Logo: to the web site of Uppsala University

For as long as nuclear power has existed, there has been a concern for effectively safeguarding nuclear material since it was learned that it could be misused. Early on, many nations were interested in nuclear power, and therefore there was an urgent need to verify that it was only used for civilian purposes. Developed after the former US President Eisenhower’s "Atoms for peace" initiative, the nuclear non-proliferation and safeguards measures were proposed to control and verify nuclear materials worldwide. The legal framework was further strengthened with the development of the Treaty on the Non-Proliferation of Nuclear Weapons (NPT) in 1970. Since then, the International Atomic Energy Agency (IAEA) has served as the legal custodian and watchdog ensuring the peaceful use of nuclear technology through a system of measures called nuclear safeguards. Over the decades, this legal and technical framework has seen amendments owing to accruing experience and advancements in science and technology. More and more advanced techniques such as artificial intelligence (AI) and machine learning (ML) have recently been adopted at various levels within the frameworks of nuclear safeguards to effectively close any gaps the system.

This thesis serves as a compendium of instances where the application of machine learning could further be used to better safeguard nuclear materials from current and future nuclear facilities. Some of the applications explored within the scope of this thesis include the possibility of using machine learning algorithms in routine inspections of spent nuclear fuel (SNF) to verify whether it has been used in line with civilian motives. This can be achieved by safeguards inspectors through independent verification of operator declarations of bunrup, enrichment, and cooling time of the spent fuel. The capabilities of these machine learning techniques have been demonstrated for verification of spent fuel from more conventional, present-day light water reactors (LWRs) as well as for irradiated fuel salts from possible future concepts of molten salt reactors (MSRs). The thesis has also underscored some of the safeguards-related concerns regarding spent fuel from MSRs as the existing verification techniques may not be directly applicable to them and require significant adaptations.