Nuclear Material Accountancy (NMA) system quantitatively evaluates whether nuclear material is diverted or not. Material balance is evaluated based on nuclear material measurements based on this system and these processes are based on statistical techniques. Therefore, it is possible to evaluate the performance based on modeling and simulation technique from the development stage. In the performance evaluation, several diversion scenarios are established, nuclear material diversion is attempted in a virtual simulation environment according to these scenarios, and the detection probability is evaluated. Therefore, one of the important things is to derive vulnerable diversion scenario in advance. However, in actual facilities, it is not easy to manually derive weak scenario because there are numerous factors that affect detection performance. In this study, reinforcement learning has been applied to automatically derive vulnerable diversion scenarios from virtual NMA system. Reinforcement learning trains agents to take optimal actions in a virtual environment, and based on this, it is possible to develop an agent that attempt to divert nuclear materials according to optimal weak scenario in the NMA system. A somewhat simple NMA system model has been considered to confirm the applicability of reinforcement learning in this study. The simple model performs 10 consecutive material balance evaluations per year and has the characteristic of increasing MUF uncertainty according to balance period. The expected vulnerable diversion scenario is a case where the amount of diverted nuclear material increases in proportion to the size of the MUF uncertainty, and total amount of diverted nuclear material was assumed to be 8 kg, which corresponds to one significant quantity of plutonium. Virtual NMA system model (environment) and a divertor (agent) attempting to divert nuclear material were modeled to apply reinforcement learning. The agent is designed to receive a negative reward if an action attempting to divert is detected by the NMA system. Reinforcement learning automatically trains the agent to receive the maximum reward, and through this, the weakest diversion scenario can be derived. As a result of the study, it was confirmed that the agent was trained to attempt to divert nuclear material in a direction with a low detection probability in this system model. Through these results, it is found that it was possible to sufficiently derive weak scenarios based on reinforcement learning. This technique considered in this study can suggest methods to derive and supplement weak diversion scenarios in NMA system in advance. However, in order to apply this technology smoothly, there are still issues to be solved, and further research will be needed in the future.

A sample size calculation algorithm was developed in a prototype version to select inspection samples in domestic bulk handling facilities. This algorithm determines sample sizes of three verification methods satisfying target detection probability for defected items corresponding to one significant quantity (8 kg of plutonium, 75 kg of uranium 235). In addition, instead of using the approximation equation-based algorithm presented in IAEA report, the sample size calculation algorithm based on hypergeometric density function capable of calculating an accurate non-detection probability is adopted. The algorithm based the exact equation evaluates non-detection probability more accurately than the existing algorithm based on the approximation equation, but there is a disadvantage that computation time is considerably longer than the existing algorithm due to the large amount of computational process. It is required to determine sample size within a few hours using laptop-level performance because sample size is generally calculated with an inspector’s portable laptop during inspection activity. Therefore, it is necessary to improve the calculation speed of the algorithm based on the exact equation. In this study, algorithm optimization was conducted to improve computation time. In order to determine optimal sample size, the initial sample size is calculated first, and the next step is to perform an iterative process by changing the sample size to find optimal result. Most of the computation time occurs in sample size optimization process performing iterative computation. First, a non-detection probability calculation algorithm according to the sample sizes of three verification methods was improved in the iterative calculation process for optimizing sample size. A computation time for each step within the algorithm was reviewed in detail, and improvement approaches were derived and applied to some areas that have major effects. In addition, the number of iterative process to find the optimal sample size was greatly reduced by applying the algorithm based on the bisection method. This method finds optimal value using a large interval at the beginning step and reduces the interval size whenever the number of repetitions increases, so the number of iterative process is less than the existing algorithm using unit interval size. Finally, the sample sizes were calculated for 219 example cases presented by the IAEA report to compare computation time. The existing algorithm took about 15 hours, but the improved algorithm took only about 41 minutes using high performance workstation (about 22 times faster). It also took 87 minutes for calculating the cases using a regular laptop. The improved algorithm through this study is expected to be able to apply the sample size determination process, which was performed based on the approximate equation due to the complexity and speed issues of the past calculation process, based on the accurate equation.

