3rd World Congress on Artificial Intelligence in Materials & Manufacturing (AIM 2025): Applied ML for Manufacturing II
Program Organizers: Remi Dingreville, Sandia National Laboratories; Ali Riza Durmaz, Fraunhofer Institute Iwm
Monday 1:30 PM
June 16, 2025
Room: Elite Ballroom 1 & 2
Location: Anaheim Marriott
Session Chair: Baishali Garai, RV University
1:30 PM Invited
ROM-Net in Additive Manufacturing: David Ryckelynck1; 1Mines Paris PSL University
ROM-net reduced order modeling (ROM) is applied to finite element thermo-mechanical simulation of metal additive manufacturing. This is a significant challenge due to the continuously evolving computational domain on which a local reduced basis is required to apply a ROM-net. The support of local reduced bases is closely related to the sampling of mechanical predictions. Considering the modeling of a directed energy deposition process, it is proposed to organize the training set of simulation snapshots according to an energy deposition length that represents the progress of the process. When the projection-based ROM-net is applied to the full-order model, the simulation data sequence allows the design of a local ROM depending on categories of input parameters. The similarity of data in a category is evaluated by a Grassmann distance between local reduced subspaces. The simulation of the construction of a turbine blade showed a computational speedup of about 100.
2:00 PM
Non-Destructive Residual Stress Prediction via Indentation Plastometry and CNN Modeling: Deunbom Chung1; Minwoo Park1; Kyeongjae Jeong2; Heung Nam Han1; 1Seoul National University; 2Sungkyunkwan University
Accurate assessment of residual stress is essential for ensuring the structural integrity and performance of advanced materials. However, conventional evaluation methods can be destructive or limited in applicability. To address these shortcomings, this study proposes non-destructive indentation plastometry integrated with a finite element-convolutional neural network (FE-CNN) to predict residual stress and plasticity. Validated finite element simulations of spherical indentation under various material properties served as the training database. The CNN effectively captures spatial information from three directional indentation profiles, achieving high accuracy in both simulation and experimental results. Its robustness was assessed through sensitivity tests on intentionally manipulated data, thereby confirming reliability under various experimental data quality. Applied to additively manufactured samples, the method demonstrated its non-destructive and highly accurate residual stress prediction capabilities, verified by neutron diffraction measurements. Consequently, the FE-CNN framework offers a versatile and robust platform for broader non-destructive stress evaluations in materials.
2:20 PM
Automated 3D Segmentation of Refractory Material Microstructures Using Deep Learning for Improved Corrosion Resistance: Johan Moncoutie1; Lalitha Raghavan1; Deniz Cetin1; Darren Rogers1; Damien Bolore2; Sunhwi Bang1; 1Saint-Gobain Research North America; 2Saint Gobain Research Provence
Material microstructure plays a critical role in determining the corrosion behavior of refractory materials. Traditional manual segmentation of 3D imaging data is often time-consuming, labor-intensive, and subject to inconsistencies. In this work, we introduce an automated segmentation approach using deep learning to streamline this process. A 3D U-Net convolutional neural network (CNN) architecture, specifically designed for volumetric data, is employed to segment 3D images of newly synthesized refractory materials, obtained via Focused Ion Beam Scanning Electron Microscopy (FIB-SEM). This automated approach not only saves material scientists significant time but also ensures more consistent and accurate results. Additionally, we demonstrate how linking microstructural features to macroscopic properties provides valuable insights, offering a robust tool for further material analysis and property prediction.
