Researchers at Brigham Young University (BYU) have demonstrated a way to cut nuclear reactor development time by a decade or more.
Currently, the designing and licensing processes for modern nuclear reactors take a long time. Licensing a new nuclear reactor design in the United States can take around 20 years and cost about $1 billion.
Besides, the construction of a modern nuclear reactor can take an additional five years and billions of dollars more after the licensing process. So, it takes roughly 25 years to develop a nuclear reactor in the US, as highlighted by the researchers.
The research team, led by professor Matthew Memmott, wants to reduce the duration and cost involved in bringing new nuclear plants online with the help of artificial intelligence (AI).
The team has demonstrated how AI can be utilized to accelerate the design and licensing processes for modern nuclear reactors.
“The idea is to shorten it, make it safer, cheaper and faster to get the nuclear power, rather than take 20 years to get the license,” the professor stated.
Rising demand for nuclear fusion
As the demand for energy and sustainability has grown ballistically, scientists across the world are working on renewable energy sources.
In this quest, nuclear fusion has emerged as the most desired option. This process, which powers the sun, is considered a key to generating limitless, clean energy on the Earth.
Therefore, both government and private organizations and institutions across the world are trying to build nuclear reactors that could successfully execute the fusion process.
“Our demand for electricity is going to skyrocket in years to come and we need to figure out how to produce additional power quickly,” Memmott said in the press release.
“The only baseload power we can make in the Gigawatt quantities needed that is completely emissions free is nuclear power.”
Complexity of nuclear reactor design
The complexity and time-consuming nature of nuclear reactor design stem from the multi-scale nature of the process.
It involves everything from neutrons on the quantum scale to macro-scale coolant flow and heat transfer. Multiple layers of physics are also tightly interconnected, which further complicates the design process.
Memmott explained these issues by citing his personal experience.
“When I worked at Westinghouse, it took a team of neutron specialists six months just to run one complete-core multi-physics model. And if they made a mistake two months in, they’d wasted two months of valuable computational time and would have to start over.”
The new research demonstrates how AI can alleviate this time burden. The team replaced a portion of the required thermal hydraulic and neutronics simulations with a trained machine learning model.
This model predicts temperature profiles based on variable geometric reactor parameters. These parameters are then optimized to create an optimal nuclear reactor design at a fraction of the computational cost of traditional methods.
Testing and validation
Interestingly, the researchers built and tested a dozen machine learning algorithms, identifying the top three and refining them until they found one that performed exceptionally well on a preliminary data set.
Their findings, published in the journal Nuclear Engineering and Design, show that their refined model can geometrically optimize design elements significantly faster than traditional methods.
For instance, the team tested their idea by using their AI algorithm to try and replicate a shield designed by a local nuclear company. Remarkably, the algorithm produced a near-perfect match of the company’s shield in just two days.
“This is amazing because within a couple of days we were able to do the same work that it took a team of engineers six months to do,” expressed the researcher.
This breakthrough could pave the way for a faster and more cost-effective expansion of nuclear power.
Most Popular
Comments / 0