Calculation of hybrid reliability of turbine disk based on self-evolutionary game model with few shot learning

Abstract

A turbine disk design based on uncertainty quantifies the risks and improves the structural reliability. An optimized design issue related to turbine disk fatigue life reliability with subjective and objective multi-source uncertainties is a three-layer nested analysis process, where each iterative step requires a double-layer nested analysis containing probability and improbability. A self-evolution game model is proposed to solve the efficiency problem of the optimized design issue under a specific accuracy requirement. In the initial state, only a few sample points are needed to start the game. In the process of its evolution, the most valuable points are searched automatically, and performance functions are identified only at the points where the game diverges. The maximum failure probability is obtained after game consistency is achieved. Accuracy and efficiency are verified with typical numerical examples. The method is adopted to analyze the hybrid low cycle fatigue reliability of a real turbine disk. It is found that, owing to the elimination of an unnecessary artificial distribution hypothesis, the reliability is lower under the same safety life, and the results are more in line with engineering practice.