Answers are grounded in this paper and cite their sources.
Industrial Control Systems (ICS) increasingly integrate both information technology (IT) networks and operational technology (OT) processes, creating complex cyber–physical environments exposed to a wide range of cyber threats. Deploying effective intrusion detection in such environments is challenging not only due to the diversity of attack surfaces across IT and OT layers, but also because ICS components typically run on resource-constrained embedded hardware with limited computational capacity. This paper extends our prior temporal-symbolic representation (Zamaïand Espes, 2025) along three axes: we replace the static position-weighted pooling with a mask-guided variable-window pooling mechanism that learns end-to-end the effective temporal horizon required for each observation; we extend the framework from IT network intrusion detection to OT anomaly detection under an unsupervised protocol on multivariate sensor time series; and we demonstrate that a single compact convolutional architecture combining these two components provides a unified intrusion detection framework operating effectively across both ICS layers. Experimental evaluation on CICIDS2017, SWaT, and WADI demonstrates competitive detection performance and a favorable trade-off between accuracy, model compactness, and embedded deployability, with model sizes below 170k parameters. ARM CPU-class deployment measurements, together with Raspberry Pi 4-class throughput estimates, confirm that all configurations sustain real-time inference well above operational constraints, supporting practical deployment on low-cost resource-constrained devices commonly used at the industrial edge.
No takes yet. Share an insight, caveat, or question.
Zamaï et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: