基于BP神经网络的SLS烧结件质量的预测

选择性激光烧结的烧结件质量预测是一个多变量、非线性的问题,采用传统的方法很难得到满意的结果。采用BP神经网络模型,在数值模拟取样的基础上,建立了烧结件质量的神经网络预测模型。该模型确定了工艺参数激光功率P、扫描速度v和预热温度T0与烧结宽度和烧结深度的关系。其预测结果与数值模拟结果相一致,说明该神经网络模型能定量地反映出工艺参数与烧结件质量之间的关系,据此可合理选择加工工艺参数。     

基于支持向量回归机的SLS制件收缩率预测与分析

结合选择性激光烧结过程的工艺特点,将收缩率作为精度的评价标准,选择预热温度、激光功率、扫描速度、铺粉层厚、支撑厚度五个工艺参数设计正交试验以获得训练样本。建立了基于支持向量回归机的SLS制件收缩率回归-预测模型,以预测不同参数组合下的制件收缩率,确定了一定参数范围下的工艺参数优化组合方案,并定性分析其参数组合的合理性。     

基于SOA-LSSVM的SLS成形工艺参数优化研究

为提高选择性激光烧结( SLS)成形精度,解决工艺参数优化试验成本高等问题,选择激光功率、预热温度、扫描速度、扫描间距以及分层厚度5个工艺参数设计正交试验以获得样本数据并建立统一目标函数。采用人群搜索算法(SOA)优化最小二乘支持向量机(LSSVM), 建立基于SOA-LSSVM的SLS成形件精度预测模型;预测不同工艺参数组合下制件的统一性能,并与采用传统BP神经网络和LSSVM模型获得的预测结果进行对比。结果表明：SOA-LSSVM模型针对小样本预测问题具有良好的泛化能力,预测值与实际值的最大相对误差仅为1.11%,可为SLS加工参数组合的选择提供参考。     

选择性激光烧结的翘曲变形与扫描方式的研究

在研究选择性激光烧结工艺(Selective Laser Sintcfing,简称SLS)成形误差的基础上,分析造成SLS成形件翘曲的原因,构建了一种新的翘曲变形模型.并以此模型作为基础,提出一种新的光栅扫描方式.通过试验验证,此扫描方式可以有效地降低SLS成形件的翘曲变形,提SLS成形件精度.     

工艺参数对316不锈钢粉末激光烧结温度场的影响

以有限元分析软件ANSYS为平台,对316不锈钢粉末激光烧结温度场分布进行了数值模拟.在考虑了材料的热物性参数随温度变化以及相变潜热等非线性情况下,建立了选择性激光烧结(SLS)三维有限元模型,利用ANSYS参数化设计语言APDL控制激光热源的热流密度、移动速度以及扫描路径,研究了工艺参数(激光功率、扫描速度、预热温度)对316不锈钢金属粉末成型过程中熔池及温度场分布产生的影响.模拟结果与前人文献实验结果相吻合,表明可以利用本模型对工艺参数进行优化,为实验工艺参数选取提供了理论依据.     

基于PSO-BP算法的3D打印制件精度预测模型

针对选择性激光烧结(SLS)中制件精度和工艺参数难以选择的问题以及BP神经网络本身缺陷,提出一种利用粒子群算法优化的BP神经网络建立SLS烧结件精度预测模型的方法。首先根据SLS成型工艺的特点,分析影响成型件精度的因素,通过实验获得不同激光功率、扫描速度、扫描间距和分层厚度条件下多组成型件精度数据,并采用多目标函数优化的单目标化思想优化目标函数,然后通过粒子群算法优化BP神经网络。用优化后的最优解作为BP神经网络算法的初始权值和阈值,利用MATLAB建立优化后的BP神经网络预测模型,对优化后的精度函数模型进行预测分析,并与传统BP神经网络获得的预测结果进行对比。结果表明:粒子群优化的神经网络模型具有良好的全局搜索能力和收敛性,精度预测更加准确,对SLS打印制件具有一定的指导作用。     

基于SLS技术的精铸蜡粉烧结件翘曲变形研究

对翘曲变形产生的根本原因进行了分析。通过大量的精铸蜡粉烧结实验,研究了激光功率、扫描速度、预热温度和铺粉厚度对翘曲变形的影响,叙述了制件翘曲程度和工艺参数之间的关系。采用正交试验方法研究了精铸蜡粉选择性激光烧结成型工艺,得到了减小烧结前期翘曲变形的最佳工艺参数。结果表明,烧结前期最佳工艺参数为预热温度50℃、扫描速度1200 mm/s,激光功率18 W、铺粉层厚0.15 mm。     

