• CN:11-2187/TH
  • ISSN:0577-6686

›› 2011, Vol. 47 ›› Issue (22): 87-92.

• 论文 • 上一篇    下一篇

连续顺序电阻点焊分流率的反向传播神经网络预测

张勇;汪帅兵;谢红霞;马铁军   

  1. 西北工业大学陕西省摩擦焊接重点实验室
  • 发布日期:2011-11-20

Prediction of Shunt Rate for Consecutive Resistance Spot Welding by Back Propagation Neural Network

ZHANG Yong;WANG Shuaibing;XIE Hongxia;MA Tiejun   

  1. Shaanxi Key Laboratory of Friction Welding Technologies, Northwestern Polytechnical University
  • Published:2011-11-20

摘要: 建立以材料电阻率、板厚、焊点间距为输入空间,分流率为输出空间的连续顺序电阻点焊分流率的3层误差反向传播(Back propagation,BP)神经网络预测模型。依据电阻点焊恒流控制的特点和点焊过程的电阻变化规律建立分流率的理论计算模型,由该模型所得数据作为样本对网络进行训练和检验。对2.0 mm 厚度的20钢及1.5 mm、1.0 mm 厚度的10钢等厚度点焊的分流进行预测,预测相对误差最大值分别为2.83%、1.77%和3.67%。验证试验结果表明,应用建立的神经网络的预测结果进行分流补偿后,在焊点间距为30 mm和50 mm时,第2~5焊点熔核直径相对第1点的平均误差,1.0 mm 厚度的10钢分别约为2.86%和3.99%,2.0 mm厚度的20钢分别约为2.46%和3.58%。证明采用建立的BP神经网络分流预测模型,对10钢和20钢恒流控制连续顺序点焊时的分流进行预测是可行的。

关键词: 电阻点焊, 分流, 恒流控制, 神经网络, 预测模型

Abstract: A prediction model for shunt rate of resistance spot welding in constant current control is developed by a three-layer back propagation(BP) neural network, where the material resistivity, material thickness and dot pitch are the input of the model, and the shunt rate is the output. According to the characteristics of the constant current control and the resistance change during the process of spot welding, a compensation model to calculate the shunt of the consecutive spot welding is established. The training and test samples of the artificial neural network(ANN) from this compensation model are used to train the network and verify the performance of the trained network. By using the trained network, the shunt rates are predicted for 20 mild steel of 2.0 mm thickness and 10 mild steel of 1.5 mm and 1.0 mm thickness respectively. The biggest relative prediction error is 2.83%, 1.77% and 3.67% respectively. The experiments with current compensation using the shunt rate of prediction on the resistance spot welding machine are done at different dot pitch. When the dot pitch is 30 mm, as for mild steel of 1mm thickness, the average relative error between the nugget diameters of the second to the fifth spot welded and the one of the first spot welded is 2.86% and as for mild steel of 2.0 mm thickness it is 2.46%. While the dot pitch is 50 mm, for the former the average relative error is 3.99%, the latter it is 3.58%. The test results indicated the proposed BP neural network prediction model has good performance to predict shunt rate for consecutive spot welding of the 10 and 20 mild steel.

Key words: Constant current control, Neural network, Prediction model, Resistance spot welding, Shunt

中图分类号: