First World Congress on Artificial Intelligence in Materials and Manufacturing (AIM 2022): Artificial Intelligence in Specific Manufacturing Processes II
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
Tuesday 1:30 PM
April 5, 2022
Room: 3' Rivers
Location: Omni William Penn Hotel
Session Chair: James Hardin, Air Force Research Laboratory/Rxms
1:30 PM Invited
Effects of Complex Die Cast Manufacturing Systems and the Critical Error Threshold on Applications of Machine Learning in Production: David Blondheim1; 1Mercury Marine
Research in machine learning (ML) for manufacturing processes is often applied to highly specific applications, typically with a focus on part quality predictions. This research is then completed in limited production settings or academic facilities. There is difficulty scaling these findings within manufacturing plants. Production manufacturing environments should be understood as highly complex systems when applying machine learning. System complexity, combined with the high-quality performance of most manufacturing systems currently achieve, means ML must exceed a Critical Error Threshold (CET) in accuracy to provide value of implementing ML in a manufacturing organization. This work will review system complexity and the CET to explain the difficulty in successful ML applications for quality predictions in die casting. Guidance on other ML applications in die casting will also be provided.
2:00 PM
MAKSAT : AI for Mining and Manufacturing Sector: Areena Khan1; Mahika Agrawal1; Sneha Tiwari1; 1VIT Bhopal
our software solves geological, topographical, geo-mechanical, engineering, mineralogy, and logging data problems to upgrade mining, using AI. It includes ML, OCR and NLP capabilities which are customizable and unified on a modular platform as per clients' needs: forecasts revenues, identifies requirement of potential products and understands customer behavior while predicting variability in manufacturing and material cost; improves order to cash process by creating an actual risk profile to customize strategy, expedite disputes' resolution, predict and prevent anomalies, enable better visibility and forecasting of cash flow; creates a directory based on data to develop targeted messaging and improves customer satisfaction; uses data from satellite imagery to predict what mineral deposits might be beneath the Earth's surface; AI combines core drill data, sample analysis results and survey reports to recommend techniques for maximizing ore deposits while accelerating returns on newly discovered ores by providing efficient solutions for extraction and processing of minerals.
2:20 PM
Automated Probabilistic Finite Element Model Calibration Tool Based on Uncertainty Quantification and Machine Learning: Joshua Fody1; Patrick Leser1; Sneha Narra2; 1NASA Langley Research Center; 2Carnegie Mellon University
Qualification and certification of safety critical parts is a hurdle to the adoption of metallic additively manufactured components for aerospace vehicle applications. Challenges include variability in part properties due to inconsistent defect distribution and microstructure. Understanding of the process through finite element modeling (FEM), and process control through in-situ monitoring, may result in significant improvements; however, solutions useful to manufacturers will require large volumes of data and automated data utilization. Toward this end, a generalizable automated FEM calibration paradigm is developed. This paradigm leverages existing and novel tools from machine learning and uncertainty quantification to enable the automatic calibration of FEMs without requiring prior knowledge of the model performance across input parameter space, including meshing and solver settings, which can require time consuming manual model probing or cause noisy and inconsistent predictions. The result is a probabilistic distribution of calibrated and validated FEM input parameters targeting measured data.
2:40 PM Break
3:10 PM
ORIGINALLY MON PM: Accelerating the Growth of Metal-Organic Framework Thin Films Guided by Pool-based Active Learning: Roberto Javier Herrera del Valle1; Luke Huelsenbeck1; Sangeun Jung1; Gaurav Giri1; Prasanna Balachandran1; 1University of Virginia
In this work, we develop a novel pool-based active learning (PAL) approach to rapidly guide the experimental growth of metal-organic framework (MOF) thin films with full coverage. The PAL approach is based on the idea that a supervised machine learning algorithm can achieve improved performance with fewer training data, provided the learning task allows the algorithm to autonomously choose data points from the vast unexplored parameter space. We implemented a PAL strategy that recommends data points based on two key measures: Diversity and Representativeness. While uncertainty-based sampling served as a measure of representative criterion, we used the maximin space-filling design to sample diverse growth parameters. The iterative PAL approach was able to efficiently guide the thin film growth towards promising regions of the vast processing space where full film coverage can be experimentally confirmed. The outcome of this work has major implications in using data-driven methods for high-throughput synthesis.