Tech News

American nuclear power plants are among the safest in the world: what if they could also be cheaper?


U.S. nuclear power plants produce more than 20% of the electricity and half of the carbon-free electricity in the United States. The country’s pressing demand for even more electricity – particularly electricity that is carbon-free and does not contribute to climate change – is driving interest in advanced nuclear technologies. It also leads many to question whether the nuclear industry, sometimes characterized as more expensive to maintain and operate than other clean energy industries, can modernize and show more compelling results.

Alex Heifetz, senior nuclear engineer at the U.S. Department of Energy’s (DOE) Argonne National Laboratory, thinks otherwise artificial intelligence And machine learning the tools and techniques will pave the way to “yes”.

Heifetz and a team of researchers used PUR-1, the NRC-commissioned all-digital research reactor at Purdue University, to conduct a series of experiments using a physics-based neural network with learning by transfer (TL-PINN). Neural networks are methods within machine learning which can be trained to recognize patterns in data. However, they are not good at making predictions outside of their field of training. Researchers like Heifetz believe PINNs can overcome this shortcoming by fusing neural networks with differential equations to solve complex scientific or engineering problems. This would help them extrapolate to new scenarios and make more accurate predictions.

In a work published in Nature Science ReportsHeifetz’s team created a model of a small modular reactor (SMR) with automated operations, then used machine learning to do something the nuclear industry would never attempt in an operating nuclear power plant: they tinkered with its data and training models.

“We focused on the specific question of algorithm performance,” Heifetz said. “The more data you have, the more your model improves in terms of prediction accuracy. With newer technologies and designs, you may not have all the data needed to make a prediction. We applied the concept of transfer learning, where you retrain the PINN with a synthetic data set, then examine the similarity between the data sets to see if you can trust the synthetic data to make accurate predictions.

Many previous studies have focused on the safety and security of nuclear reactors, but this research focused specifically on cost. The overall goal, Heifetz said, is to reduce operation and maintenance costs and enable the development and commercialization of low-cost SMRs. Small modular reactors produce 300 megawatts of power or less, compared to today’s reactors, which are rated at about 1,000 megawatts (enough power to provide electricity to about 1,000,000 homes).

Many believe that SMRs could significantly increase the impact of nuclear energy. They take up less space than existing nuclear reactors, are less expensive to build, operate and maintain, and have modern features such as digital gauges, devices and sensors, which most existing reactors did not initially possess. ‘origin. They can be used to provide electricity to remote areas of the United States, or possibly in futuristic stations allowing the electrification of long distance trucking is more plausible.

“Building a new nuclear reactor is difficult, that’s why we don’t want to shut down the ones we have,” Heifetz said. “The challenge is to take a $10 billion nuclear reactor built in the 1950s and – while preserving all its safety features – make it more modern, efficient and self-sustaining, with fewer operators and less maintenance time. ‘stop. But we also want to test new models, such as SMRs, for which there is not enough data. By developing a physics-based neural network that does not require a lot of time to train, we may be able to reduce the cost of training new models.

This new work is based on previous collaborations between Argonne and Purdue University. The two recently entered into an agreement that will allow certain researchers to share affiliations with both institutions. Under this joint appointment framework agreementdesignated researchers will have access to the facilities and expertise of a national laboratory and Purdue University.

DOE’s Advanced Research Projects Agency-Energy (ARPA-E) funded the TL-PINN research. Additional support was made possible through a donation to AI Systems Lab at Purdue University by Goldman Sachs Gives.


Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button