Neural Network-Based Dynamic Stress Prediction Method for Bladed Disks

This study addresses the challenge of monitoring dynamic stresses in high-pressure turbine blades of aviation engines by designing a non-contact dynamic stress prediction method based on a BP neural network for rotating Disks. Firstly, the research focuses on a bladed disk, simulating blade tip displacement and blade root strain under various excitation conditions, and calculating stress outcomes using Hooke’s Law. Subsequently, an input-output neural network surrogate model is constructed with disk excitation conditions as inputs and blade tip displacement and monitoring point stress as outputs. Finally, the input and output data are divided into training and testing sets, and the surrogate model is trained and validated, ultimately establishing a non-contact dynamic stress prediction model for the disk under the influence of multi-parameter coupling. The data for building the neural network surrogate model is obtained through flat-bladed disk simulations, with model prediction errors within 6%.