Recovery systems architecture for cyber-manufacturing systems against cyber-manufacturing attacks: Reinforcement learning approach

Cyber-manufacturing attacks are cyber-attacks that identify a vulnerability in cyber entities and exploit it to impact manufacturing systems. The cyber entities inside manufacturing systems consist of operational and information technology. The number of cyber-manufacturing attacks continues to rise every year by taking advantage of weaknesses in cyber entities. The widespread impact of these attacks has attracted researchers towards developing solutions against cyber-manufacturing attacks. Additionally, it is recognized that some of the cyber-manufacturing attacks cannot be prevented, and solely adopting prevention measures are not sufficient. On the other hand, detection measures do not guarantee the functioning of the cyber-manufacturing systems after the cyber-manufacturing attacks. Hence, to ensure the attacked manufacturing systems are achieving operational goals a systematic recovery measure should be implemented. To implement a successful recovery measure an architecture is required. Thus, this work proposes a four-layer recovery architecture. The architecture consists of a systems layer, attack identification layer, data auditing and detection layer, and reinforcement learning based recovery layer. This work explains the advantages of implementing a reinforcement learning based recovery layer compared to the state of the art recovery methods. The implementation of the recovery architecture is demonstrated by conducting a simulation of the conveyor system constructed inside the Unity3D physics platform.