3rd World Congress on Artificial Intelligence in Materials & Manufacturing (AIM 2025): Optimization of Manufacturing Processes I
Program Organizers: Remi Dingreville, Sandia National Laboratories; Ali Riza Durmaz, Fraunhofer Institute Iwm
Wednesday 9:10 AM
June 18, 2025
Room: Elite Ballroom 1 & 2
Location: Anaheim Marriott
Session Chair: Stephen Price, Worcester Polytechnic Institute
9:10 AM
Application of Deep Learning Approaches to Model
the Heat Treatment Process-Microstructure-Property Relationship: Hoheok Kim1; Junwoo Kang1; Sehyeok Oh1; Jaimyun Jung1; Sejong Kim1; Ho Won Lee1; Seong-Hoon Kang1; 1Korea Institute of Materials Science (KIMS)
Designing materials to meet specific demands is essential in materials science, requiring a deep understanding of the process-structure-property (PSP) relationship. Traditional methods rely on feature engineering, while deep learning offers a framework that eliminates this need and enhances performance. This study introduces a deep learning framework to establish the PSP linkage for the heat treatment, microstructures, and mechanical properties of 42CrMo4 steel. We employed a conditional StyleGAN to generate microstructure images based on tempering temperatures and a ResNet algorithm to predict yield strength, tensile strength, and elongation from these images. Samples were heat-treated at various temperatures, revealing that lower temperatures resulted in tempered martensite, while higher temperatures increased ferrite content. Strength values decreased with rising tempering temperatures for both observed and generated images. The ResNet predictions aligned well with actual measurements, demonstrating the framework's ability to generate plausible microstructures and accurately predict properties under new conditions.
9:30 AM
ANN-Based Prediction of Steel Hardenability: Hai-Lin Chen1; Yunpeng Ma1; Qing Chen1; 1Thermo-Calc Software
We present a general artificial neural network (ANN) model for predicting microstructure formation and hardness in continuously cooled steels based on composition and austenitization condition. The model includes specialized sub-models for start and finish transformation temperatures, fractions of individual phases (ferrite, pearlite, bainite, and martensite), and hardness, as well as additional critical quantities. Thermo-Calc calculations and principles of physical metallurgy were integrated to enhance model accuracy and generalization. Additionally, a robust algorithm was developed to enable rapid generation of Continuous Cooling Transformation (CCT) diagrams, detailing transformation temperatures and critical cooling rates, for any specific steel compositions and austenitization conditions. These predicted diagrams, along with phase fraction and hardness estimations, provide valuable guidance for optimizing steel heat treatments to achieve desired mechanical properties.
9:50 AM
A Study on Digital Tools for the Safe and Sustainable Design of Materials: Andrea Gregores Coto1; Christian Precker1; Santiago Muiños Landín1; Leticia Hernando Rodríguez2; 1AIMEN; 2UPV/EHU
The urgent need to replace toxic hard chromium coatings has led the MOZART project to propose nickel-matrix nanocomposite coatings, which provide high corrosion and wear resistance without harmful environmental impacts. However, these coatings typically rely on boric acid, which conflicts with EU REACH standards. Our approach leverages a Conditional Variational Autoencoder to map the latent space of potential molecular alternatives, with Genetic Algorithms guiding the search to regions likely to yield safe and sustainable substitutes. Additionally, we assess the importance of molecular representation to improve model training. This innovative use of machine learning to optimize latent space demonstrates a powerful pathway for sustainable materials discovery in manufacturing.