2:40 PM Cancelled
Bayesian Calibration and Uncertainty Quantification of a Cohesive Zone Model for Metal-Oxide Interfaces: Revanth Mattey1; Jason Schulthess1; Alexander Swearingen1; James Cole1; 1Idaho National Laboratory
Plate-type nuclear fuel fabricated through hot isostatic pressing (HIP) consists of a U–10Mo-based fuel foil encapsulated in an aluminum alloy cladding. Understanding the failure mechanisms of these fuel plates is crucial for safe reactor operation. A key potential failure mechanism is de-bonding between the aluminum cladding, which may be exacerbated by second phase precipitates forming along the interface during the HIP fabrication. The precipitates' shape and size, influenced by peak temperatures and cooling rates, can degrade bond strength. To model the degradation, a cohesive zone model is adopted. Calibrating interface properties with a finite element forward model and Bayesian approaches is computationally intensive. Thus, surrogate models like Gaussian Process (GP) regression are developed to predict the mechanical response which are then utilized to calibrate the fracture properties of the interface through Bayesian inversion. The total uncertainty is estimated by combining the forward propagation of parameter uncertainty and model discrepancies.
3:00 PM Break
3:30 PM
Long-Term Durability of GFRP Rebars in the Alkaline Environment Under Sustained Loading Using Machine Learning and Empirical Modelling: Mudassir Iqbal1; Daxu Zhang2; Xiao Lin Zhao1; 1The Hong Kong Polytechnic University; 2Shanghai Jiao Tong University
Corrosion in marine environments poses a threat to the safety of engineering structures. The corrosion of steel bars induced by the penetration of chloride ions is the main reason for the deterioration of reinforced concrete structures. Novel fibre reinforced polymer (FRP) reinforced concrete structures are increasingly being developed and used in civil engineering to ensure their safety. The current American Concrete Institute (ACI) 440.1 R-15 guidelines for environmental reduction factors and long-term durability of GFRP (glass fibre reinforced polymers) rebars in harsh environments are based on limited experimental data and are deemed conservative. This study evaluated the GFRP rebars subjected to sustained loading under accelerated aging. Due to its eminent nature, tensile strength reduction of aged GFRP rebar was studied for durability evaluation using XGBoost model. Experimental data of 308 samples [1-5] were used in the investigation. The tensile strength reduction factor was studied as a function of the type of glass fibers, rebar surface, exposure type (bare / concrete embedded), the magnitude of sustained loading, type of resin, size of rebar, the volume fraction of fibers, pH of alkaline solution, temperature and duration of conditioning. The XGboost model was trained using the best hyperparameters obtained from grid tuning. The model displayed reliable results in terms of R-squared and mean absolute error for the training (0.98, 1.19 %) and test data (0.82, 3.87%), respectively. The developed model was used to study the behaviour of input variables towards tensile strength reduction using Shapely Additive explanations. The degradation of GFRP rebars is strongly influenced by key factors such as temperature, conditioning duration, solution pH, and the magnitude of sustained loading. It was observed that when the sustained load exceeded 20% of the ultimate tensile strength, the degradation process accelerated significantly. This finding is consistent with prior research, further validating the newly developed model in this study. The environmental reduction factor was determined using a machine learning approach combined with the Arrhenius relationship. Both methods were compared for accuracy. Additionally, a web-based application was developed to predict tensile strength degradation in alkaline environments. The findings and conclusions from this research contribute to a deeper understanding of the durability of GFRP rebars in challenging conditions.
3:50 PM
Harnessing AI for Precision Manufacturing of Nanofibers Membranes: Zeeshan Khatri1; 1Mehran University of Engineering & Technology
Producing nanofibers is very challenging in terms of quality, precision, and efficiency since process parameters like voltage, flow rate, and distance are sensitive. The current paper deals with the integration of AI with the electrospinning process to bring a paradigm shift in the production of nanofibers. Through extensive data collection on critical manufacturing variables and nanofiber properties, we train a prediction model of AI for ideal parameters in the nanofibers that will have better diameter uniformity and improved mechanical strength. The laboratory confirmation proves the capability of AI to improve control of process, to reduce variability, and to minimize resources wastage. This method opens up a smarter and more efficient operation in nanofiber manufacture that offers transformative insights into areas of material science and industrial-scale applications.