Over the years I have worked on several projects in physics and engineering, with a strong focus on optimization, modeling and experimental validation. A non-extensive list of my research interests is:
- Bayesian Optimization of Fusion Reactor Scenarios (in progress)
- Integrated Modeling of SPARC Burning Plasmas
- Perturbative Transport in Tokamaks
- Surrogate Techniques for Model Validation
- Turbine Optimization via Evolutionary Algorithms
Integrated Modeling of SPARC Burning Plasmas
As a postdoctoral researcher at the Plasma Science and Fusion Center, I have been collaborating with the company Commonwealth Fusion Systems on the design of SPARC, a new tokamak that will start operations in 2025. Thanks to innovative superconductor technology, SPARC will be the first device to generate net energy (Q>1), a very important milestone in the quest for fusion energy.
Integrated modeling simulations allow us to dive into the physics of SPARC plasmas, to have an early look into what regimes to expect and what performance SPARC will be able to achieve. My role in the project has been to lead the integrated modeling efforts to inform the design of this unique machine, and provide predictions of kinetic profiles and other plasma quantities to use in further physics studies.
Want to know more about the SPARC project?
– M. Greenwald et al 2018
– B. Mumgaard et al APS-DPP 2018
– P. Rodriguez-Fernandez et al 2019 Nuclear España SNE (in Spanish)
– MIT News 2018, “MIT and newly formed company launch novel approach to fusion power“
Perturbative Transport in Tokamaks
During my years as a doctoral student at MIT under the supervision of Prof. White, I worked on unveiling one of the most prominent mysteries in core plasma transport physics: Why does the temperature at the center of the tokamak increases when the edge is cooled down?
As a result of three and a half years of work, and thanks to fruitful collaborations with scientists at MIT, General Atomics, Princeton and Max Planck Institute, we provided an explanation to this 20-year-old problem. We published five peer-reviewed journal papers, and the work has been featured in about a dozen talks at international conferences and workshops. Our work includes experiments in three world-class devices (Alcator C-Mod, DIII-D and ASDEX Upgrade) and simulations with two of the leading transport solvers in our community (TRANSP and ASTRA).
C. Angioni et al 2019 Nucl. Fusion 59 106007
P. Rodriguez-Fernandez et al 2019 Phys. Plasmas 26 062503
P. Rodriguez-Fernandez et al 2019 Nucl. Fusion 59 066017
P. Rodriguez-Fernandez et al 2018 Phys. Rev. Lett. 120 075001
P. Rodriguez-Fernandez et al 2017 Nucl. Fusion 57 074001
P. Rodriguez-Fernandez 2019 Ph.D. Thesis, MIT
Surrogate Techniques for Model Validation
As a side project during my Ph.D., I developed a technique to efficiently explore the vast parameter space required for the validation of plasma transport models. The tool was named VITALS (Validation via Iterative Training of Active Learning Surrogates), and it makes use of the same genetic algorithms aided by surrogate modeling that I developed for turbine optimization.
Publications using VITALS:
A. J. Creely et al 2019 Plasma Phys. Control. Fusion 61 085022
A. J. Creely 2019 Ph.D. Thesis, MIT
VITALS example (by Prof. White):
Turbine Optimization via Evolutionary Algorithms
For my Master’s thesis project, I had the pleasure to work at the Laboratorio di Fluidodinamica delle Macchine from the Politecnico di Milano, Italy. I worked alongside Prof. Persico on the development of optimization techniques for the shape of turbomachinery blades, particularly those used in Organic Rankine Cycles (ORC). We employed genetic algorithms for multi-objective, constrained optimization problems, and gaussian process regression models to reduce the number of Computational Fluid Dynamics (CFD) simulations required.
We demonstrated the benefit of using this technique on a ORC benchmark case. With respect to the baseline design, our workflow reduced single-blade pressure losses by half, an improvement of 20% with respect to previous gradient-based methods.