Cyber Attacks Detection and Prevention using Cryptography Algorithms for Industrial Automation

Most industries and societies couldn’t function without industrial control systems (ICSs). If they collapse, it might have a major effect on the economy and on people’s lives. This makes these systems vulnerable to both cyber and physical assaults. Although several techniques have been presented for detecting attacks, most either have a poor detection rate, a high false positive rate, or are system dependent. In this research, we investigate a technique for attack detection that uses 1D convolutions and autoencoders, two types of basic and lightweight neural networks. We apply these networks to the data in both the time and frequency domains and talk about the benefits and drawbacks of each. We test the proposed approach on three widely used public datasets and find that it not only achieves detection rates comparable to or higher than those of previously reported detection findings, but also benefits from a compact footprint, fast training and detection durations, and broad applicability.We also show how PCA may be useful, since it can provide excellent attack detection scores with the right quantity of data preprocessing and feature selection. We conclude with an examination of the suggested method’s resistance against adversarial assaults, which take use of the neural networks’ own blind spots to remain undetected while producing their desired physical impact. Our findings demonstrate the robustness of the proposed approach against such evasion attempts, since the attacker must forego the intended physical effect on the system in order to avoid detection. Based on these results, it seems that neural networks that are trained in accordance with the rules of physics may be relied upon more than those that are taught in more lax environments.