Power-Based Online Anomaly Detection Of Industrial Robots

With the accelerating integration of information and industrial technology,industrial control systems(ICSs)security is more and more important.As important ICS devices,industrial robots are facing many security problems,and are susceptible to cyberattacks,such as hosting,malware injection,and communication hijacking.For example,by tampering with the control command and replaying the normal trajectory data,attackers result in motion trajectory deviation and evade the anomaly detection.However,the existing intrusion detection systems for industrial robots are still in their infancy,with the following deficiencies.First,the motion states-based methods are easy to be evaded by replaying the historical states.Secondly,the side-information-based information,such as sound and vibration,is hard to utilize because of the pervasive environmental noise in ICSs.Thirdly,the offline detection methods are not suitable for real-time industrial robots.In this work,inspired by the cross-checking,we find that power consumption strongly correlates to the motion state,which not only ensures the authenticity of the motion state data,but also is insensitive to environmental noise.In this paper,we design an online anomaly detection system using the power consumption information,and evaluate it using three types of industrial robots.The main contributions are summarized as follows.1.As it is difficult to collect power consumption information from industrial robots,we sample instantaneous current and voltage signals to replace the power consumption.To achieve this goal,we design a hardware power acquisition module system on the NI's C series functional modules and the Compact RIO controllers.Then we drive it on the LabVIEW development platform.At last,we verify the substitutability of current to power information through experiments.2.To achieve the adaptability of the anomaly detection model,the protocol format and network settings of three robots(i.e.,ABB,UR,and KUKA)are analyzed.After that we build a position acquisition module for multiple robot types,and further confirm the effectiveness of the power consumption,through several experiments.3.For the online anomaly detection task,a node-jumping time alignment mechanism is proposed.Specifically,we use the time domain conversion to match data from multiple clock sources.Then,we establish a mathematical model between the motion state and power consumption based on robot dynamic analysis.Meanwhile,considering the complex dynamic characteristics of that model,a linearization scheme is proposed to realize the decoupling of the unknown coefficients.Finally,we implement the power-based online anomaly detection module,and build the verification environment on ABB,UR,and KUKA.The real-time anomaly detection performance and system adaptability are confirmed.