CASPER: Context-Aware Anomaly Detection System for Industrial Robotic Arms

摘要

With the arrival of industry 4.0, industrial control systems are converted into “smart” industrial cyber-physical systems that depend on high interconnectivity enabled by ubiquitous applications. As these applications can significantly reduce maintenance and supervision costs, the integration of these applications is done with the “cost” being the focus overlooking the security aspect that suffers from the vulnerabilities that occurred due to increased attack surface. The adversaries aim to create physical alterations by exploiting these cyber vulnerabilities via so-called “cyber-physical” attacks. In this work, we introduce CASPER, a context-aware ubiquitous machine learning-based anomaly detection infrastructure that utilizes ubiquitous computing to detect anomalies of an industrial robotic arm. CASPER monitors the robotic arm's movements to ensure the arm follows a predetermined trajectory. The CASPER can reach an accuracy and F1 score of 97% which is promising for an industrial domain. We modify the joint velocity of an industrial robotic arm to create anomalies which we detect via CASPER.(Demo Video)