Last month, the Nuclear Energy University Program (NEUP) at the US Department of Energy (DOE) pledged more than $56 million to support nuclear energy researchers and programs across the country. The funding announcement also included results
from this year’s NEUP Innovations in Nuclear Energy R&D Student Competition, which recognizes students making innovative contributions to nuclear research and development.
Ezgi Gursel, an industrial and systems engineering PhD student working with Associate Professor Anahita Khojandi, was one of the nation’s 14 winners.
“It’s a great honor to receive this award,” said Gursel. “I’m very grateful to the NEUP for their recognition and support.”
Gursel received the award for using artificial intelligence (AI) to detect human errors in nuclear power plants (NPPs). While automation has made many aspects of NPP operation more reliable, human-made records still occasionally disagree with automatic sensor readings.
“Human errors can lead to a range of problems in NPPs, including downtime, delayed responses, and the need for unexpected and costly corrective measures,” Gursel said. “Those issues can impact the safety, reliability, and operational efficiency of a plant.”
To catch those errors, NPPs often employ anomaly detection algorithms—but those algorithms can struggle to find patterns in the highly detailed datasets, where each timestamp can correspond to dozens of sensor readings.
Gursel and her team realized that a more sophisticated approach was needed. A generative adversarial network (GAN) consists of two competing AIs; one pollutes the dataset with convincing forgeries while the other tries to determine which data points are real.
Over time, the network learns to accurately highlight points that do not fit with the true data—like those derived from human error.
“GANs are commonly used in fields such as cybersecurity, medicine, and power systems,” said Gursel, “but to the best of our knowledge, ours is the first work using GANs to detect human errors in NPPs.”
By incorporating a GAN, Gursel’s project outperformed the algorithms previously considered state-of-the-art.
Gursel’s NEUP award includes a $3,500 honorarium and opportunities to travel and present her research.
Contact
Izzie Gall (865-974-7203, egall4@utk.edu)