Safety and Security for Ethernet-based Visual Servoing Systems under hard Real-time Constraints

Description

Various manufacturing processes in industry use Visual Servoing Systems (VSS), which use real-time camera streams with high data rates to make control decisions. This thesis investigates how these streams can be securely delivered and maintain real time requirements even in case of link failures and adversaries inside the network. We developed a network topology and a corresponding network protocol. The protocol combines redundancy with fast integrity checks based on Keccak, the winner of the SHA-3 competition. Our topology is based on network coding and spreads important information over locally independent links. For proof of concept, we implemented a protocol converter for existing GigE Vision cameras on top of the Zedboard by AVNET. The resource intensive parts of the packet processing were accelerated by using a Field Programmable Array (FPGA), co-located with the ARM CPUs on the Zedboard. This allows an embedded device with limited CPU power to achieve high packet processing speeds while adding negligible latency to the overall system. Our implementation introduces only 159 us additional latency and can provide tamper detection for networking traffic for throughputs up to 565 Mbit/s. It outperforms the existing IPSec infrastructure of Linux on the Zedboard in all cases, and even a modern off-the-shelf X86-PC in selected cases. Our tests suggest that this throughput can be further increased as our developed hardware module is already able to process network traffic up to maximal line-rate of Gigabit Ethernet.