Research On Prediction Model Of Plate Roller Straightening Process Based On GA-BP Neural Network Algorithm

With the rapid development of science and technology,the demand for highstrength performance steel sheets and strips in infrastructure,transportation and aerospace fields has increased,but at present,the automation of domestic straightening machines and leveling machines is low,and the production process mainly relies on empirical parameter settings.Straightening is characterized by geometric nonlinearity,material nonlinearity,complex deformation process,difficult to solve,and time-consuming finite element simulation,all of which are not applicable to field production.In this paper,we establish a neural network(BP)model and perform genetic algorithm(GA)optimization to achieve fast and efficient formulation of the production process,and the main research contents and conclusions are as follows:(1)Based on the process simulator model,analyze the straightening parameters,construct a BP neural network model of the straightening process and force-energy parameters,and train the process database calculated using the curvature integral model.A 3-input 14-output model with plate thickness,Young’s modulus,and yield strength as inputs,and head-to-tail roll depression,residual curvature ratio,and straightening force as outputs was established,and the prediction results showed that the prediction accuracy gradually improved as the amount of data increased;the training model error was larger under smaller data amounts.(2)The BP neural network prediction model is optimized by genetic algorithm and greedy strategy,and the research results show that: under the same amount of data,the error under the first and last roll press is reduced to within 0.2mm,the error of residual curvature ratio is reduced from 20% to 32% to within5%,and the error of straightening force is reduced from 18% to 36% to within 7%after the model optimization,and the accuracy and stability of its prediction have been very obviously The accuracy and stability of its prediction have been improved significantly.(3)Self-learning of field data was established,and the field data was obtained by GA-BP model,and the error was found to be within 3% after the model was continuously revised.By comparing and analyzing the results of GA-BP prediction,finite element simulation and straightening experiment,the error between the finite element simulation value and the experimental value is within14%,the error between the prediction value and the experimental value is within6%,and the error between the prediction value and the finite element simulation value is within 8%.