基于机器学习的前轴辊锻工艺参数预测

Prediction on roll forging process parameters of front axle based on machine learning

使用基于混合核函数的最小二乘支持向量机算法来进行前轴第1道次辊锻工艺参数的预测,构造混合函数以提高预测模型的预测精度。对工艺参数预测模型进行实验验证,结果表明,与基于单独RBF核的LS-SVM算法相比,混合核函数LSSVM算法构建的预测模型具有更高的预测精度,由3组不同核函数参数构成的预测模型对最大成形载荷及展宽的平均预测精度分别为91.8%和92.1%、89.4%和90.1%以及94.5%和93.2%,并且确定了最优的核函数参数:RBF核函数系数γ=4.52、惩罚系数c=276.4、Polynomial核函数核阶数q=1.21以及混合权重系数a=0.28。通过3组工艺参数实验对所研究的前轴辊锻工艺参数预测方法的可行性进行验证,所使用的预测方法预测得到的最大成形载荷和展宽与实际值误差在10%以内。

The roll forging process parameters of the first pass for front axle were predicted by the least square support vector machine algorithm based on the mixed kernel function,and the mixed function was constructed to improve the prediction accuracy of predicted model.Then,the process parameter predicted model was verified by experiment.The results show that compared with LS-SVM algorithm based on the single RBF kernel,the predicted model constructed by the mixed kernel function LS-SVM algorithm has higher prediction accuracy,and the average prediction accuracies of three groups of predicted models with different kernel function parameters for the maximum forming load and widening are 91.8%and 92.1%,89.4%and 90.1%,94.5%and 93.2%respectively.Furthermore,the optimal kernel function parameters are determined with RBF kernel function coefficientγ=4.52,penalty coefficient c=276.4,Polynomial kernel function kernel order q=1.21 and mixed weight coefficient a=0.28.Finally,the feasibility of prediction method for the studied front axle roll forging process parameters was verified by three groups of process parameter experiments,and the errors between the predicted maximum forming load and widening by the prediction method and the actual values are less than 10%.