Research On Forming Accuracy And Springback Model Of BP Neural Network For The Aluminum Parts Of Three-dimensional Stretch-bending

At present,aluminum parts of three-dimensional stretch-bending are widely used in automobiles,high-speed rail,aircraft and other fields as lightweight structural members.With the rapid development of China's modern industry,the demand for aluminum parts of three-dimensional stretch-bending is increasing.However,the conventional stretch-bending process generally adopts an integral die.It can only be bent and formed on one plane of curvature.The multi-point three-dimensional stretch-bending forming process is a new forming technology that combines the traditional bending process with the computer.So it can bend the profile into the aluminum parts of three-dimensional stretch-bending without unloading the profile.In the three-dimensional stretch-bending process,since the profile needs to be stretched and bent on different planes,the forming accuracy of the parts cannot be accurately controlled.At the same time,it is easy to produce defects that affect the forming accuracy of the parts,such as springback,wrinkling,cross-sectional distortion,etc.In order to ensure the forming precision of aluminum parts of three-dimension stretch-bending,it is necessary to select and optimize the parameters in the three-dimensional stretch-bending process,and to reasonably estimate and predict the springback of the aluminum parts of three-dimensional stretch-bending.In this paper,firstly,the finite element model was established based on the characteristics of finite element theory and three-dimensional stretch-bending process.Then,the front frame of the high-speed train head was carried out as the research object for three-dimensional stretch-bending forming.The ABAQUS finite element simulation software was used to simulate the stretch-bending process of the profile.The paper studied the influence of the distribution density of die units on the forming accuracy of complex section profiles in three-dimensional stretch-bending process.Secondly,the springback could affect the forming accuracy of the parts,because the profile could produce the springback after three-dimensional stretch-bending forming.Therefore,the shape error of the formed part and the target part was reduced by using the method of gradually adjusting die units surface.The result showed that under the premise that other parameters remain unchanged,with the increase of the number of die units,the forming precision of the parts could be improved,but the increase of the number of die units could also increase the manufacturing cost.In this paper,the number of die units was 25,which was the most suitable three-dimensional stretch-bending process parameters for the high-speed train head frame structure.At the same time,in the finite element simulation,the method of gradually adjusting die units surface could reduce the maximum shape error of the formed part from 18.82 mm to 2.46 mm.In the actual stretch-bending tests,the maximum shape error was reduced from 27.26 mm to 6.03 mm.The result showed that the finite element simulation could effectively predict the springback of the profile.The method of gradually adjusting die units surface could effectively improve the forming accuracy of the parts.So this can provide guidance for the determination and optimization of three-dimensional stretch-bending forming parameters in the future.In addition,this paper also took high-speed train head structure as an example.Based on the design of orthogonal experiment,the ABAQUS software was used to simulate the horizontal springback along the x-y plane,the vertical springback along the x-z plane,and the total springback.The range analysis of orthogonal test showed that the material parameters had the greatest influence on the springback of aluminum parts.At the same time,the influence of each parameter on the springback of aluminum parts was analyzed.So the optimal factor level was obtained.The combination was: 6063 aluminum alloy material,the horizontal bending angle was 14°,the vertical bending angle was 14°,the number of die units was 10,and the friction coefficient was 0.15.Finally,this paper established the BP neural network springback prediction model of aluminum parts of three-dimensional stretch-bending through the data obtained by orthogonal test.The springback of the parts was predicted by BP neural network.At the same time,comparing the predicted results,finite element simulation and experimental values,it was found that the error was within 11%,which indicated that BP neural network could effectively predict the amount of springback.This provided an idea for predicting the springback during the three-dimensional stretch-bending forming process.