The data analysis subsystem of an Unmanned Aerial Vehicle (UAV) includes two main modules: a data acquisition module for data processing and a normalization module. One of the main features of an adaptive UAV protection system is the analysis of its cyber-physical parameters. An attack on a general-purpose computer system mainly affects the integrity, confidentiality and availability of important information. By contrast, an attack on a Cyber-Physical System (CPS), such as a UAV, affects the functionality of the system and may disrupt its operation, ultimately preventing it from fulfilling its tasks correctly. Cyber-physical parameters are the internal parameters of a system node, including the states of its computing resources, data storage, actuators and sensor system. Here, we develop a data normalization technique that additionally allows us to identify the signs of a cyber-attack. In addition, we define sets of parameters that can highlight an attack and define a new database format to support intrusion detection for UAVs. To achieve these goals, we performed an experimental study of the impact of attacks on UAV parameters and developed a software module for collecting data from UAVs, as well as a technique for normalizing and presenting data for detecting attacks on UAVs. Data analysis and the evaluation of the quality of a parameter (whether the parameter changes normally, or abrupt anomalous changes are observed) are facilitated by converting different types of data to the same format. The resulting formalized CPS model allows us to identify the nature of an attack and its potential impact on UAV subsystems. In the future, such a model could be the basis of a CPS digital twin in terms of security. The presented normalization technique supports processing raw data, as well as classifying data sets for their use in machine learning (ML) analyses in the future. The data normalization technique can also help to immediately determine the presence and signs of an attack, which allows classifying raw data automatically by dividing it into different categories. Such a technique could form the basis of an intrusion detection system for CPSs. Thus, the obtained results can be used to classify attacks, including attack detection systems based on machine learning methods, and the data normalization technique can be used as an independent method for detecting attacks.
We consider the trusted operation of cyber-physical processes based on an assessment of the system’s state and operating mode and present a method for detecting anomalies in the behavior of a cyber-physical system (CPS) based on the analysis of the data transmitted by its sensory subsystem. Probability theory and mathematical statistics are used to process and normalize the data in order to determine whether or not the system is in the correct operating mode and control process state. To describe the mode-specific control processes of a CPS, the paradigm of using cyber-physical parameters is taken as a basis, as it is the feature that most clearly reflects the system’s interaction with physical processes. In this study, two metrics were taken as a sign of an anomaly: the probability of falling into the sensor values’ confidence interval and parameter change monitoring. These two metrics, as well as the current mode evaluation, produce a final probability function for our trust in the CPS’s currently executing control process, which is, in turn, determined by the operating mode of the system. Based on the results of this trust assessment, it is possible to draw a conclusion about the processing state in which the system is operating. If the score is higher than 0.6, it means the system is in a trusted state. If the score is equal to 0.6, it means the system is in an uncertain state. If the trust score tends towards zero, then the system can be interpreted as unstable or under stress due to a system failure or deliberate attack. Through a case study using cyber-attack data for an unmanned aerial vehicle (UAV), it was found that the method works well. When we were evaluating the normal flight mode, there were no false positive anomaly estimates. When we were evaluating the UAV’s state during an attack, a deviation and an untrusted state were detected. This method can be used to implement software solutions aimed at detecting system faults and cyber-attacks, and thus make decisions about the presence of malfunctions in the operation of a CPS, thereby minimizing the amount of knowledge and initial data about the system.