Research On Springback Prediction Method Of Profile Roll Forming Based On Machine Learning

Profile bending parts are widely used in aerospace and other fields,with many types of parts,complex molding and high precision requirements.In the traditional production and processing process,it is often necessary to process a large number of molds,resulting in high production costs.Rolling equipment is processed by the rolling process,and no special mold is required,which has broad market prospects.Because roll forming is affected by the coupling of material performance parameters,geometric parameters,structural parameters,process parameters and other complex factors,coupled with the current lack of control of rolling equipment,rolling process,molding control,especially in the production of variable curvature rolling parts Because of its timing and dynamics,it is difficult to solve the problem that the springback of the molded parts is difficult to solve.Therefore,the key core technologies are in urgent need of breakthrough.To this end,this paper has conducted in-depth research on key technologies of profile rebound based on machine learning.The main research contents are as follows:Firstly,the coupling effect and springback effect of the bending material parameters,equipment parameters and process parameters on the forming curvature were studied.The mechanism of the roll forming and the springback constraint of the curvature were analyzed,and the characterization of the roll forming feature vector was established.The method analyzes the correlation of the main factors on the forming springback.On this basis,the main features are extracted and normalized in combination with the forming principle.Secondly,according to the rebound resilience relationship and the need of small sample analysis,a machine learning method is proposed to propose a fixed curvature rebound prediction model based on support vector regression to analyze and mine the correlation between fixed curvature and springback under the influence of complex factors.relationship.The kernel function is introduced into the model,and the high-dimensional feature problem of the rollback springback is mapped to the Hilbert space analytical rebound relationship.Thirdly,based on the dynamic change of the curvature of the variable curvature forming process,a variable curvature springback analysis method based on bidirectional long-short-term memory neural network(BiLSTM)algorithm is proposed based on the study of the fixed-curvature springback relationship.Process loop iteration analysis,forgetting and passing,and optimizing the fitting of the variable curvature rebound relationship.Finally,based on the three-roller symmetrical bending machine,the analysis model of the 6061 aluminum alloy profile bending springback is carried out.The experimental data analysis is used to establish the springback analytical relationship under different curvature distributions,and the springback is predicted for different input curvatures.The effectiveness and feasibility of the proposed machine learning-based bending rebound prediction method are verified.