A short arc machining technique belongs to the special processing industry of EDM technology category. It is especially suitable for super hard, super strength, high toughness of difficult machining in efficient processing.However, the technical charac-teristics of the workpiece surface (surface modification, hardness, residual stress, surface layer defects, etc.) need to be further studied.In order to obtain the good processing results of the short arc milling process, a traditional BP algorithm and a improved Levenberg-Marquardt (L-M) algorithm are introduced to build the model of the surface quality of the short arc milling process.By analyzing the factors of influencing on the surface quality,the discharge voltage,frequency,the pressure,pulse time are selected as the input of the model in this paper.Meanwhile,the surface roughness,metamorphic layer thickness,workpiece material removal rate are selected as output,comparing the prediction accuracy of two models.The results show that using BP algorithm based on the im-proved L-M neural network,the average prediction errors of surface roughness,metamorphic layer thickness,workpiece material re-moval rate are 2.9%,9.4% and 4.6%,respectively,which are lower than that of the traditional neural network.Comparing with the traditional BP neural network,the improved LM-BP neural network model can improve the prediction accuracy which can be used to optimize the process parameters in practical engineering.%短电弧铣削加工技术属于特种加工行业中电加工的技术范畴，尤其适用于特硬、超强、高韧性等难加工材料的高效加工。但工件加工表面的技术特性（表面变质层、硬度、残余应力、表面层缺陷等）还有待于深入研究。为获得短电弧铣削加工良好的工艺效果，引入传统BP算法和Levenberg-Marquardt（简称L-M）算法，构建短电弧铣削加工表面质量模型。通过分析表面质量的影响因素，选取放电电压、频率、气压、脉冲时间为模型的输入，表面粗糙度、变质层厚度、工件材料去除率为输出，比较两种模型的预测精度。结果表明，基于L-M算法的BP神经网络对表面粗糙度、变质层厚度、材料去除率的平均预测误差分别为2．9％、9．4％、4．6％，低于传统的BP神经网络。相比传统的BP神经网络，改进的LM-BP 神经网络模型提高了预测精度，实际工程中可用于优化工艺参数。
展开▼
