中文摘要：

低速单向走丝电火花线切割在钛合金加工领域有着不可替代的作用和地位,但微观放电的复杂性决定了其难以建立有效的表面粗糙度数学模型,同时现有机床系统中并没有针对钛合金材料的加工参数。以钛合金TC4为试验研究对象,采用Design-Expert设计Box-Behnken试验并通过三维轮廓仪和扫描电子显微镜对加工后的表面形貌、功率频谱和重熔层进行分析。观测结果表明：电火花加工表面没有明显纹理,为各向同性,不同于磨削加工表面;当峰值电流为40 A,开路电压为100 V,脉冲宽度为18μs时,裂纹延伸至TC4基体。利用响应曲面法通过模型选择和显著性检验得出三维表面粗糙度的2阶数学模型,能正确地映射出低速单向电火花线切割钛合金的工艺规律。为了提高模型预测精度和泛化能力,引入BP神经网络建立组合模型,试验验证结果表明：样本内相对误差均值由4.33%降低到3.26%,样本外相对误差均值由13.31%降低到8.50%,为电火花加工工艺仿真提供新的方法和途径。

英文摘要：

Low speed wire electrical discharge machining （LS-WEDM） has an irreplaceable role in tita- nium alloy machining, but the complexity of microcosmic discharge makes it difficult to build the surface roughness model. Furthermore, the existing machine systems can not provide any machining parameters of titanium alloy. Box-Behnken experiment of TC4 as research object is designed by Design-Expert soft- ware. The three-dimensional contourgraph and scanning electron microscope are used to analyze the sur- face morphology, power spectrum and re-solidified layer, and the finished surface is isotropic and differ- ent from grinding surface. The observed result shows that the open circuit voltage is 100 V with the peak current of 40 A, and the crack extends to TC4 matrix with the pulse width of 18 μs. A quadratic model of three-dimensional surface roughness is established after model selection and significance test by using re- sponse surface method. In order to improve the prediction accuracy and generalization ability, BP neural network is applied to build a combined model. The verification experiments mean absolute error from 4. 33% to 3.26% for in-sample and from 13.31% show that it can reduce the to 8.50% for out-sample, which provides a new method for electrical discharge machining simulation.