基于BP神经网络的DP600点焊熔核参数建模

以焊接电压、焊接电流构造输入向量,熔核直径、热影响区外径和焊接区焊后厚度为输出量,建立DP600高强钢电阻点焊的熔核参数模型。推导了梯度下降法、动量梯度法和共轭梯度法三种权值算法,并用实际试验数据对模型进行训练和预测。结果表明,共轭梯度法训练后的预测结果误差率最低,所有参数的误差在8%内,平均误差在4%内,可用于在线检测来提高产品质量。

Modeling of the DP600 resistance spot welding nugget parameter based on BP neural network

The input vector is constructed by welding voltage and welding current,nugget diameter,heat-affected zone diameter and thickness of weld zone after welding as output,establish DP600 high strength steel resistance spot welding nugget diameter model.Deduce gradient descent method,gradient descent method with momentum and conjugate gradient training method,and use actual test data to train and predict.The result shows that the model which uses conjugate gradient training method has smallest error rate,the overall error of parameters within 8％,the average error is 4％,which can be applied to quality control.