当前位置: X-MOL 学术Integr. Mater. Manuf. Innov. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
A Computer Vision Approach to Evaluate Powder Flowability for Metal Additive Manufacturing
Integrating Materials and Manufacturing Innovation ( IF 3.3 ) Pub Date : 2021-08-17 , DOI: 10.1007/s40192-021-00226-3
Jiahui Zhang 1 , Zhiying Liu 1 , Tianyi Lyu 1 , Yu Zou 1 , Mahdi Habibnejad-korayem 2 , Qiang Sun 3
Affiliation  

Additive manufacturing (AM) is a transformative technology to many industries that enables the fabrication of parts with complex geometries. A vast majority of powder-bed metal AM techniques use powder as feedstock. The powder packing behavior and flowability significantly influence the defect density of as-built parts and, eventually, affect their reliability and mechanical performance. The experimental characterization methods of powder flowability, for example, Hausner ratio, Carr index, and angle of repose, are rather time-consuming and cost-inefficient. Here, we show a rapid-deployed, low-cost, and reliable computer vision approach to evaluate powder flowability based on scanning electron microscopy images. We have trained seven machine learning models using 2,212 SEM images from 16 types of commonly used plasma-atomized metal powders in AM. Our results indicate that the vector of locally aggregated descriptors model with speedup robust features performs best among the models, represented by about 12 ± 7%. Mean absolute percentage error value is lower than traditional convolutional neural network model. The image analysis model can be implemented without a powerful computing system. The performance of such model is robust to the changes of image brightness. This study also demonstrates that our model can successfully predict the flowability of metal powder that does not exist in the original dataset. Such a computer vision approach provides an effective and efficient tool to evaluate and predict the powder flowability for AM.



中文翻译:

一种评估金属增材制造粉末流动性的计算机视觉方法

增材制造 (AM) 是许多行业的变革性技术,可以制造具有复杂几何形状的零件。绝大多数粉末床金属增材制造技术使用粉末作为原料。粉末填充行为和流动性显着影响成品零件的缺陷密度,并最终影响其可靠性和机械性能。粉末流动性的实验表征方法,例如豪斯纳比、卡尔指数和休止角,相当耗时且成本低效。在这里,我们展示了一种快速部署、低成本且可靠的计算机视觉方法,用于基于扫描电子显微镜图像评估粉末流动性。我们使用 AM 中 16 种常用等离子雾化金属粉末的 2,212 张 SEM 图像训练了七个机器学习模型。我们的结果表明,具有加速鲁棒特征的局部聚合描述符模型的向量在模型中表现最好,大约为 12 ± 7%。平均绝对百分比误差值低于传统的卷积神经网络模型。图像分析模型无需强大的计算系统即可实现。这种模型的性能对图像亮度的变化具有鲁棒性。这项研究还表明,我们的模型可以成功预测原始数据集中不存在的金属粉末的流动性。这种计算机视觉方法为评估和预测增材制造的粉末流动性提供了一种有效且高效的工具。约 12 ± 7%。平均绝对百分比误差值低于传统的卷积神经网络模型。图像分析模型无需强大的计算系统即可实现。这种模型的性能对图像亮度的变化具有鲁棒性。这项研究还表明,我们的模型可以成功预测原始数据集中不存在的金属粉末的流动性。这种计算机视觉方法为评估和预测增材制造的粉末流动性提供了一种有效且高效的工具。约 12 ± 7%。平均绝对百分比误差值低于传统的卷积神经网络模型。图像分析模型无需强大的计算系统即可实现。这种模型的性能对图像亮度的变化具有鲁棒性。这项研究还表明,我们的模型可以成功预测原始数据集中不存在的金属粉末的流动性。这种计算机视觉方法为评估和预测增材制造的粉末流动性提供了一种有效且高效的工具。这项研究还表明,我们的模型可以成功预测原始数据集中不存在的金属粉末的流动性。这种计算机视觉方法为评估和预测增材制造的粉末流动性提供了一种有效且高效的工具。这项研究还表明,我们的模型可以成功预测原始数据集中不存在的金属粉末的流动性。这种计算机视觉方法为评估和预测增材制造的粉末流动性提供了一种有效且高效的工具。

更新日期:2021-08-19
down
wechat
bug