10:10 AM
Chemical and Materials Informatics for Rapid Toxin-Free Product Development: James Saal1; 1Citrine Informatics
Discovering and/or designing replacements for toxic chemicals such phthalates and per- and polyfluoroalkyl substances (PFAS) from coatings and materials requires modeling capabilities to not only identify discrete small molecules that can mimic PFAS properties but also predict the highly-nonlinear interactions of these additives in coating and materials formulations that give rise to performance metrics of interest (e.g., resistance to wear, plasticization, and flammability). Citrine Informatics is a materials and chemicals informatics software company for accelerated design of novel products and manufacturing processes. Several relevant Citrine success stories will be shared, including the discovery of high-conductivity organic semiconductors with Panasonic and the development of non-phthalate plasticizers for polystyrene plastics with MIT. In each case, Citrine’s tools were used to connect physics-based simulations and existing experimental data to predict real-world chemicals and materials performance and rapidly identify candidate materials for lab-scale validation, resulting in a significant acceleration of product development.
10:30 AM Break
10:50 AM
Physics-Informed ML for Crystal Growth Manufacturing: Katherine Colbaugh1; Crystal Zhu2; Petia Koutev1; 1Crystal Growth Solutions, LLC; 2Case Western Reserve University
Single crystal materials are the functional components in a variety of applications including energy, semiconductor, medical imaging, nuclear detection, electrification, and telecommunication. The demands for crystal materials are rising, driving an increase in specifications that require higher levels of testing, design, data analysis, and operation. However, traditional data-driven AI methods are not effective due to the long cycle times - days to months, high cost of experimentation - high-temperature, precious metals, and rare-earth elements, and complexity of the crystal growth processes – multidisciplinary, many control parameters and measured variables. Here we have developed a deep-tech approach to combine physical domain knowledge and ML models. This benchmark study on industrial data validates the predictive capabilities of physics-informed ML for use in crystal growth manufacturing, including melt growth, solid-state synthesis, vapor deposition, and epitaxy. The methods developed in this work enable data-driven decision-making to increase yield, optimize parameters, and reduce crystal defects.
11:10 AM Cancelled
AI-Simulation Workflow to Accelerate Computational Discovery of Graphitization Product of Detonation Nanodiamonds: Xiaoli Yan1; Millicent Firestone1; Álvaro Vázquez-Mayagoitia1; Murat Keçeli1; Eliu Huerta1; 1Argonne National Laboratory
Detonation nanodiamonds (DNDs) are known to exhibit diverse morphologies, including carbon dots and nano-onion structures, which depend on various post-detonation processing parameters. While experimental techniques used to study these structures are widely regarded as accurate, they are costly, labor-intensive, and often impractical for exploring the full design space of process parameters. In this work, we introduce an AI-assisted molecular dynamics simulation framework to accelerate the optimization and refinement of process parameters for DND synthesis. ReaxFF-based simulations are performed on nanodiamonds with different morphologies, enabling the exploration of their structural evolution under varying process conditions. To predict time-dependent morphological transitions, we develop a graph-diffusion model that integrates these simulation results, offering a predictive tool for understanding the impact of parameter combinations on nanodiamond properties. This AI-driven approach significantly enhances the efficiency of the design process, reducing reliance on expensive experimental trials and opening new avenues for tailored nanodiamond production.
11:30 AM
Performance Analysis of Different Shaped Tool Electrodes During Electrical Discharge Machining of Inconel 718: Shankar Singh1; 1Sant Longowal Institute of Engineering & Technology (SLIET), Longowal
Electrical discharge machining is used for machining electrically conducting material by thermal energy, which leads to local heating followed by melting and evaporation of the work surface, resulting in small craters. Complex geometries involve different shapes, and tool electrode plays a major role in reproducing the shape on component. The aim of the current study is to explore the effects of copper tool electrode shapes along with other main EDM process factors on the basic machining performance, during EDM of Inconel 718. The different cross sections of electrode used in the work include circular, square and hexagonal. The effect of different tool electrode geometry (TEG) (circular, square and hexagonal), pulse current (Ip), pulse-on time, gap voltage on material removal rate, tool wear rate, and surface roughness have been studied. L9 (34) orthogonal array was used to perform experimental runs, and the results were analysed using Regression Analysis.