System and method for particle morphology classification
A method is provided for particle morphology classification. The method includes obtaining input imagery of a particle sample that includes powder particles that are sintered together. The method also includes generating an input dataset for clustering based on the input imagery, including (i) detecting and segmenting powder particles, (ii) extracting and standardizing powder particle images, and (iii) calculating morphology metrics of the powder particles. The method also includes identifying categories, based on geometry or morphology-based similarities between different particles, using K-means clustering on Hu invariant moments of the powder particle images. Some implementations include receiving labels for the categories from a user, and subsequently using the categories to analyze or quantify future batches of particulate based on those labels. Some implementations include pre-categorizing the powder particles, based on size or other characteristics prior to clustering, in order to eliminate larger-scale differences between the powder particles.
1 . A method for particle morphology classification, the method comprising:
obtaining input imagery of a particle sample that includes powder particles that are sintered together;
generating an input dataset for clustering based on the input imagery, including (i) detecting and segmenting the powder particles in the input imagery, (ii) extracting and standardizing powder particle images, and (iii) calculating morphology metrics of the powder particles;
applying K-means clustering on Hu invariant moments of the input imagery to create categories by clustering a number of vectors/points into a number of categories based on proximity to other vectors;
identifying the categories based on geometry or morphology-based similarities between different particles; and
sorting the powder particles based on the identified categories into accepted powder particles and rejected powder particles,
wherein the accepted powder particles are inputs to a manufacturing operation, wherein the method further comprises:
training random forest or support vector machine (SVM) classifiers on the Hu invariant moments, and using the trained random forest or SVM classifiers to further categorize the identified categories based on fundamental attributes of the identified categories, and
training a convolutional neural network or a deep neural network on the input imagery based on the identified categories, to identify features that are more geometrically complex than the identified categories and are more accurate than metrics-based measurements of the Hu invariant moments, and using the trained convolutional neural network or the deep neural network to identify the features for particle samples.
2 . The method of claim 1 , further comprising:
receiving labels for the categories from a user; and
subsequently using the categories to analyze or quantify future batches of particulates based on the labels.
3 . The method of claim 1 , further comprising:
pre-categorizing the powder particles, based on size or other characteristics prior to clustering, in order to eliminate larger-scale differences between the powder particles.
4 . The method of claim 3 , wherein the pre-categorizing is performed using a size classifier that classifies the powder particles into a plurality of size categories based on particle size distribution.
5 . The method of claim 4 , wherein the plurality of size categories includes (i) a category for small particles, which accounts for a minimal amount of a total volume of the particle sample, (ii) a category for particles within predetermined size specifications, and (iii) a category for particles that are outside of the predetermined size specifications or are defective.
6 . The method of claim 1 , further comprising:
using multiple levels of unsupervised clustering to create groups within groups, when identifying the categories.
7 . The method of claim 1 , wherein identifying the categories is performed using a multi-tiered classifier that incorporates a plurality of types of classifiers.
8 . The method of claim 1 , further comprising:
displaying identified categories to a user;
receiving labels for the identified categories and a new set of categories after removal of redundant categories, from the user; and
saving a model comprising the new set of categories and the labels for subsequent categorization of particle samples.
9 . The method of claim 1 , wherein the input imagery is obtained from an optical particle measurement system, an in-line imaging system, or similar flow-based particle imaging system.
10 . The method of claim 1 , wherein the K-means clustering ignores a metric of the Hu invariant moments that differentiates based on reflection, for particle morphology.
11 . A computer system for physics simulation, comprising:
one or more processors; and
memory;
wherein the memory stores one or more programs configured for execution by the one or more processors, and the one or more programs comprise instructions for performing the method of claim 1 .
12 . A non-transitory computer readable storage medium storing one or more programs configured for execution by a computer system having one or more processors and memory, the one or more programs comprising instructions for performing the method of claim 1 .
13 . The method of claim 1 , wherein the rejected powder particles are recycled.
14 . The method of claim 1 , wherein the manufacturing operation is additive manufacturing.
15 . The method of claim 1 , wherein the manufacturing operation is solid fuel production.