基于广义回归神经网络的选区激光烧结成形收缩率研究

针对激光烧结增材制造过程中出现的几何尺寸误差问题,采用正交实验与测量的方法获得训练样本,依据广义回归神经网络,建立了选择性激光烧结过程中工艺参数与成形收缩率之间的定量模型,以预测收缩率。定性分析了预热温度与支撑厚度对收缩率的影响,得到了各因素对收缩率影响的权重,并分析了主要因素间的交互作用。通过定性分析与定量预测,可为烧结过程中优化控制收缩提供一个新思路。     

选择性激光烧结工艺参数对影响铸造砂型精度的优化设计

阐述了选择性激光烧结(SLS)的成型原理,并将此成型方式运用于直接砂型制作。通过用自主研制的快速成型机进行烧结试验,采用正交试验和方差分析,对影响铸造砂型精度的工艺参数进行了优化设计,得到激光功率、扫描速度、扫描间距及铺粉厚度工艺参数的最佳组合,可为基于选择性激光烧结铸造砂型的制造提供指导和参考。     

基于BP神经网络的覆膜砂选择性激光烧结件精度预测

采用正交试验与反向传播(BP)神经网络结合的方法对覆膜砂选择性激光烧结的工艺进行优化。运用BP神经网络建立尺寸精度(xy向收缩)与选择性激光烧结工艺参数之间的预测模型,利用正交试验样本对所建立的神经网络进行训练,形成输入与输出之间的高度映射关系,并验证神经网络模型的可行性。在此基础上,预测各因素在不同水平下覆膜砂选择性激光烧结件的最优精度。     

磨损失效零件激光增材再制造性神经网络量化评价方法EI北大核心CSCD

针对现有激光增材再制造性评价方法主观性强、效率低等不足,提出一种失效数据驱动的磨损失效退役零件增材再制造性神经网络量化评价方法。根据磨损失效退役零件的激光增材再制造修复难易程度与失效模式相关的特点,以修复路径规划可行性、运动轨迹规划可行性、激光熔覆材料选择、再制造时长、再制造经济性为评价指标,构建磨损失效退役零件激光增材再制造性层次评价模型;通过退役零件失效区域的修复路径规划、修复设备运动轨迹模拟和碰撞检测进行修复路径可行性和运动轨迹可行性指标量化,并定义再制造时长、经济性指标的量化公式。以多个同类零件为对象,通过上述量化评价方法构建样本空间,基于神经网络训练获得再制造性神经网络量化评价模型,实现同类零件的快速激光增材再制造性评估。最后,以碎煤机端盘为例对所提方法进行验证,结果表明了该方法的可行性与有效性。根据失效数据与退役零件再制造性之间的映射关系,可实现磨损失效零件增材再制造性的快速量化评价,为其工程应用提供技术支撑。     

激光熔覆镍基熔覆层截面形貌预测

利用BP神经网络建立激光熔覆关键工艺参数(激光功率、扫描速度、送粉电压、送粉载气流量)与熔覆层截面形貌(熔宽、余高)的预测模型。以激光熔覆工艺参数为输入,熔覆层的截面形貌为输出,利用工艺试验数据对网络进行训练,实现对输入和输出的高度映射。结果表明,BP神经网络可以较好地对熔覆层形貌进行预测,同时双隐藏层BP神经网络模型预测结果误差波动更小,表现出优良的稳定性,最大预测误差相比单隐藏层神经网络大大降低。     

大功率碟形激光焊支持向量回归熔宽预测算法

大功率碟形激光焊接作为一种重要的新型激光加工技术在制造业领域得到日益广泛的应用.针对焊接过程多变量强耦合性以及反馈的实时性要求,提出了用支持向量机(SVR)对焊接过程中熔宽变化量进行预测的新方法,并根据焊接试验数据对其性能进行验证.同时分析对比了支持向量机与BP神经网络的预测效果.结果表明,BP神经网络和支持向量机的训练和单步预测效果良好,均可以应用在大功率碟形激光焊接过程中,但SVR模型要更加适应于大功率碟形激光焊接过程的样本训练和预测.当采样点数N值取10时预测效果最优.     

