基于神经网络的焊机参数预测方法

针对脉冲MIG焊参数众多，不易调节的特点，提出了一种基于神经网络的焊机参数预测方法. 该方法采用LM(levenberg-marquarlt)算法建立了焊机参数的BP(back propagation)神经网络模型，充分利用已知的理想数据对网络进行训练，实现了焊接过程中任一给定焊接电流状态下焊机输出参数的预测；利用焊接参数的预测值分别对单、双脉冲MIG焊进行了试焊. 结果表明，基于神经网络的焊机参数预测方法精度较高，焊接过程稳定，焊缝成形美观，能够实现良好的一元化调节.

Prediction method of welding machine parameters based on neural network

In view of the fact that pulse MIG welding has many parameters and is difficult to adjust, a welding parameter prediction method based on neural network is proposed. This method, having established BP neural network model of welding parameters by adopting LM(levenberg-marquarlt) algorithm, and making full use of the known data to train the network, have realized the prediction of the output parameters in any given welding current state, and then conduct test weld on single and double pulse MIG welding respectively by using the predicted values of welding parameters. The results show that the prediction method of welding parameters based on neural network is of high accuracy, that the welding process is stable, and that the seams can be well-formed, thus achieving a good unified adjustment.