To enhance the transient response and modeling accuracy of compressor blade vibration characteristics in time domain dynamic reliability analysis, a method called the radial basis function extremum neural network (RBFENN) is proposed. This method combines the radial basis function neural network (RBFNN) with the extreme response surface method (ERSM). By considering the time domain and frequency domain characteristics of unsteady aerodynamic excitation on blades affected by multi-row stator/rotor interference, the stress response distribution of the blades is determined through deterministic analysis. The RBFENN mathematical model is established using RBFNN to address the high nonlinearity in surrogate modeling. The ERSM is used to process the transient problem of motion reliability analysis, taking extreme values into account throughout the response process. In evaluating the vibrational reliability of the compressor blades, the maximum stress is considered as the primary focus of the study, taking into account the stochastic nature of aerodynamic excitation frequency, rotational velocity, and material parameters. It is observed that the unsteady aerodynamic excitation induces higher-order harmonic resonances in the blade velocity margin, resulting in increased stress at the leading edge of the blade root. This stress distribution follows a normal distribution pattern. A comparison among Monte Carlo methods, ERSM, and RBFENN demonstrates the superiority of the proposed RBFENN method in terms of computational accuracy and efficiency. This study introduces the RBFENN method as a learning-based approach for blade vibration reliability design, enriching the theory of mechanical reliability.
Read full abstract