High-Throughput Unsupervised Profiling of the Morphology of 316L Powder Particles for Use in Additive Manufacturing

TL;DR

<3-5 sentence high-level summary> The paper tackles the problem of capturing industrial-scale variability in 316L powder morphology for selective laser melting. It introduces three unsupervised pipelines—O(2)-invariant VAE, descriptor-based (CDF/FD/ZM), and functional-data clustering (GPmix)—applied to ~126k images to profile particle shapes. Among them, Fourier descriptor + k-means provides the best balance of cluster separation and computational efficiency, achieving sub-millisecond per-particle processing and enabling real-time feedstock monitoring possibilities. The framework supports tracking morphology across reuse cycles and offers a transferable approach for other powder chemistries, potentially informing flowability, packing density, and part quality in SLM workflows.

Abstract

Selective Laser Melting (SLM) is a powder-bed additive manufacturing technique whose part quality depends critically on feedstock morphology. However, conventional powder characterization methods are low-throughput and qualitative, failing to capture the heterogeneity of industrial-scale batches. We present an automated, machine learning framework that couples high-throughput imaging with shape extraction and clustering to profile metallic powder morphology at scale. We develop and evaluate three clustering pipelines: an autoencoder pipeline, a shape-descriptor pipeline, and a functional-data pipeline. Across a dataset of approximately 126,000 powder images (0.5-102 micrometer diameter), internal validity metrics identify the Fourier-descriptor + k-means pipeline as the most effective, achieving the lowest Davies-Bouldin index and highest Calinski-Harabasz score while maintaining sub-millisecond runtime per particle on a standard desktop workstation. Although the present work focuses on establishing the morphological-clustering framework, the resulting shape groups form a basis for future studies examining their relationship to flowability, packing density, and SLM part quality. Overall, this unsupervised learning framework enables rapid, automated assessment of powder morphology and supports tracking of shape evolution across reuse cycles, offering a path toward real-time feedstock monitoring in SLM workflows.

Figures (10)

• Figure 1: (a) Original grayscale micrograph of a single particle captured on the Malvern Morphologi 4. (b) Intensity histogram with Otsu's threshold (vertical dashed line) used to separate foreground particles from background. (c) Binary mask generated by applying the threshold, isolating the particle silhouette. (d) Final normalized input: the mask is centered, padded, and resized to a 128 × 128-pixel canvas, ensuring consistent spatial dimensions for VAE.
• Figure 2: Left: An irregular 2D shape with its centroid marked. Radial lines drawn at various angles $\theta$ extend from the centroid to the boundary, each labeled with a distance $D_i$. Right: A plot of distance $D(\theta)$ versus angle $\theta$, showing how the CDF captures shape variations around the perimeter.
• Figure 3: Original Shape (left): The closed contour sampled as complex points $z(t)=x(t)+iy(t)$. Fourier Transform (center): The formula $C_n=\frac{1}{N}\sum z(t)e^{-i2\pi nt/N}$ above a stem plot of coefficient magnitudes $|C_n|$. Reconstruction (right): The approximated contour $z'(t)=\sum C_ne^{i2\pi nt/N}$ showing the effect of using a limited number of harmonics.
• Figure 4: Schematic of the O(2)-invariant VAE. $\mathbf{\oplus}$ denotes the concatenation of output feature fields. The encoder uses six E(2)-steerable convolutions (with batch normalization and nonlinearity), inserts equivariant pooling after layers 2 and 4, and applies global group pooling after layer 6. The encoder's MLP block is a four-stage sequence: a linear projection, a 1D batch-normalization layer, an ELU activation, and a final linear layer. The resulting embeddings are passed to a decoder built from a standard convolutional architecture.
• Figure 5: Clustering via VAE embedding.