Research On Key Problems Of Cyber Physical System Security

Cyber Physical System(CPS)is an integrated system that provides computing,networking,and physical processes.As a key area of research,CPS will play an important role in the design and development of future systems.However,in terms of CPS hardware,there is an unintentional security hole at every level of popular microcomputing devices from the smallest unit logic gate of large-scale integrated circuits.In terms of CPS software,from the small-scale assembly drivers to large-scale operation systems,there are also various potential security threats that are continuously exposed and not yet exposed at all levels of the system.Moreover,with the division of labor in information technology manufacturing,malicious manufacturers or designers in any one of the many links in the supply chain may insert their own background Trojans.These unintentional and intentional security loopholes pose a huge threat to human society.This paper will study the key problems of CPS security in terms of hardware security,network security and data processing security:(1)CPS Hardware security on integrated circuits.In a CPS,the main equipment is the terminal equipment responsible for data collection.The data collected by these devices will be sent to the computing center for further calculations or processing.Therefore,the security of the CPS hardware is the cornerstone of the entire CPS system security.Only by ensuring the safety of the CPS equipment that collects data,the collected data can be safe and reliable.The terminal equipment in the CPS system is characterized by low hardware configuration,and its computing capacity and power capacity may be limited.This article will explore stream ciphers and scan-based test designs for resource-constrained devices.This article presents a simple but powerful scan-based attack that uses less than 85 scan vectors to reveal keys and/or states stored in chips that implement the most advanced stream ciphers.In addition,with the prevalence of outsourced production of hardware chips,designed chips are easily inserted with hardware trojans during third-party production.Finding the node in the chip most likely to be inserted into the HT before chip production will greatly improve the process of detecting chip safety after chip production.The proposed method of searching inactive nets is low in complexity,high in accuracy,and tested on popular benchmarks and large circuits.(2)CPS network security on abnormal network usage detection.After collecting the data,CPS devices can process the data locally collaboratively or send the data to remote computing centers for further processing based on the requirements of the application.Either way,CPS transmits large amounts of data on its network.Therefore,ensuring the network security of the CPS system is another necessary condition for the security of the CPS system.This paper will explore how to detect the state of the entire network of a CPS system.We apply image processing and machine learning to the use of network resources to extract and identify communication patterns.The phenotypic analysis method is tested in four practical distributed applications.In the case of continuous network resources with appropriate length sampling,the overall recognition accuracy is about 99%.In addition,the identification error rate is used to detect abnormal network traffic.The experimental results show that the proposed anomaly detection method can effectively detect the abnormal network resource usage of each intensity.(3)CPS data processing security on malicious deep learning model.After collecting and transmitting data,the final step is processing the data.Data processing security is therefore an indispensable part of CPS system security.Because the data collected by the CPS system is massive,the currently popular deep learning model is helpful for analyzing and extracting known and unknown patterns or rules in this massive data.As the core technology behind deep learning,the security of neural networks has received little attention.For the first time,this paper proposes a novel method to design and insert a powerful neural network Trojan in the pre-trained neural network model.Most of the time,Po Trojans remains inactive and does not affect the normal functionality of its host model.However,once activated,PoTrojans may cause the host model to fail,or misprediction or classification,which poses a significant threat to CPS using deep learning to process data.We will explain the principles of Po Trojans and show how to design and insert Po Trojans in a pre-trained deep learning model.PoTrojans does not modify the existing architecture or parameters of pre-trained models and does not require retraining.Therefore,the proposed method is very effective.The proposed attack methods or countermeasures are focusd on the key issues of CPS security.This work greatly contribute to building a secury and trustworthy CPS.