News Story
UMD’s Zhao to Lead New Dynamic PRA Study
University of Maryland (UMD) Department of Mechanical Engineering faculty member Yunfei Zhao and a multi-institutional team of collaborators have been awarded a grant from the Department of Energy (DOE) to address some of the key challenges to dynamic probabilistic risk assessment (PRA), which is used to support safe operation of nuclear power plants.
Zhao, the PI on the grant, will be working with a multi-institutional team of collaborators, including UMD Professor Mohammad Modarres and Associate Professor Katrina Groth, to improve the computational efficiency of dynamic PRA by developing a new algorithm based on reinforcement learning.
![](https://enme.umd.edu/sites/enme.umd.edu/files/Modarres250.png)
Mohammad Modarres, Professor of Mechanical Engineering and Director, Center for Risk and Reliability, University of Maryland.
The grant was awarded as part of the DOE’s Nuclear Energy University Program (NEUP), which supports research and development, infrastructure enhancement, and student education related to nuclear energy.
PRA, widely utilized in the nuclear industry to assess risks and make risk-informed decisions about design, operation, and safety protocols, comes in two types. While conventional “static” PRA makes use of manually-developed event trees, “dynamic” PRA utilizes physics-based simulations, incorporating stochastic models in order to yield a more complete picture of possible accident sequences.
![](https://enme.umd.edu/sites/enme.umd.edu/files/Groth%20revised2.png)
Katrina Groth, Associate Professor of Mechanical Engineering and Associate Director, Center for Risk and Reliability, University of Maryland.
"Dynamic PRA offers many advantages over conventional PRA, but the high computational cost and the lack of a user-friendly approach for its use in risk-informed decision-making have been major barriers to its widespread adoption,” Zhao said. “In this project we aim to address these two challenges by leveraging advances in machine learning, in particular reinforcement learning. In addition to nuclear power plants, the methodologies developed in this project will be generic and applicable to other safety-critical systems."
As a secondary objective, the team aims to improve the usability of dynamic PRA by developing a user-friendly question-answering system that can answer a variety of risk-related questions. Comprehensive case studies will be conducted to demonstrate and verify the proposed algorithm and the question-answering system.
Zhao, Groth, and Modarres are part of the university’s Center for Risk and Reliability (CRR), a cross-disciplinary research hub. This project is the third NEUP project led by faculty in CRR in the past five years. Additional members of the research team on this project include Zeyun Wu (Virginia Commonwealth University) and Aaron Epiney (Idaho National Laboratory).
Zhao, who joined the UMD faculty in 2023, specializes in risk and decision analysis for complex industrial systems, including nuclear plants. He analyzes a wide range of potential risks, including component failures, human error, natural hazards, and attacks staged by malicious actors.
Published August 21, 2024