Research On Springback Of Automobile Girder In Stamping Forming Based On Deep Learning

Beam parts are one of the most important structural parts in body parts.Its existence ensures the carrying capacity of the body in white and the rigidity of the body itself.As the automobile lightweight technology becoming more and more mature,advanced high-strength steel has widely used for making automobile beam-shaped structural parts,and the springback phenomenon of automobile beam-like structural parts have serious springback problem during manufacturing process.It has become more and more important in automotive industry to find the solution of springback problem.As far as we can know,it is the solution of springback problem to predict the springback accurately.This paper study a specific beam part,based on the theory of springback two-dimensional section method and finite element simulation technology,uses deep learning network model and image recognition technology,integrates sheet material factors and stamping process factors,and studies beam structures The springback problem at different sections of parts improves the quality of those beam parts.The main work of this paper are mentioned as follow:1.Two-dimensional discretization and numerical conversion of three-dimensional beam parts.Based on the two-dimensional springback theory,this paper simplifies the three-dimensional springback problem into two-dimensional springback problem,and one three-dimensional beam section are separated to several two-dimensional sections.The double-plane projection method and the image binarization method are used to convert the section curve of the beam parts into Two-channel image data that the neural network model can recognize.2.Based on Latin hypercube sampling to study the rebound size under different material parameters and process parameters.Eight parameters like mold gap,friction coefficient,blank holder force,material thickness,material strengthening factor,material hardening coefficient,initial yield strain,and anisotropy coefficient are used as stamping process parameters and sheet material parameter variables for Latin hypercube sampling,Use Dynaform software to perform CAE rebound simulation,obtain training samples,and obtain subsequent training samples of artificial intelligence networks.3.Research on the springback of beam parts stamping based on deep learning network.This article suggest to use deep learning image recognition technology to predict the springback of beam parts.First,optimize the model parameters based on the LeNet-5 deep learning model,the AlexNet deep learning model,and the NiN deep learning model,and use the moldel to recognize the cross-sectional image,then make the network model link with a fully connected neural network and sheet material factors and stamping Process factors are coupled,and finally the springback algorithm model of beam parts in this paper is obtained.4.Rebound type classification and algorithm model verification based on Gaussian mixture clustering.Taking an automobile beam-like structural part as the research object,the rebound samples are obtained through CAE rebound simulation.According to the rebound value of the samples,the rebound samples are classified into large rebound,medium rebound,and medium rebound using the method of Gaussian mixture classification.Three types of big rebound.Each type of rebound sample was trained and optimized through the Le Net-5 deep learning model,NiN deep learning model,AlexNet deep learning model,and those models’ performance was compared through the sample validation set.The results of confirmatory experiments prove that the accuracy of the deep learning model based on AlexNet is the best,and the algorithm is also more robust than the other two.