The surface quality and mechanical properties of AlSi10Mg parts manufactured using laser powder bed fusion (LPBF) rely heavily on the process parameters. Such needs to be tuned to improve the overall reliability of the LPBF devices. To model and forecast the performance of the process, an artificial neural network (ANN) model with feedforward backpropagation algorithm was created and tested. Wherein, laser power (LP), point distance (PD) and exposure time (ET), and surface roughness (S_a), Vickers microhardness (H_v) were introduced as inputs and outputs of the network, respectively. This ANN tool will also be beneficial for dealing with the optimization of process parameters. Comparing the consequential ANN networks, it was discovered that, ANN architecture using "trainlm," "tansig"–"Logsig," as the training algorithm and transfer functions in the hidden and output layers, respectively, with eight hidden neurons and 400 training epoch, produces the best simulation result with the lowest mean square error (MSE). A model is developed to predict the optimal process parameters for producing AlSi10Mg components with desired surface roughness and Vickers microhardness.

使用激光粉末床熔合 (LPBF) 制造的 AlSi10Mg 零件的表面质量和机械性能在很大程度上取决于工艺参数。需要对此进行调整以提高 LPBF 器件的整体可靠性。为了对过程的性能进行建模和预测，创建并测试了具有前馈反向传播算法的人工神经网络 (ANN) 模型。其中，激光功率（LP）、点距（PD）和曝光时间（ET）、表面粗糙度（S _a）、维氏显微硬度（H _v) 分别作为网络的输入和输出引入。这个人工神经网络工具也将有利于处理工艺参数的优化。比较结果式 ANN 网络，发现 ANN 架构使用“trainlm”、“tansig”-“Logsig”作为训练算法，在隐藏层和输出层分别使用传递函数，有 8 个隐藏神经元和 400 个训练 epoch ，产生具有最低均方误差 (MSE) 的最佳模拟结果。开发了一个模型来预测生产具有所需表面粗糙度和维氏显微硬度的 AlSi10Mg 部件的最佳工艺参数。