About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Modeling, AI Applications, and Method Development in Nuclear Materials
|
| Presentation Title |
A Physics-Informed Neural Network-Based Phase-Field Model for Simulating Hydride Assisted Cracking |
| Author(s) |
Shyam Kishor Sharma, Bhanu Kumar Mishra, Indeevar Singh |
| On-Site Speaker (Planned) |
Shyam Kishor Sharma |
| Abstract Scope |
Hydride assisted cracking (HAC) is the most common failure mechanism for the failure of pressure tube made up of zirconium alloy in nuclear reactor. HAC is a time-dependent fracture process in which hydride is formed near the crack tip due to stress gradient and hydrogen concentration in the material. The phase-field modelling (PFM) approach for fracture has emerged as a promising energy-based method to overcome the limitations of discontinuity-based methods. However, the computational time for the PFM is large, and the multiphysics nature of DHC modelling further increases the simulation time. A promising approach for reducing computational time is the application of Physics-Informed Neural Networks (PINNs) to solve the partial differential equations (PDE) of phase-field evolution and hydrogen diffusion. In this study, we developed a hybrid FEM–PINN framework for DHC simulation in zirconium alloys by combining the strengths of conventional FEM and PINN. The developed numerical model is validated by comparing the load-displacement response and critical stress intensity factors with the experimental results. A structured description of the physics involved and its numerical implementation is presented for better understanding. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Nuclear Materials, Modeling and Simulation, Machine Learning |