基于神经网络的短粗转轴裂纹诊断研究

STUDY ON CRACK DIAGNOSIS FOR PODGY SHAFT BASED ON NEURAL NETWORK

构造Rayleigh-Timoshenko区间B样条小波梁单元,建立短粗转子系统小波有限元模型。求解与不同裂纹相对位置和相对深度相对应的裂纹转子前三阶固有频率。将裂纹相对位置、相对深度、前三阶固有频率作为神经网络的训练样本,训练出用于裂纹定量诊断的神经网络。以实测固有频率作为训练好的神经网络的输入,定量诊断测出裂纹的相对位置和深度。试验研究表明,所提出的裂纹诊断方法具有较好的精度和鲁棒性,且易于在工程实践中进行短粗裂纹转子定量诊断。

The Rayleigh-Timoshenko beam element based on B-spline wavelet on the interval (BSWI) was constructed for discrete podgy shaft and stiffness disc. The wavelet-based finite element model of rotor system was built up to solve the first three natural frequencies which are functions of relative crack location and depth. The relative crack location, relative crack depth and the first three natural frequencies were employed as the training samples to achieve the neural network for crack diagnosis. Measured natural frequencies were served as inputs of the trained neural network and the relative crack location and depth could be identified. The experimental results verify the validity of the method, which is feasible for practical application to crack diagnosis in podgy rotor system.