以轿车用低碳钢薄板为实验样品，分析了脉冲激光焊接产生的主要变形方式。利用径向基函数神经网络对薄板焊接产生的横向收缩变形和横向弯曲变形进行预测。采用响应面法对实验参数进行优化设计。将脉冲频率、脉宽、聚焦镜焦距、离焦量、工件移动速度、保护气体种类、工件温度波动和光功率波动作为神经网络输入，提高了焊接变形预测的准确度。通过对比6种神经网络对薄板焊接变形预测的结果得出了最佳的网络结构。实验证明该神经网络对薄板焊接产生的变形有较高的预测准确度。

A set of mild steel thin plate specimens used for automotive industry are used as laboratory samples. Different types of distortions are analyzed. Radial basis function neural network (RBFN) models have been developed to predict transverse shrinkage and longitudinal bending distortion of welded plates. Response surface method is used to set up the experimental parameters matrix. Pulse frequency, pulse width, focal distance, defocus distance, moving speed of welded plates, shielded gas, workpiece temperature fluctuation and laser power fluctuation are used as input variables of these models to increase the prediction accuracy. Six different types of RBFN models have been developed to predict the distortion of welded plates. The best one is selected from them and resulted in better output prediction.