Vibration reliability analysis of turbine blade based on ANN and Monte Carlo simulation

Dynamic stress of turbine blade has great influence on its reliability and fatigue life. In order to decrease the magnitude of dynamic stress, frequency modulation method is often used to avoid resonance, which implies the frequency of active force must be kept away from the inherent vibration frequency of blade. At present, many models of calculating inherent vibration frequency of blade are deterministic, which didn´t consider the randomness of many parameters (such as loading parameters, geometric parameters, material parameters) in practical operation. In this paper, a variable-section torsion blade is investigated and a new approach which is the combination of finite element method (FEM), artificial neural network (ANN) and Monte Carlo simulation method (MCS) is put forward to solve the vibration reliability analysis of blade. Based on the finite element parametrical model of torsion blade and central composite design (CCD) experiment design, analysis file of the blade is complied by deterministic finite element method and applied to be loop life to create sample points. A BP network is chosen to fitting these samples and employed as a surrogate of numerical solver to drastically reduce the number of solvers call. Then Monte Carlo simulation method is used to obtain the statistical characteristics and cumulative distribution function of static frequencies and dynamic frequencies of blade. Aiming to the blade´s dangerous mode of vibration, performance function is created and the vibration reliability analysis is carried out. Moreover, the proposed method (FEM-ANN-MCS) in this paper is compared with the FEM-RSM-MCS method and Latin Hypercube samples Mont Carlo simulation method (LH-MCS) which is acted as relative precision method. The comparison result shows that FEM-ANN-MCS is a better approach for the vibration reliability analysis of the blade and it has more stable and high regression accuracy than FEM-RSM-MCS method.