Machine learning-based probabilistic fatigue assessment of turbine bladed disks under multisource uncertainties

Abstract

PurposeThe multisource uncertainties, including material dispersion, load fluctuation and geometrical tolerance, have crucial effects on fatigue performance of turbine bladed disks. In view of the aim of this paper, it is essential to develop an advanced approach to efficiently quantify their influences and evaluate the fatigue life of turbine bladed disks.Design/methodology/approachIn this study, a novel combined machine learning strategy is performed to fatigue assessment of turbine bladed disks. Proposed model consists of two modeling phases in terms of response surface method (RSM) and support vector regression (SVR), namely RSM-SVR. Two different input sets obtained from basic variables were used as the inputs of RSM, then the predicted results by RSM in first phase is used as inputs of SVR model by using a group data-handling strategy. By this way, the nonlinear flexibility of SVR inputs is improved and RSM-SVR model presents the high-tendency and efficiency characteristics.FindingsThe accuracy and tendency of the RSM-SVR model, applied to the fatigue life estimation of turbine bladed disks, are validated. The results indicate that the proposed model is capable of accurately simulating the nonlinear response of turbine bladed disks under multisource uncertainties, and SVR-RSM model provides an accurate prediction strategy compared to RSM and SVR for fatigue analysis of complex structures.Originality/valueThe results indicate that the proposed model is capable of accurately simulate the nonlinear response of turbine bladed disks under multisource uncertainties, and SVR-RSM model provides an accurate prediction compared to RSM and SVRE for fatigue analysis of turbine bladed disk.

A novel reliability analysis methodology based on IPSO-MCopula model for gears with multiple failure modes

“…Zhou et al 18 proposed a non-empirical model based on the Nataf transformation to quantitatively analyze the common-cause failure probability of load-dependent systems, which transformed one complex correlation problem into two independent problems. Zhu et al 19 proposed a combined machine learning strategy to evaluate the fatigue of turbine blade disks. Luo et al 20 proposed a method combining enhanced uniform importance sampling coupled and support vector regression (EUIS-SVR) for structural reliability analysis with low failure probability.…”