基于人工神经网络的拼焊板成形极限图预测

Prediction of forming limit diagram for tailor welded blanks based on artificial neural network

成形极限图(FLD)是评价金属板材成形能力的重要工具。为快速的建立拼焊板(TWB)成形极限图,建立基于人工神经网络(ANN)拼焊板FLD的预测模型。采用试验设计和有限元法获得训练样本,L-M算法对样本数据进行训练,建立了FLD预测模型并与物理试验结果对比。基于预测模型,分析了摩擦系数对拼焊板最小极限应变的影响。结果表明,基于ANN预测的拼焊板FLD与试验结果吻合,主应变的相对误差最大为8.71%。摩擦系数f对最小极限应变影响较大,f从0增大到0.12时,最小极限应变先增大后减小,并在摩擦系数f=0.06附近出现极小值。

The forming limit diagram （FLD） is a very effective tool to evaluate the formability of the sheet metal . To quickly create the FLD for tailor welded blank （TWB） ,we proposed a prediction model based on an artificial neural network .The design of experiment and finite element method were used to gain the training samples , which were trained by the Levenberg-Marquardt （L-M ） algorithm . The prediction model of FLD was built and validated using the experimental data . Furthermore , the effect of the friction coefficient on the minimum ultimate strain was analyzed by the presented prediction model . The results show the FLD by the prediction model is consistent with that from the experiment data ,and the maximum relative error between the experiments and the predictions is 8.71% . The friction coefficient has a marked effect on the limit strain of TWB . The least limit strain first increases then decreases with the increasing of the friction coefficient form 0 to 0.12 , reaching the minimal value when the friction coefficient is near 0.06 .