Interval prediction of the fatigue crack cycle using evidence theory and the Kriging meta-model

Abstract

The fatigue crack cycle is calculated via a specimen experiment under specific stress conditions; however, the experimental results are scattered due to various causes (e.g., experimental conditions, material specifications). Therefore, the fatigue crack cycle is evaluated by fitting it to a specific probability model after carrying out a sufficient amount of testing. If the number of tests is insufficient, a specific probability model is unobtainable. The uncertainty due to an insufficient number of tests is referred to as the epistemic uncertainty. In this paper, evidence theory is employed to deal with epistemic uncertainty. Through the use of evidence theory, belief and plausibility functions are obtained. Belief and plausibility may be implied as the lower and upper bounds for the limited state function. Here, the belief and plausibility functions are approximated to the Kriging meta-model using sample data generated by the Cartesian product. These approximated functions can then be used to calculate the optimal solution using a genetic algorithm. Finally, the fatigue crack cycle is predicted in terms of the interval (belief and plausibility).