First World Congress on Artificial Intelligence in Materials and Manufacturing (AIM 2022): Computer Vision for Materials and Manufacturing R&D I
Program Organizers: Taylor Sparks, University of Utah; Michael Dawson-Haggerty, Kerfed, Inc.; Elizabeth Holm, University of Michigan; Jin Kocsis, Purdue University; Adam Kopper, Mercury Marine; Benji Maruyama; James Warren, National Institute of Standards and Technology
Monday 9:30 AM
April 4, 2022
Room: 3' Rivers
Location: Omni William Penn Hotel
Session Chair: Joshua Stuckner, NASA Glenn Research Center
9:30 AM Break
10:00 AM
Feature Anomaly Detection System (FADS) for Intelligent Manufacturing: Anthony Garland1; Kevin Potter1; Matthew Smith1; 1Sandia National Labs
Anomaly detection is important for industrial automation and part quality assurance, and while humans are able to easily detect anomalies in components given a few examples, designing a generic automated system that can perform at human or above human capabilities remains a challenge. In this work, we present a simple new anomaly detection algorithm called FADS (feature based anomaly detection system) which leverages a pretrained convolutional neural network (CNN) to generate a statistical model of what a normal part should look like. By using a pretrained network, FADS demonstrates excellent performance similar to or better than other machine learning approaches to anomaly detection while at the same time FADS requires no tuning of the CNN weights. We demonstrate FADS’ ability by detecting process parameter changes on a custom dataset of additively manufactured lattices. In addition, we test FADS on benchmark datasets, such as MVTec, and report good results.
10:20 AM
An Exploratory Analysis of Low and High Temperature Tempered Steel Micrographs Using Machine Learning: Nicholas Amano1; Nan Gao1; Elizabeth Holm1; 1Carnegie Mellon University
Image based property predictions enable quick and relatively noninvasive sample testing of metals for a variety of applications. This study tries to predict physical properties, processing temperatures, and additive compositions of nominally 0.28% carbon steel using scanning electron micrographs. The image processing pipeline began with feature extraction, which used convolutional neural networks which employed transfer learning. Dimensional reduction was then conducted on the feature vectors to minimize computational overhead. Finally, a variety of regressions and classifications models were used to predict properties. Physical and processing properties were effectively classified and regressed upon by multiple different predictive models, while chemical properties were universally a challenge. The interaction between processing, microstructure, and physical properties is well understood, particularly compared to composition's impact on microstructure. The different compositions of additives lead to nonlinear and complex changes within microstructures, which lead to difficulties in modeling.
10:40 AM
Orientation Imaging Microscopy Grain Reconstruction Using Deep Learning: Patxi Fernandez-Zelaia1; Andrés Márquez Rossy1; Quinn Campbell1; Andrzej Nycz1; Christopher Ledford1; Michael Kirka1; 1Oak Ridge National Laboratory
Allotropic phase transformations take place in many commonly used structural materials following solidification. Phase reconstruction algorithms, which make inferences based on spatial structure present in orientation micrographs, are commonly used to estimate the underlying parent phase crystal structure. In this work we present a deep convolutional neural network architecture to estimate the prior austenite structure from observed martensite electron backscatter diffraction micrographs. A novel data augmentation strategy enables the training of our model using only four micrographs. The model generalizes well when tested on micrographs of a different material but its efficacy depends on the scale of microstructural features and the receptive field of the vision model. This work demonstrates that modern computer vision approaches can quantify complex spatial-orientation patterns present in orientation imaging micrographs.