Joint Sessions of AIM, ICME, & 3DMS: Digital Twins for Manufacturing
Program Organizers: TMS Administration

Tuesday 1:30 PM
June 17, 2025
Room: Platinum Ballroom 7&8
Location: Anaheim Marriott

Session Chair: Chinnapat Panwisawas, Queen Mary University of London

1:30 PM  Invited
Toward Sentient Manufacturing: Christopher Spadaccini1; Aldair Gongora1; 1Lawrence Livermore National Laboratory
    By integrating manufacturing process simulation, computationally driven design approaches, in-situ diagnostics, non-destructive evaluation, rapid characterization, data analytics and machine learning, and digital twins, we may be able to reduce fabrication timescales and costs, while improving quality and repeatability of components. Individually, each of these areas are helping to advance manufacturing and materials deployment but together could be even more powerful. Ultimately, we envision component design through manufacturing to be part of a sentient ecosystem where autonomous, on-the-fly process adaptation and even design changes are possible during the manufacturing process itself. Each area that contributes to this vision can be individually explored; however, the true power of the concept will be in the integration. Integrating these elements of the ecosystem to create a more aware, or sentient, materials and manufacturing enterprise, would allow for fabrication corrections and design changes on-the-fly, radically advancing the state of the art.

2:00 PM  
Digital Twins for Accelerated Materials Innovation: Surya Kalidindi1; 1Georgia Institute of Technology
    This presentation will expound the challenges involved in the generation of digital twins (DT) as valuable tools for supporting innovation and providing informed decision support for the optimization of material properties and/or performance of advanced heterogeneous material systems. This presentation will describe the foundational AI/ML (artificial intelligence/machine learning) concepts and frameworks needed to formulate and continuously update the DT of a selected material system. The central challenge comes from the need to establish reliable models for predicting the effective (macroscale) functional response of the heterogeneous material system, which is expected to exhibit highly complex, stochastic, nonlinear behavior. This task demands a rigorous statistical treatment and fusion of insights extracted from inherently incomplete, uncertain, and disparate data used in calibrating the multiscale material model. This presentation will illustrate with examples how a suitably designed Bayesian framework combined with emergent AI/ML toolsets can uniquely address this challenge.

2:20 PM  
Manufacturing and Control of Fiber Reinforced Polymer Composites Through FMEA-Based Digital Twin: Arshdeep Singh1; Soban Babu Beemaraj1; Sooriyan Senguttuvan1; Amit Salvi1; 1Tata Consultancy Services
    Manufacturing high-quality polymer composite parts without rejection is critical, especially for aerospace and wind energy structures requiring precise temperature control. This paper presents a Failure Mode and Effect Analysis (FMEA) framework for composite manufacturing, implemented through a digital twin that monitors and controls the process. Using this digital twin, multiple process deviations and potential failure scenarios in composite manufacturing are modeled, and optimal corrective actions are computed to create the FMEA table. This digitally generated FMEA table is then used to detect anomalies in the physical manufacturing process. To simulate this process, a multi-scale cure kinetics model is developed to capture the part's thermo-chemical-mechanical state. Additionally, the Joule heating effect is modeled for resistive heating elements embedded in a PID-controlled molding tool. Demonstrated on a tapered laminated composite with multiple heating zones, this approach enhances quality control and ensures more efficient composite manufacturing.

2:40 PM  
Harnessing Deep Learning Conditional Diffusion Models for Microscopy Modality Transfer of Light Optical Microscopy to Electron Backscattering Microscopy Diffraction Misorientations: Nicholas Amano1; Bo Lei2; Elizabeth Holm1; Dominik Britz3; Martin Müller3; 1University of Michigan; 2Lawrence Livermore National Laboratory; 3Steinbeis-Forschungzentrum Material Engineering Center Saarland
    Analyzing microstructures is essential in metallurgical science and manufacturing, prompting significant investment in the preparation and imaging of structured materials. In this research, we present a deep learning method that employs conditional diffusion models to generate the electron backscattered diffraction microscopy (EBSD) misorientation maps of quenched and tempered steel images from light optical microscopy (LOM) data. By leveraging the cost effective and relatively easy to produce LOM micrographs to generate high quality EBSD misorientation maps, we hope to accelerate the characterization step during steel manufacturing. This work is supported by a unique dataset of synchronized LOM and EBSD misorientation micrographs taken from the same sample locations and scales. We showcase diffusion models applicability to materials science imaging by reproducing EBSD misorientations from LOM images of highly complex multiphase steel. Our results indicate that diffusion models produce plausible and internally consistent EBSD misorientation mappings, but their absolute values are somewhat unreliable.

3:00 PM  
Generalized Graph Foundation Models as Versatile Data-Driven Digital Twins for Complex Technological Systems: Pawan Tripathi1; Benjamin Pierce1; Hein Aung1; Tommy Ciardi1; Kristen Hernandez1; Raymond Wieser1; Yangxin Fan1; Weiqi Yue1; Erika Barcelos1; Jayvic Jimenez2; Brian Giera2; Robert Gao1; Mengjie Li3; Kristopher Davis3; Laura Bruckman1; Roger French1; Quynh Tran1; 1Case Western Reserve University; 2Lawrence Livermore National Laboratory; 3University of Central Florida
    Generalized graph foundation models offer a flexible approach to constructing data-driven digital twins (ddDTs) for complex technological systems. Unlike traditional, physics-based digital twins that require idealized models built from first principles, ddDTs leverage real-world data streams to provide adaptable and modular representations of system behavior. By using spatiotemporal graph neural networks (st-GNNs) as a foundation, ddDTs capture dynamic interactions and performance characteristics, allowing for accurate monitoring and prediction across a range of applications. This work introduces a unified pipeline to develop graph-based foundation models for diverse systems, including solar photovoltaic fleets, direct ink write additive manufacturing, and laser powder bed fusion. The proposed approach avoids the constraints of physics-based assumptions, enabling a single ddDT architecture to address various performance issues and operational questions without extensive reconfiguration. These foundation models streamline digital twin implementation, supporting efficient, data-driven decision-making in technologically complex environments.