Novel adaptive surrogate model based on LRPIM for probabilistic analysis of turbine disc

Probabilistic lifetime assessment for aero-engine turbine disc is required to ensure structural safety and reliability. For probabilistic analysis of aero-engine turbine disc, a large amount of random variables involving load, geometry, and material properties result in a high dimensional nonlinear state function for the fatigue lifetime, which can become prohibitively expensive. This paper presents a novel adaptive surrogate model for the probabilistic analysis of an aero-engine turbine disc by integrating the local radial point interpolation method (LRPIM) and directional sampling technique. The directional sampling technique includes initial sampling, limit state recognition and subsequent sampling. In order to implement the high-dimension-probabilistic analysis for the turbine disc, an adaptive scheme is proposed involving three parts, i.e. scale adjustment of local support domain, convergence test and repeated procedure of subsequent sampling. Applied to an aero-engine turbine disc probabilistic analysis problem with 11 dimensional random variables, it is demonstrated that the novel approach proposed improves the accuracy and computational efficiency with reduced sampling amount as compared to other models such as response surface method (RSM), Kriging model (KM) and artificial neural network model (ANNM). A leave-one-out (LOO) validation test is performed to verify the robustness of the prediction of the adaptive surrogate model in the probabilistic analysis process of aero-engine turbine discs. (C) 2017 Elsevier Masson SAS. All rights reserved.