Significant progress in stress prediction research using artificial intelligence has been made by researchers at the National Center for Supercomputing Applications (NCSA) and The Grainger College of Engineering at the University of Illinois Urbana-Champaign (UIUC). Working with the lens of deep operator network (DeepONet) implementations, the researchers have aimed to enhance stress response predictions in complicated geometries, such as those noticed in additive manufacturing. Employing the NCSA’s Delta system, they have achieved results significantly quicker than those obtained with the traditional finite element methods.
Through Illinois Computes, a program that provides comprehensive computing and data storage resources, the research was conducted by the team. This program has enabled collaboration across multiple disciplines, integrating machine learning and computational mechanics. The Delta system, noted for its superior GPU computing capabilities, played a key role in training deep neural networks and creating training data with the help of Abaqus software.
This research has so far resulted in two significant publications. The initial publication, featured in “Computer Methods in Applied Mechanics and Engineering,” introduces a new DeepOnet that uses a residual U-Net (ResUNet) to code complex geometries. This methodology signifies the first usage of ResUNet in the DeepONet architecture, showing an enhancement in memory efficiency and adaptability compared to traditional methods.
The second paper, published in “Engineering Applications of Artificial Intelligence,” details an advanced version of DeepONet known as S-DeepONet. This advanced version utilises cutting-edge sequential learning methods and provides improved accuracy for multi-physics solutions under diverse thermal and mechanical loads.
Iwona Jasiuk, a professor of mechanical science and engineering at UIUC, highlights the revolutionary nature of additive manufacturing, stating that it provides almost limitless implementation opportunities.
DeepONet is described as a powerful and speedy computational tool capable of simulating the additive manufacturing process at different spatial and temporal scales. This kind of simulation is an essential part of understanding the additive manufacturing process, and in applying and tracking it.
The research is not just an advancement in AI applications, it also has extensive implications for advanced manufacturing processes and the development of digital twins. This cooperation between NCSA and MechSE emphasises the combining effect of multi-disciplinary skills and innovative technology.
Come and let us know your thoughts on our Facebook, Twitter, and LinkedIn pages, and don’t forget to sign up for our weekly additive manufacturing newsletter to get all the latest stories delivered right to your inbox.
“Why did the 3D printer go to therapy? Because it had too many layers of unresolved issues!”
0 Comments