Prediction method of non-stationary random vibration fatigue reliability of turbine runner blade based on transfer learning

Abstract

In order to solve the problems such as lack of fault information, sample variation with time and expensive calculation in the estimation of the vibration fatigue reliability of the turbine runner blade under the non-stationary hydraulic excitation. A prediction method of non-stationary random vibration fatigue reliability of the turbine runner blade based on transfer learning is proposed in this paper. Firstly, the dynamics model of the cracked turbine runner blade under the non-stationary hydraulic excitation is established to analyze the characteristics of the non-stationary random vibration fatigue of the turbine runner blade. Secondly, the transformation matrix between the source domain and target domain in the hidden space is found by the transfer learning method of balanced distribution adaptation (BDA). The adaptation of active learning and Kriging-based system reliability method (AK-SYSi) is applied to estimate the non-stationary random vibration fatigue reliability of the turbine runner blade with multi-failure-mode. Finally, an example is analyzed, and the Monte Carlo simulation (MCS) is used to verify the correctness of the proposed method. The results show that the method proposed in this paper can effectively predict the failure probability of the non-stationary vibration fatigue of the turbine runner blade in future time.