Cyber-attacks on industrial facilities often employ the insertion of artificial false signals into the facilities' control systems as an attack method, and are thus capable of causing significant damage to these facilities, both economically and in human life.
In this project, our goal was to investigate whether it is possible to find and extract a "signature" from time-series signals originating from sensors; We used real industrial signals provided by Aperio Systems.
This signature will make it possible to identify the source of the signal and the type of signal (since each sensor produces a number of signals of different types), and thus can serve as a detection mechanism for false signals, assuming they do not have a known signature.
We present a deep learning-based technique of identifying the signature, using an LSTM-type network, which has long-term learning capability. The resulting network has a relatively high differentiation capability and is able to classify signals with a sufficient level of confidence.