Bio
I am a researcher at National Renewable Energy Laboratory (NREL) and earned my Ph.D. from the University of Michigan in Aerospace Engineering.
My work revolves around Scientific Machine Learning, the simulation of Complex fluid flows with High-Performance Computing, Uncertainty Quantification and Adversarial robustness. I develop methods to improve the efficiency and reliability of wind turbines, batteries, deposition reactors, bio reactors, inverters, and efficient engines.
Scientific Machine Learning (SciML)
The amount of data routinely generated by scientific calculations is an opportunity to improve our models, make them more efficient and extract novel information. I look for methods that intelligently use data and ML-inherited methods to solve problems that could not be tackled before. I work on:
- Probabilistic data augmentation
- Surrogate models and reduced-order models
- Information extraction from large database
- Data reduction methods
Complex fluid flows and High-Performance Computing (HPC)
Many engineering applications depend on some form of fluid mixing possibly coupled with small-scale phenomena, whether it be a bubbling flow in a bubble column reactor, a surface reaction in a deposition reactor, or an ignition kernel in an aircraft engine. Efficient numerical simulations that take advantage of novel computing architectures can enable affordable design optimization and allows gaining a deeper understanding of the limits and potential of the system at stake. I work on:
- Minimally dissipative methods
- Turbulent combustion modeling
- Chaotic dynamics of turbulence
- Surface reaction modeling
- Analytically reduced chemistry for HPC
Uncertainty quantification (UQ)
Decisions in energy systems typically involve financial an societal consequences. Therefore, appropriate quantification of uncertainty can enable critical decision-making. Uncertainty estimates are useful to objectively assess whether more information is needed and avoid over-confident conclusions. When decicions about extreme and rare events are needed, uncertainty estimates are even more instrumental but can be particularly challenging to compute. I work on:
- Bayesian inference
- Rare event probability estimation
- Uncertainty propagation
Adversarial robustness
Energy systems are typically subject to perturbations (for ex. fluctuation of load and production of renewable energy) that may lead to operating the system in untested settings. Such perturbations are often harmless but can also lead to catastrophic outcomes. At the same time, energy systems, like the power grid, keep evolving. Thus, pathways to catastrophic outcomes are also likely to change. A further complication is that some pathways to failure could be exploited by a capable malicious agent. I work on methods that can rapidly and automatically assess what vulnerabilities exist and how to patch them. Specifically, I work on:
- Single-agent and multi-agent reinforcement learning
- Anomaly detection