VULNERABILITY ASSESSMENT OF A PHYSICAL PROTECTION SYSTEM
DESIGN USING A MULTI PATH ANALYSIS AND MOVING CRITICAL DETECTION
POINTS

The probability of interruption (PI) offered by physical protection systems (PPS) at nuclear facilities against an adversary attacking the facility was assessed by a modified Monte-Carlo-based multi-path adversary analysis method. Based on an adversary sequence diagram (ASD), a Monte Carlo script was developed to perform a multi-path adversary attach analysis. The analysis of adversary interruption used three types of distributions (Gaussian, Poisson, and Uniform) to determine the differences in choosing the probabilities of detection (PD) provided by the PPS elements. Compared to the deterministic approach used by the estimate adversary sequence interruption (EASI) model, the multi-path analysis approach presented in this study was not limited to the adversary's single path analysis. The PPS performance is not accurately represented by the EASI model because uncertainty cannot be estimated. Furthermore, unlike the EASI model, this model did not fix the critical detection point (CDP) at the same protection layer for all the attack scenarios. The CDP was moved to enable the analysis of the types of actions adversaries take to achieve their goals in response to their perceptions of the PPS. Several types of adversary actions, including random, rushing, covert, deep penetration, and most vulnerable path (MVP) were analyzed. According to the path selected by the adversary, the script developed was able to move the CDP. PI values and their associated uncertainties were more realistic because of this type of CDP movements. By eliminating the corresponding detection or delay elements of the PPS for the chosen adversary path, the threats from insiders were also modeled in the code. The script was integrated with the price of each PPS element, such as sensors and cameras present in the PPS. The relationship between cost and PI was examined by taking into account the unit price of the detection elements. Following the sampling of PD values from three different distributions, a PI value distribution was generated, and their uncertainties were compared for each sampling strategy, which were found to be not largely different.