基于BP人工神经网络的板料自冲铆接的底切量仿真预测

Simulation prediction on under-cut for self-piercing riveting of sheet based on BP artificial neural network

以Al Mg3铝合金板料的自冲铆接作为研究对象,利用DEFORM-2D有限元软件与正交试验,对两层2 mm的Al Mg3铝合金板料的自冲铆接过程进行有限元仿真,基于BP人工神经网络及仿真数据,对板料自冲铆接的成形质量进行仿真预测。以模具深度、模具凸台高度、模具宽度以及铆接速度为输入层,将底切量作为输出层,建立了4-11-1的3层BP人工神经网络。以仿真结果为样本进行反复的训练,得到BP人工神经网络的预测值,与有限元仿真值相比两者的最大误差为0. 97%。此外,借助自冲铆接设备及安装模具,进行自冲铆接试验,试验值与BP人工神经网络预测值之间的相对误差为6. 57%,验证了BP人工神经网络应用于板料自冲铆接成形质量预测的正确性与可靠性,为企业的实际生产提供重要的参考。

For the self-piercing riveting of AlMg3 aluminum alloy sheet,the self-piercing riveting process of two layers of AlMg3 aluminum alloy sheet with thickness of 2 mm was simulated by DEFORM-2D and orthogonal experiment,and the forming quality of self-piercing riveting for sheet was predicted based on BP artifical neural network and simulation results.Then,the die depth,height of die boss,die width and riveting speed were taken as the input layer,the under-cut was taken as the output layer,and the BP artificial neural network with three layers of 4-11-1 was established.The predictive value of BP artifical artificial neural network were obtained by using the simulation results as samples for repeated training.Compared with the finite element simulation results,the maximum error between the two methods was 0.97%.In addition,the self-piercing riveting tests were conducted by the self-piercing riveting equipment and die,and the relative error between prediction value of BP artifical neural network and experimental value was 6.57%.Therefore,the correctness and reliability of BP artifical neural network applied to the prediction on the forming quality of self-piercing riveting for sheet are verified which is an important reference on the actual production of enterprises.