Implementing SCADA Scenarios and Introducing Attacks to Obtain Training Data for Intrusion Detection Methods
This provides a valuable resource for security researchers to train and test anomaly detection methods in industrial control systems, though it is incremental as it focuses on dataset creation rather than new detection algorithms.
The authors tackled the lack of public datasets for evaluating intrusion detection algorithms in industrial SCADA systems by creating a labelled dataset from a real-world hardware setup involving Siemens PLCs and OPC UA protocol, including three types of attacks such as synthetic and live malicious behaviour, with sensors capturing temperature, sound, and flow data for correlation.
There are hardly any data sets publicly available that can be used to evaluate intrusion detection algorithms. The biggest threat for industrial applications arises from state-sponsored and criminal groups. Often, formerly unknown exploits are employed by these attackers, so-called 0-day exploits. They cannot be discovered with signature-based intrusion detection. Thus, statistical or machine learning based anomaly detection lends itself readily. These methods especially, however, need a large amount of labelled training data. In this work, an exemplary industrial use case with real-world industrial hardware is presented. Siemens S7 Programmable Logic Controllers are used to control a real world-based control application using the OPC UA protocol: A pump, filling and emptying water tanks. This scenario is used to generate application specific network data. Furthermore, attacks are introduced into this data set. This is done in three ways: First, the normal process is monitored and captured. Common attacks are then synthetically introduced into this data set. Second, malicious behaviour is implemented on the Programmable Logic Controller program and executed live, the traffic is captured as well. Third, malicious behaviour is implemented on the Programmable Logic Controller while still keeping the same output behaviour as in normal operation. An attacker could exploit an application but forge valid sensor output so that no anomaly is detected. Sensors are employed, capturing temperature, sound and flow of water to create data that can be correlated to the network data and used to still detect the attack. All data is labelled, containing the ground truth, meaning all attacks are known and no unknown attacks occur. This makes them perfect for training of anomaly detection algorithms. The data is published to enable security researchers to evaluate intrusion detection solutions.