Improving accuracy through shrinkage modelling by using Taguchi method in selective laser sintering

Selective laser sintering (SLS) is a powder-based rapid prototyping process in which parts are built by selective sintering of layers of powder using CO₂ laser. Nowadays, SLS is emerging as a rapid manufacturing technique, which produces functional parts in small batches, particularly in aerospace application and rapid tooling. Therefore, SLS prototypes should have high accuracy in order to satisfy functional requirements. Shrinkage is one of the major factors which influence the accuracy of the SLS parts. To compensate for shrinkage, the material shrinkage coefficient or scaling factor is to be calculated in each direction and is to be applied to STL file. The amount of shrinkage encountered is found to be governed by the process parameters during processing and cannot be kept constant as it is a usual practice in today's SLS technology. In the present work, the relationship between shrinkage and the various process parameters namely laser power, beam speed, hatch spacing, part bed temperature and scan length in SLS have been investigated. Cuboids with suitable dimensions are fabricated rather than fabricating long parts along X, Y and Z directions in order to study shrinkage as it is expected that the shrinkage along X, Y and Z direction is not independent. Optimum shrinkage conditions are obtained by maximizing signal-to-noise (S/N) ratio and analysis of variance (ANOVA) is used to understand the significance of process variables affecting shrinkage. Laser power and scan length are found to be most significant process variables influencing shrinkage in X- direction. Along Y-direction laser power and beam speed are the significant parameters and along Z-direction beam speed, hatch spacing and part bed temperature are found to be most significant factors influencing shrinkage. Empirical models for predicting shrinkage along X, Y and Z directions are derived using regression. Obtained results are validated and they are found in good agreement with experiments. One case study of bench marking part is also presented to show that shrinkage model developed in the present work confine more accurate parts.     

焊缝偏差RBF神经网络预测算法

以10 kW大功率光纤激光焊接304奥氏体不锈钢板为试验对象,研究一种焊缝偏差预测算法.利用红外摄像机摄取焊接过程中的熔池红外图像,提取匙孔质心、匙孔形状参数和热堆积效应参数等反映激光束与焊缝位置偏差的特征量作为径向基函数RBF神经网络预测模型的输入量,建立焊缝偏差RBF神经网络预测模型.选择焊缝偏差特征量作为训练样本并对预测模型进行训练,建立焊缝偏差预测模型.结果表明,该模型能够对大功率光纤激光焊接过程中的激光束与焊缝位置之间的偏差进行有效预测.     

基于选择性激光烧结技术的快速铸造

介绍了用选择性激光烧结（ＳＬＳ）快速成型技术直接制造覆膜砂铸型（芯）的特点及工艺过程。结合铸件生产,分析研究了ＳＬＳ铸型（芯）的工艺设计、三维实体造型及铸型后处理等过程中所遇到的问题。     

Al-Cu-Mg(2519)合金高温变形本构关系的神经网络模型

2519铝合金是一种新型的装甲材料。变形时,各热力学参数之间存在着非常复杂的非线性关系。本文采用Gleeble 1500热模拟机上圆柱体轴对称高温压缩试验数据建立了该合金本构关系神经网络模型。利用所建立的网络模型对其他一些热力学状态下材料的流变应力进行了预测,发现预测数据与实验数据吻合良好(总拟合度为2．6％),表明该本构关系神经网络模型有较高的预测精度。     

An application of neural network solutions to laser assisted paint stripping process of hybrid epoxy-polyester coatings on aluminum substrates

A neural network approach is used to model a paint stripping process performed using a 1.5 kW continuous diode laser source on an aluminum substrate coated with approximately 80 μm of a hybrid epoxy-polyester resin. Two different coating colors, namely RAL 8087 (dark) and RAL 1013 (bright), are examined in order to analyze the influence of the difference in absorption of the laser energy by the surface on process.

Experimental analysis was performed first in order to find the trend of paint stripping factor (PSF) according to leading process parameters such as laser power, scan speed, defocus length, and number of passes. A statistical approach is used to discuss the experimental data found. Two neural network models are investigated, namely Multi-Layer Perceptron (MLP) and Radial Basic Function (RBF), with MLP being more reliable and effective in modeling experimental results.

A sensitivity analysis on the MLP model is used to show the significance of all the input data employed. As a result of sensitivity analysis, a check between experimental and calculated trends for each investigated variable was performed, which revealed an appreciable fit between data displayed. Following this, a regression model to predict the trend of PSF according to laser fluence was developed. Finally, the regression model and the MLP model were compared and showed the high degree of accuracy of neural network solution in predicting the experimental results.     

Weld pool shape prediction in plasma augmented laser welded steel

Abstract

A welding process that combined plasma arc welding with laser welding was used to make autogenous bead on plate welds on a sheet stock of a carbon steel. A wide range of welding parameters (arc current, laser power, weld speed) was employed. The experimental weld pool shapes were analysed and the data were used to train a neural network to predict weld pool shape as a function of process conditions. The predictions of the neural network model showed excellent agreement with the experimental results, indicating that a neural network model is a viable means for predicting weld pool shape. Using the model, a parametric study was carried out to examine the influence of process conditions on the final weld pool profile.