A. Stein, V. I. Perumal, R. Frye, D. Beckett, C. B.S. Woodruff
UHV3D Inc,
United States
Keywords: IFE, fusion, advanced energy
Summary:
Inertial Fusion Energy (IFE) has been found to be a promising approach to achieving net- gain. Scaling up fusion energy production using IFE will depend on, among other things, the ability to push the limits of mechanical design of fusion targets to extract maximum performance. Target production, currently, is an expensive endeavor (~$1M/target) with long lead times (~weeks) [1]. Additionally, it is projected that up to a million fusion targets will be needed per day per fusion power plant to meet the needs of the fusion industry by 2035. To meet such demands in a cost-effective manner, extreme multidisciplinary design considerations for the fusion target are to be made to extract maximum performance from the target material at a practical manufacturing cost. Such multidisciplinary and multiphysics design considerations are difficult to be achieved using conventional simulation techniques, such as finite element (FE) or finite volume (FV) methods due to the computational costs. An AI-based design optimization approach enables a change in this paradigm by generating surrogate models that emulate the response surface of more computationally expensive models [2]. We propose the use of a multidisciplinary and multiscale design optimization approach accelerated using AI-based models that optimizes the fusion target design. In this approach, the objective and constraints of the optimization are defined based on the fluidic, thermal and structural performance of the target. Thereafter, high-fidelity FE and FV simulations based on the accurate mathematical description of the various loads during the manufacturing, fueling, and injection of the targets. These simulations will then be used to train AI models, such as convolutional, graph, recurrent and physics-informed neural networks (NN) to develop accurate surrogates for the physics-based models. NNs are known to efficiently learn and capture the underlying complex relationships between the input parameters, such as fluid, material, topology, and the resulting physical responses [3]. Our AI surrogates will be used in place of the FE and FV approaches to compute the field responses, such as fluid pressure, temperatures, displacements, and stresses, and the respective gradients. These values will be fed into a gradient-based multiobjective multiscale topology optimization. The optimizer will call the AI surrogates to compute the various response fields at the macroscale and use a graph-based NN to synthesize appropriate metamaterials to meet the design requirements at the microscale. Using AI in place of the standard solvers can drastically reduce the computational time for iterative approaches such as optimization while ensuring the desired accuracy [3]. [1] Alex Haid, Neil Alexander, Mike Farrell, Rick Olson, Mark Schmitt, Brian Haines, Cliff Thomas, Elijah Kemp, Brent Blue, Mike Campbell. Additive Manufacturing for Inertial Fusion Energy Target Production System White Paper 2022. https://lasers.llnl.gov/content/assets/docs/nif-workshops/ife-workshop-2021/white-papers/haid- GA-IFE-workshop-2022.pdf [2] Jeffrey Hittinger and Simon Woodruff. Artificial Intelligence & Machine Learning, Chapter 8 of the Report of The Fusion Energy Sciences Workshop on Inertial Fusion Energy https://events.bizzabo.com/IFEBRN2022/home [3] Perumal, V, Abueidda, D, Koric, S, Kontsos, A. (2023). Temporal convolutional networks for data- driven thermal modeling of directed energy deposition. J of Mfg. Processes, 85, 405-416.