Technology
Tech Breakthrough
Charles Darwin University Proposes Multi-Modal Deep Learning Framework for Rapid Powder Flowability Assessment
2026-05-27
A team from Charles Darwin University in Australia has introduced an explainable multi-modal deep learning framework for rapid quality control of powder flowability in additive manufacturing. Using 183,120 scanning electron microscopy (SEM) image-record pairs from 38 metal powders, the study trained five convolutional neural network backbones to predict angle of repose, Hausner ratio, Carr index, and Hall flow rate. The best model, RegNetY-400MF, achieved an optimal accuracy-efficiency balance (MAE=0.186, RMSE=0.351, R²=0.988), with multi-modal fusion significantly improving prediction accuracy over single-image models. By directly correlating particle morphological features with powder flowability through SEM images, this research provides a new approach for rapid inter-batch quality consistency assessment of metal powders.
📰 News Source
This article was translated from a Chinese source.
Source URL not available
Show Chinese original
查尔斯·达尔文大学提出基于多模态深度学习的粉末流动性快速评估框架
澳大利亚查尔斯·达尔文大学团队提出一种用于增材制造粉末流动性快速质量控制的可解释多模态深度学习框架。研究使用来自38种金属粉末的183,120张扫描电子显微镜图像（SEM）-记录对，训练了五种卷积神经网络主干用于预测休止角、豪斯纳比、卡尔指数和霍尔流速。最佳模型RegNetY-400MF实现最优精度-效率平衡（MAE=0.186，RMSE=0.351，R²=0.988），相比单一图像模型，多模态融合显著提高了预测准确性。该研究通过SEM图像直接关联颗粒形态学特征与粉末流动性，为金属粉末批次间质量一致性快速评估提供了新方法。