Research On Prediction Method Of Profile Rolling And Bending Based On Graph Genetic Programming And Error Correction

Profile rolling and bending workpiece is widely used in aerospace,automotive,shipbuilding and other fields,and is mostly used to make rocket,ship and other frame structures.Profile roll-bending forming technology is widely used compared with stretch bending technology,which has the characteristics of no need to make a specific mould,no limit on the length of profile and more flexible in making workpiece with variable curvature.The springback problem is one of the important reasons affecting the precision of roll-bending.The traditional calculation method of springback of roll bending based on mechanics theory needs to be based on many assumptions,which is affected by many factors in actual production,resulting in inaccurate prediction.This paper studies the springback problem in rolling bending of profiles based on traditional forming theory and advanced machine learning methods.Firstly,the influence factors of springback deformation in rolling bending are studied.Based on the analysis of coupling effect among several factors in the rolling and bending process of profiles,the characterization method and quantitative description of factors affecting rolling and bending rebound of profiles were established.A method of rolling and bending springback characteristic based on hierarchical clustering was proposed.By mining the similarity between rolling and bending data,different performance regions of profiles were divided and quantitative description of rolling and bending springback characteristic of profiles was given.Secondly,to solve the problem of constant curvature roll bending,we proposed a predict model based on the differential graph genetic programming.This model can through genetic iterations,analytical the coupling relationship of influence factors and profile radius after springback.Given the approximate function expressions of them.It provides the possibility of reverse control through inverse analytic expression.Thirdly,a prediction model of variable curvature rolling and bending was proposed based on graph genetic programming model and error correction network group.Based on the analytical ability of the coupling relationship of the constant curvature model,the relationship between the regions before and after the profile forming was mined,and the error correction quantity was synthesized by multiple groups of neural networks to predict the bending radius after springback in the rolling bending process with variable curvature.Finally,based on the three-roller symmetrical rolling bending machine,the roll bending rebound of aluminum alloy profiles under various processes was tested,and the concrete realization method of systematic measurement,model training and prediction was established.The accuracy of the constant curvature prediction model and the variable curvature model is verified,and the importance of the correlation relationship in the variable curvature model is proved by experiments.