First World Congress on Artificial Intelligence in Materials and Manufacturing (AIM 2022): Human-AI Collaboration for Materials and Manufacturing Problems
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 9:30 AM
April 5, 2022
Room: 3' Rivers
Location: Omni William Penn Hotel
Session Chair: Anesia Auguste, Air Force Research Laboratory
9:30 AM Break
10:00 AM Invited
Teaching Tools to be Teammates: Digital Manufacturing Research at AFRL: Sean Donegan1; James Hardin1; Andrew Gillman1; Ezra Ameperosa1; Dan Berrigan1; 1Air Force Research Laboratory
The Digital Manufacturing Research Team (DMRT) at the Air Force Research Laboratory researches and develops approaches for state-of-the-art digital manufacturing relevant to the Department of the Air Force. The team’s core vision is teaching tools to be teammates: developing techniques that effectively integrate software and hardware tools with designers, technicians, and end-users to establish efficient and effective human-machine-factory teams. This presentation will cover the DMRT’s overall strategy while highlighting collaborative work with academic and industrial partners across the nation. Specifically, we will brief current work on robotic learning from human demonstrations, agile automation for AI-driven closed-loop feedback, and optimization approaches for digital manufacturing. Finally, the DMRT plans for future directions will be covered.
10:30 AM
Human-Robot Collaboration for the Application of Speckle Patterns Through Learning from Demonstration Techniques: Anesia Auguste1; Jennifer Ruddock1; Erick Braham2; Ezra Ameperosa2; James Hardin2; Andrew Gillman2; 1UES, inc; 2Air Force Research Laboratory
Industry is incorporating more robotic machines and broadening their scope of use. Artisanal processes, which require human experts, are major challenges to automation because robots are typically actuated in a different manner and have disparate sensor packages than human experts. To explore these limits, we explore integrating human users and robots to teach a robotic system to apply high-quality speckle patterns via spray painting for Digital Image Correlation (DIC). DIC is a non-destructive optical method to measure local deformation in a specimen. The accuracy is dependent on the quality of the artisanal applied speckle pattern. By utilizing learning from demonstration (LfD) techniques like Task-Parameterized Gaussian Mixture Model, which empowers experts with little robotics background to intuitively interact with the robot, we developed generalized spray trajectories based on expert kinesthetic teaching of the robot. Overall, we are combining human intuition with machine precision in order to produce high-quality speckle patterns.
10:50 AM
Developing Autonomous Spray Processes with Deep Reinforcement Learning Guided by Human Demonstration: Ezra Ameperosa1; Anesia Auguste2; Jennifer Ruddock2; Erick Braham1; James Hardin1; Andrew Gillman1; 1Air Force Research Lab; 2UES, Inc.
Deep Reinforcement Learning (DRL) has recently shown promise for agile robotic control in real world applications but is often impractical due to the amount of data required and expensive trial-and-error learning. In principle, providing foresight to these algorithms through expert demonstration helps bridge developing control policies for otherwise difficult to automate tasks. We investigate this concept for applying speckle-patterns for Digital Image Correlation, a process known for its reliance on the expert knowledge and skills. Here, we consider the use of DRL in a custom spraying game/simulation and ways to efficiently accelerate learning with human demonstrations. We examined various methods for capturing information from an ‘expert’ performing the task, then how ‘expert’ data is infused into training to boost speed and performance of learning. We further aim to generalize these strategies with the potential of improving the efficiency of a more diverse range of complex manufacturing processes.
11:10 AM
Experimental Validation of Materials Discovered by Autonomous Intelligent Agents.: Richard Padbury1; 1Lucideon
One of the biggest challenges facing industry is the identification of new materials to meet emerging industrial demands. However, gaps in advanced materials properties remain due to the intractable challenge of screening a vast material composition search space. To overcome this challenge, TOYOTA Research Institute (TRI) has developed an end-to-end platform that leverages artificial intelligence-based agents to explore high-dimensional search spaces for new materials. The computational autonomy for materials discovery (CAMD) system uses a combination of past knowledge, surrogate models and heuristic rules to promote cost-effective exploration-exploitation strategies that guide which experiments to perform next, aiding the pursuit of new stable compounds. In collaboration with Lucideon we present a complete workflow, from initial discovery campaign to materials synthesis, to demonstrate how new inorganic materials can be identified by CAMD, down selected for further analysis, experimentally validated using appropriate in-situ characterization techniques and subsequently, processed via solid state synthesis routes.
11:30 AM
Teaching the Machine: Characterizing Speckle Patterns for Virtual Demonstrations
: Jennifer Ruddock1; 1UES, Inc
Learning from Demonstration models trained in virtual environments are a potential pathway to teach brittle robotic programming to perform dynamic, artisanal processes. Virtual demonstrations allow humans to share natural motions with robotic systems. The spray application of speckle-patterns used in digital image correlation provides a representative stochastic process, typically applied by human experts, and inherently difficult to automate -- especially on non-planar geometries. Critical to developing an accurate virtual environment is the rapid and accurate linkage between process parameters and the applied speckle pattern. In this work, we developed image processing tools for analysis and characterization of speckle patterns applied using a robotic spray and six process parameters. The speckle pattern features were then modeled with a support vector machine and verified by predicting the process parameters for desired spray pattern features. The results provide a method for rapid characterization and enable a more accurate simulation needed for virtual demonstrations.
11:50 AM
Profit-Driven Methodology for Servo Press Motion Selection Under Material Variability: Luke Mohr1; Nozomu Okuda1; Alex Kitt1; Hyunok Kim1; 1EWI
Servo presses enable new types of forming motion profiles that can be used to stamp difficult materials such as high strength steels. EWI's newly-developed Smart Forming Algorithm is an application of Bayesian statistics to intelligently select which motion profile maximizes the expected utility given the properties of the incoming material. Bayesian logistic regression is used in conjunction with expected utility to estimate manufacturing returns, which can be used to make informed process decisions. This presentation will describe the statistical methodology used to build the algorithm, how an operator can interpret results from the algorithm to influence process decisions, and a use case which demonstrates that this approach can increase expected returns by more than 20%.