中国激光, 2011, 38 (8): 0803004, 网络出版: 2011-07-04
基于人工神经网络的激光立体成形件成形表面质量预测
Component′s Surface Quality Predictions by Laser Rapid Forming Based on Artificial Neural Networks
光学制造 激光立体成形 人工神经网络 表面质量 工艺参数 optical fabrication laser solid forming artificial neural network surface quality process parameters
摘要
通过建立单道折线扫描数学模型,推导出了折线扫描拐角处重叠区域数学描述,从理论上对激光立体成形(LSF)表面质量的影响因素进行了分析,并得出影响折线扫描路径试样表面质量的主要因素是折线角度和扫描速度的结论。建立了适用于激光立体成形件表面质量预测的人工神经网络(ANN)模型,以激光立体成形过程中扫描速度和折线的角度为模型输入,输出成形件表面质量评估参数。经过实验数据训练后的神经网络模型可以实现对不同扫描速度及不同扫描角度成形件表面质量的预测,网络预测值和试验测得值之间的均方差(MSE)小于0.01。
Abstract
A single-channel polyline scanning mathematical model is established, and the mathematical description of surface quality is analyzed theoretically. It is drew the conclusion that the angle degree and the scanning speed are the main important process parameters for part surface quality. The artificial neutral networks (ANN) model is established to predict part surface quality based on laser solid forming (LSF). The input of this model is the angle degree of corner joint and the scanning speed which is the most important factor in LSF process. The output of this model is the data of surface characterization which is the difference between corner joint height and layers′ height. The ANN model could predict parts′ surface quality data under different corner joint′ angle degree conditions after trainied by experimental data. The mean squared error (MSE) is less than 0.01 between prediction data and experimental data.
杨东辉, 马良, 黄卫东. 基于人工神经网络的激光立体成形件成形表面质量预测[J]. 中国激光, 2011, 38(8): 0803004. Yang Donghui, Ma Liang, Huang Weidong. Component′s Surface Quality Predictions by Laser Rapid Forming Based on Artificial Neural Networks[J]. Chinese Journal of Lasers, 2011, 38(8): 0803004.