the GP approach can significantly reduce the number of tests required to model the fatigue crack growth data. Two sets of experimental data are used to validate the method.