Implementation of cnn based algorithm for cyber-attacks detection on a real-world control system

Abstract

The emergence of the Industry 4.0 concept leads to crucial changes in manufacturing by building advanced industrial systems and applications based on Cyber-Physical Systems (CPS), as the core of this approach. Using CPS, manufacturing assets are designed in the form of systems of systems through interconnection of smart devices with integrated computation and communication capabilities. System control logic is distributed over a large number of resources, and its performance is achieved through their coordinated work and ubiquitous communication raising the issue of cyber-attacks by malicious adversaries. Since cybersecurity within industrial control systems is safety related, it is necessary to timely detect cyber-attacks on industrial assets; for these purposes a number of different approaches have been developed. As a technique of choice, deep learning (DL) based methods emerge, providing good online performances. In this work, we focus on the implementation of a DL based cyber-attack detection algorithm on an electro-pneumatic positioning system containing smart sensor and smart actuator. In particular, we employ cyber-attack detection procedure based on 1D Convolutional Neural Network (CNN) at the local controller of the smart actuator. The implemented algorithm can successfully detect cyber-attacks in real-time, as will be experimentally demonstrated.