Access level control in SCADA systems with YOLO algorithm

In today's industrial production systems, dependence on manpower is gradually decreasing as a result of technological developments in control and automation systems. This trend increases the efficiency of production processes, while at the same time strengthening operational independence. In factory conditions, effective management and supervision of machine operators is critical for the appropriate functioning of production processes. In mass production facilities, during the adaptation of production lines to different products, machine parameters must be updated by the operators via Human-Machine Interfaces (HMI) within the Supervisory Control and Data Acquisition (SCADA) systems. Among operators working in multi-shift systems, it can sometimes be difficult to determine who is making these parameter changes. Traditional username and password-based login systems can create security risks and operational disruptions among operators. And passwords are no longer a foolproof way to keep people's information secure. Microsoft reports about 1,287 password attacks every second, approximately 111 million per day. Although it is recommended to use both numbers and letters in encryption for security, different authentication methods are recommended as a second layer of security. This thesis aims to optimize the user authentication process of operators at system logins by integrating face recognition technology into HMIs in SCADA systems. The proposed system enables operators to access authorized pages by simply looking at the camera, while at the same time detecting, alarming and recording potential identity fraud attempts (for example, an attempt to access the system using another operator's photo). In this study, the YOLOv8 model, which was trained on our own dataset created from the faces of different individuals for the purpose, achieved a 90 percent success rate with a detection time of 0.5 seconds. In addition, the FaceNet model used for face recognition, which achieved a success rate of 99.63 percent on the LFW dataset, was integrated into our application and very successful results were obtained with a detection time of 0.6 seconds. This approach aims to increase system security, speed up the access processes of operators and increase the level of operational control. The tests conducted within the scope of the thesis show that the developed system provides effective security with a 90 percent accuracy rate in a total analysis time of less than 2 seconds (spoofing detection and face analysis). It is clear that this application designed for industrial automation systems can be effectively used in other access-limited industrial areas.