惠记庄,骆伟,阎志强,王俊杰,吕景祥,郭许,张浩.AlSi10Mg选区激光熔化表面粗糙度预测、优化及表面形貌分析[J].表面技术,2024,53(15):129-140, 151.
HUI Jizhuang,LUO Wei,YAN Zhiqiang,WANG Junjie,LYU Jingxiang,GUO Xu,ZHANG Hao.Surface Roughness Prediction, Optimization and Surface Morphology Analysis of AlSi10Mg by Selective Laser Melting[J].Surface Technology,2024,53(15):129-140, 151 |
| AlSi10Mg选区激光熔化表面粗糙度预测、优化及表面形貌分析 |
| Surface Roughness Prediction, Optimization and Surface Morphology Analysis of AlSi10Mg by Selective Laser Melting |
| 投稿时间：2023-10-20 修订日期：2023-12-21 |
| DOI：10.16490/j.cnki.issn.1001-3660.2024.15.012 |
| 中文关键词: 表面粗糙度 选区激光熔化 AlSi10Mg 工艺参数优化 表面形貌 预测模型 |
| 英文关键词:surface roughness selective laser melting AlSi10Mg process parameter optimization surface morphology prediction model |
| 基金项目:陕西省秦创原“科学家+工程师”队伍建设项目(2022KXJ-150);长安大学研究生科研创新实践项目(300103724009) |
|
|
| Author | Institution |
| HUI Jizhuang | School of Construction Machinery, Chang'an University, Xi'an 710064, China |
| LUO Wei | School of Construction Machinery, Chang'an University, Xi'an 710064, China |
| YAN Zhiqiang | School of Construction Machinery, Chang'an University, Xi'an 710064, China |
| WANG Junjie | School of Construction Machinery, Chang'an University, Xi'an 710064, China |
| LYU Jingxiang | School of Construction Machinery, Chang'an University, Xi'an 710064, China |
| GUO Xu | School of Construction Machinery, Chang'an University, Xi'an 710064, China |
| ZHANG Hao | School of Construction Machinery, Chang'an University, Xi'an 710064, China |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 选区激光熔化制造过程相当复杂，通过理论模型去研究表面粗糙度较为困难，因此采用数据驱动的方式进行研究是一种可行的方案。方法 基于麻雀算法优化双向长短期记忆网络来预测表面粗糙度，并对比验证该模型的适用性。首先进行三因素四水平全因素试验，其次，以激光功率、扫描速度、扫描间距为输入，以粗糙度为输出，建立模型。然后，利用遗传算法优化预测模型，从而获得最佳工艺参数组合。最后，分析不同工艺参数下成形零件的表面形貌，探究各参数及其耦合关系对表面质量的影响。结果 最佳工艺参数为扫描间距0.12 mm、扫描速度1 800 mm/s、激光功率280 W，预测表面粗糙度为10.407 μm，调整工艺参数进行实验，得到的样件的平均表面粗糙度为10.897 μm，与预测值相比，误差仅为4.5%。工艺参数对表面形貌的影响从大到小的顺序为扫描速度、激光功率、扫描间距，各因素间存在耦合作用，且共同影响激光能量密度，能量密度过高、过低均会使表面形貌恶化。结论 基于麻雀算法优化双向长短期记忆网络构建的数据驱动预测模型适用于粗糙度的预测与优化，能够实现对样件表面粗糙度的精准预测，可以指导实践，保证加工质量。 |
| 英文摘要: |
| The surface roughness of the part manufactured by selective laser melting (SLM) is a macroscopic manifestation of the surface morphology, and this indicator significantly affects the fatigue life of the part. Due to the complex and multi-physical intersection of in the SLM process, it is challenging and obstructive to establish the theoretical model between surface roughness and the process parameters. Therefore, the data-driven approach can be a better choice to study this phenomenon. In this work, the Bi-directional Long Short-Term Memory network combined with the Sparrow Search Algorithm was used to predict the surface roughness. Meantime, the applicability of the proposed method was demonstrated through the comparison with other models. Firstly, a three-factor and four-level full-factor test was carried out to AlSi10Mg materials by SLM technology, followed by the measurement of the surface roughness via a surface roughness meter. Then, the data-driven model was established based on the Bi-directional Long Short-Term Memory network and the test data were obtained, during which the laser power, scanning speed, and hatch spacing were input variable, and the surface roughness value was output variable. Next, the Genetic Algorithm was applied to obtain the best combination of process parameters, and the reliability of the prediction was verified experimentally. Finally, the surface morphology of the formed parts under different scanning speed, hatch spacing, and laser power was analyzed to investigate the effect of each process parameter and their coupling effect on the surface quality. The constructed model had Coefficient of Determination of 0.920 3, which indicated that the model had higher goodness of fit. The Root Mean Squared Error of 0.872 3 and the Mean Absolute Error of 0.785 7 meant that model had better applicability, and Residual Predictive Deviation of 2.429 6 meant higher reliability and stability. This model reduced the training time for a large number of model parameters, enhanced the global search ability, and effectively improved the training efficiency and generalization ability. The optimal configuration of hatch spacing, scanning speed and laser power was 0.12 mm, 1 800 mm/s and 280 W, respectively. By adjusting the process parameter combination for the experiment, the average surface roughness of the sample obtained was 10.897 μm, and the error was only 4.5% compared with the predicted value. The effect of process parameters on the surface morphology was in such an order:scanning speed>laser power>hatch spacing. In addition, each factor had a great effect on the surface morphology of the molded parts and the coupling of the three factors together determined the laser energy density. Too low laser energy density led to insufficient melting of the powder, resulting in powder adhesion, the molten pool was difficult to tightly overlap and the porosity also increased dramatically. Too high energy density made the molten pool state extremely unstable, prone to molten pool splash, porosity, overburning and other effects of the molten pool. Therefore, the data-driven prediction model constructed based on the optimized Bidirectional Long Short-Term Memory network of the Sparrow Search Algorithm is suitable for surface roughness prediction and optimization, which can achieve accurate prediction of surface roughness of the sample parts and guide the practice to ensure the machining quality. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
渝公网安备 50010702501715号