Material balance evaluation is an important measure to determine whether or not nuclear material is diverted. A prototype code to evaluate material balance has been developed for uranium fuel fabrication facility. However, it is difficult to analyze the code’s functionality and performance because the utilization of real facility data related to material balance evaluation is very limited. It is also restricted to deliberately implement various abnormal situations based on real facility data, such as nuclear diversion condition. In this study, process flow simulator of uranium fuel fabrication facility has been developed to produce various process data required for material balance evaluation. The process flow simulator was developed on the basis of the Simulink-SimEvents framework of the MathWorks. This framework is suitable for batch-based process modeling like uranium fuel fabrication facility. It dynamically simulates the movement of nuclear material according to the time function and provides process data such as nuclear material amount at inputs, outputs, and inventories required for Material Unaccounted For (MUF) and MUF uncertainty calculation. The process flow simulator code provides these data to the material balance evaluation code. And then the material balance evaluation code calculates MUF and MUF uncertainty to evaluate whether or not nuclear material is diverted. The process flow simulator code can simulate the movement of nuclear material for any abnormal situation which is difficult to implement with real process data. This code is expected to contribute to checking and improving the functionality and performance of the prototype code of material balance evaluation by simulating process data for various operation scenarios.

The measurement activities to evaluate material balance of nuclear material are usually performed by operator. It is because that the IAEA does not have enough manpower to carry out nuclear measurement accountancy of all nuclear materials in the world. Therefore, the IAEA should consider scenarios which facility operator tries to divert nuclear material for misuse by distorting measurement record. It is required to verify the operator’s measurement data whether it is normal or not. IAEA measures inventory items using their own equipment which is independent of facility operator equipment for verification. Since all inventory lists cannot be verified due to limited resources, the number of items to be verified is determined through statistical method which is called as sample size calculation. They measure for the selected items using their own equipment and compares with operator’s record. The IAEA determines sample size by comprehensively considering targeted diverted nuclear material amount and targeted non-detection probability and performance of measurement equipment. In general, the targeted diverted nuclear material amount is considered significant quantity (plutonium: 8 kg, uranium-235: 75 kg). If the targeted non-detection probability or the performance of the verification equipment is low, the sample size increases, and on the contrary, in the case of high non-detection probability or good performance of verification equipment, even a small sample size is satisfied. It cannot be determined from a single sample size calculation because there are so many sample size combinations for each verification equipment and there are many diversion scenarios to be considered. So, IAEA estimates initial sample size based on statistical method to reduce calculation load. And then they calculate non-detection probability for a combination of initial sample size. Through the iteration calculation, the sample size that satisfies the closest to the target value is derived. The sample size calculation code has been developed to review IAEA’s calculation method. The main difference is that IAEA calculates sample size based on approximate equation, while in this study, sample size is calculated by exact equation. The benchmarking study was performed on reference materials. The data obtained by the code show similar results to the reference materials within an acceptable range. The calculation method developed in this study will be applied to support IAEA and domestic inspection activities in uranium fuel fabrication facility.

It is important to ensure worker’s safety from radiation hazard in decommissioning site. Real-time tracking of worker’s location is one of the factors necessary to detect radiation hazard in advance. In this study, the integrated algorithm for worker tracking has been developed to ensure the safety of workers. There are three essential techniques needed to track worker’s location, which are object detection, object tracking, and estimating location (stereo vision). Above all, object detection performance is most important factor in this study because the performance of tracking and estimating location is depended on worker detection level. YOLO (You Only Look Once version 5) model capable of real-time object detection was applied for worker detection. Among the various YOLO models, a model specialized for person detection was considered to maximize performance. This model showed good performance for distinguishing and detecting workers in various occlusion situations that are difficult to detect correctly. Deep SORT (Simple Online and Realtime Tracking) algorithm which uses deep learning technique has been considered for object tracking. Deep SORT is an algorithm that supplements the existing SORT method by utilizing the appearance information based on deep learning. It showed good tracking performance in the various occlusion situations. The last step is to estimate worker’s location (x-y-z coordinates). The stereo vision technique has been considered to estimate location. It predicts xyz location using two images obtained from stereo camera like human eyes. Two images are obtained from stereo camera and these images are rectified based on camera calibration information in the integrated algorithm. And then workers are detected from the two rectified images and the Deep SORT tracks workers based on worker’s position and appearance between previous frames and current frames. Two points of workers having same ID in two rectified images give xzy information by calculating depth estimation of stereo vision. The integrated algorithm developed in this study showed sufficient possibility to track workers in real time. It also showed fast speed to enable real-time application, showing about 0.08 sec per two frames to detect workers on a laptop with high-performance GPU (RTX 3080 laptop version). Therefore, it is expected that this algorithm can be sufficiently used to track workers in real decommissioning site by performing additional parameter optimization.