Life reliability assessment of twin-web disk using the active learning kriging model

Abstract

Compared with the traditional single turbine disk, the novel twin-web turbine disk (TWD) has a bright future in the application of the aero high-pressure disk due to its advantages in weight loss and heat transfer. With the strong coupling of multiple disciplines and the uncertain factors during the manufacturing process, it is necessary to analyze the multidisciplinary reliability of its fatigue life cycles in an efficient way. In this paper, based on the stratified sampling, a new sampling method called modified quadrangular grid (MQG) method, which has better efficiency and stability of sampling, is applied with the active learning Kriging (ALK) surrogate model and the Monte Carlo simulation. The distance based on sampling efficiency (DBSE) and information entropy are proposed in this paper as the criteria to evaluate the current sampling methods. A platform is then built for analyzing its fluid characteristic and structural strength performance. And the reliability of low-cycle fatigue (LCF) life is analyzed by Monte Carlo simulation with ALK model (ALK-MC). Results show the failure probability of current design is higher than the criterion which needs further reliability-based optimization design.