New research with Pacific Northwest National Laboratory (PNNL) Uses machine learning, data analysis and artificial intelligence to identify potential nuclear threats.
Benjamin Wilson, non -POLIGERIAL PNNL analyst It has a unique opportunity to combine data exploration and machine learning techniques with nuclear analysis.
According to Wilson: “prevention of nuclear proliferation requires vigilance. It covers work, from nuclear material audits to research, who deals with nuclear materials. Techniques based on data analysis can be used to make it easier.”
With the support of the National Nuclear Security Administration (NNSA), mathematics initiatives for artificial reasoning in science (Mars) and the Department of Defense PNNL researchers are working on several projects to improve the effectiveness of nuclear non -syntraction and security measures. Below are the main theses of some of them.
Detection of nuclear materials leakage
Nuclear objects collect issued nuclear fuel and divide them into waste. The products are then used to produce compounds that can be processed into new fuel for nuclear reactors. These compounds contain uranium and platoon and can be used to produce nuclear weapons. Maea monitors nuclear objects to make sure that none of the nuclear materials is used for nuclear weapons. These are long -term regular controls, as well as collecting samples for further analysis.
“We could save a lot of time and labor costs if we could create a system that automatically detects irregularities based on data on the process of objects,” said Wilson.
In a study published in International Journal of Nuclear Safeguards and In-ProfationWilson collaborated with scientists in Sandia National Laboratories to create a virtual replica of the re -processing plant. Then they trained the artificial intelligence model to detect patterns in procedural data representing the leakage of nuclear materials. In this simulated environment, the model showed encouraging results. “Although it is unlikely that this approach be used in the near future, our system is a promising start to supplement existing security,,” said Wilson.
Analysis of texts to search for signs of nuclear weapons proliferation
Scientists from PNNL have developed a machine learning tool based on Google Bert: Language model trained on Wikipedia data for general queries. Language models allow computers to “understand” human languages - they can read texts and distinguish important information from them, including context and nuance. People can ask Bert questions, such as “Switzerland's population?” And get the right answer.
Although the model trained by Wikipedia is perfect in answering general questions, it lacks knowledge in the nuclear domain. The team was created by Ajax, a helper to fill this gap of knowledge.
“Although Ajax is still at an early stage, he can save analysts many hours of work, giving both a direct answer to queries and evidence of this answer,” said Subramanian. Evidence is particularly intriguing for scientists, because most machine learning models are often referred to as “black boxes” that leave no trace of evidence to their answers, even if they are correct. Ajax aims to ensure control capacity by receiving documents containing evidence.
According to subramans: “When the domain is just as important as detecting nuclear proliferation, it is very important to know where our information comes from.”
This development was published in International Journal of Nuclear Safeguards and Petopora.
Currently, Maea analysts spend many hours reading research articles and manually analyzing data rice containing information on nuclear proliferation. Scientists hope that in the future you can ask Ajax questions and get not only an answer, but also a link to the source of information. This will significantly simplify the task of analysts.
Image analysis to determine the origin of nuclear materials
Sometimes law enforcement officers come across nuclear materials that are beyond regulatory control and unknown origin. It is extremely important to find out where the material comes from and where it was created. After all, there is always a possibility that a separate sample can only be part of the material in illegal circulation. The criminal analysis of nuclear materials is one of the analytical tools used in this important work.
PNNL researchers, in cooperation with the University of Utah, Lawrence Livermore National Laboratory and Los Alamos National Laboratory, have developed a machine learning algorithm for criminal analysis of these samples. Their method uses images of an electron microscope to compare the microstructures of nuclear samples. Different materials contain subtle differences that can be detected by machine learning.
“Imagine that the synthesis of nuclear materials was like baking cookies,” said Elizabeth Jurrus, running Mars Initiative. “Two people can use the same recipe and end with differently looking cookies. The same with nuclear materials.”
Many factors, such as local humidity and purity of initial materials, can be influenced by the synthesis of these materials. As a result, nuclear materials produced in a specific enterprise acquire a special structure – a “characteristic appearance” that can be seen in an electron microscope.
Research is published in Journal of Nuclear Materials.
Scientists have created a library of paintings of various nuclear samples. They used machine learning to compare images from their library with unknown samples, and thus determine the origin of the unknowns.
This will help nuclear analysts to determine the source of materials and direct further tests.
It will probably take some time before agencies such as Maea include machine learning techniques in nuclear hazard detection methods. However, such tests can definitely affect this process and optimize it.
“Although we do not expect that machine learning will replace someone's work, we perceive it as a way to facilitate their work,” says scientists. “We can use machine learning to identify important information so that analysts can focus on what is most significant.”
