ASIDS: Acoustic side-channel based intrusion detection system for industrial robotic arms

Industrial robotic arms play a vital role in manufacturing systems. However, they are susceptible to attackers executing malicious mechanical movements, thereby presenting significant threats to both industrial manufacturing and human safety. Existing techniques attempt to detect the abnormal signals within a manufacturing network to mitigate these attacks. However, these signals are unreliable since they might be deliberately tampered with by network attackers, including trajectory signals, and thus bypass anomaly detection. In this work, we propose ASIDS, a novel acoustic side-channel intrusion detection system to protect industrial robotic arms against data tampering attacks. We take advantage of an important insight that the acoustic side-channel signal emitted by an industrial robotic arm during a mechanical movement is unique, which could be used to reconstruct industrial robotic arms’ trajectory and detect abnormal movements. In particular, we extract the time-domain and frequency-domain features of the sounds emitted by the industrial robotic arm during a movement and reconstruct its trajectory by using a neural network. The data tampering attack can be detected by identifying the discrepancy between the reconstructed trajectory and the fake trajectory tampered with by the attackers through network traffic. To validate the performance of ASIDS, we have conducted real-world experiments on three industrial robotic arms, testing across more than 25,000 operational cycles. The experimental results indicate that ASIDS can accurately reconstruct trajectories and detect the attacks, achieving an average reconstruction error of 2.36% and an average detection rate of 95.9%.