Steam Turbine Last Stage Lifetime Prediction Based On Neural Networks

Last stage blades are key parts, the lifetime prediction of blades is very important. Based on Finite Element Method, analyzing the dynamic stress and static stress equation, after that analyzing the lifetime of blades with improved local stress—strain method, main influencing factor of stress and distribution of stress could be got. Get the training samples with the three-dimensional finite element model and improved local stress—strain method. The maximal stress and lifetime shell be got with the trained neural network.Based on Finite Element Method, analyzing element kinetic energy\ complementary strain energy\stiffness and derive the motion differential equation. The motion differential equation derives dynamic stress and static stress equation. Analyzing the dynamic stress and static stress equation, the main influencing factor shell be got. Base on the Neuber rule, analyzing blade local stress—strain equation, improve the strain—lifetime curve with machining/assembly/running, the actual strain and lifetime curve shell be got and the main influencing factor shell be got.Analyzing the structure of the last stage blade, build the model of root and body separately. Because shape of blade body is complex, NURBS method build character section curve, divide the curve with equality define point. Using sweep tool and sew tool build the 3D model. Base on the 3D model, analyzing the moving frequency and stress. Compare test result to the 3D model result, confirm the correct of model. The stress neural network sample is each character section maximum stress. Base on improved local stress—strain method, get the lifetime neural network sample.With the learning and training of neural network using MATLAB neural network toolbox by samples, the mapping relation between main influencing factor and stress/fatigue lifetime could be confirmed. The calculation result is compared with the data which calculated by finite element method and local stress—strain method, which prove correct and valid of BP neural networks in stress analyzing and lifetime prediction. After that, base on OLE interface, Compile software with VB and MATLAB, the software system contain stress analyzing and lifetime prediction model and general neural networks model which can training and testing the networks. After testing, the software system is tested and verified.