激光选区熔化组织分析及人工神经网络力学性能预测

采用激光选区熔化技术（selective laser melting，SLM）制备18Ni300时效模具钢.通过扫描电子显微镜（scanning electron microscope，SEM），研究试样的枝晶生长取向和凝固组织状态.利用人工神经网络对激光功率、扫描速度和扫描间距进行重要性分析，同时采用BP （back propagation，BP）神经网络以工艺参数为特征对材料的抗拉强度进行预测，应用遗传算法（genetic algorithm，GA）对神经网络权值和阈值进行寻优.结果表明，试样组织主要呈树枝柱状生长，外延生长明显，组织取向主要取决于熔池底部的凝固条件；熔池顶部易发生柱状晶向等轴晶转变（columnar to equiaxed transition，CET），可以通过调节工艺参数来控制转变区的大小；热毛细对流导致熔池其它区域也出现枝向转变区.人工神经网络重要性预测结果由大到小的顺序是激光功率、扫描速度、扫描间距，BP拟合结果与实际结果较为接近，决定系数R2=0.73.

Microstructure based on selective laser melting and mechanical properties prediction through artificial neural net

Selective laser melting has been applied to fabricate 18Ni300. SEM is used to observe dendritic growth orientation and solidification structure. Artificial neural network is applied to rank the respective importance of laser power, scanning speed and scanning space for mechanical properties, while BP neural net with improved weight by genetic algorithm is applied to the prediction of tensile property. Results show that the main structure of the specimen is columnar dendritic crystals with significant epitaxial growth. The orientation of the growth is determined by the solidification condition at the bottom of the molten pool. CET can easily take place on the top of the melting pool. Meanwhile, there is transition zone in other places contributed by the thermo capillary convection. The result of the importance prediction by artificial neural network shows:They order from high to low is laser power, scanning speed and scanning space.Since the prediction results agree with the actual ones, BP neural net can effectively predict actual results. The determination coefficient R²=0.73.