Resources

Case study

Water treatment facilities are essential to public health and environmental safety, providing clean and safe water to communities worldwide. The continuous and reliable operation of these facilities is crucial, as any interruption can lead to immediate and far-reaching consequences for public health and the environment. These plants utilise automation technologies, predominantly Programmable Logic Controllers (PLCs), to manage the multiple processes involved in treating and distributing water.

PLCs are integral to the automation of water treatment processes, controlling operations such as chemical dosing, filtration, and disinfection. They continuously receive and process data from various sensors that monitor critical parameters like flow rates, pressure, and water quality, adjusting the treatment processes in real time to ensure compliance with safety and quality standards. Despite the robust design and automation, water treatment plants are vulnerable to a range of issues, from physical component failures to sophisticated cyber-attacks. These threats can lead to operational disruptions or compromises in water quality, underscoring the need for swift and accurate anomaly detection and response mechanisms to prevent service interruptions.

The rise of data-centric anomaly detectors, often powered by Al, represents a significant advancement in monitoring and protecting critical infrastructure. However, the effectiveness of these Al systems is frequently undermined by false positives – erroneous alerts triggered by normal fluctuations in operational data. Such false alarms can lead to operational inefficiencies, as they compel operators to divert time and resources to investigate and resolve non-existent issues. This not only burdens the system but can also result in operators becoming desensitised to alarms, potentially overlooking real and serious threats.

Consequently, there is a pressing need for the next generation of Al-driven anomaly detectors that not only possess an ultra-high detection rate but also maintain an ultra-low rate of false alarms. Achieving this balance is essential for ensuring that operators can trust and respond effectively to the system’s alerts, thereby enhancing the security, reliability, and efficiency of water treatment operations.

Challenge

The AI.CI Challenge is a machine learning challenge where participants are tasked with developing algorithms capable of detecting anomalies in the operation of Secure Water Treatment (SWaT) testbed, an industrial-grade facility housed in Trust. Producing 5 gallons per minute (approximately 19 litres per minute) of treated water, this testbed replicates the complexities of large-scale water treatment facilities, serving as an invaluable resource for developing, testing and validating cyber security solutions.

Participants will have a unique opportunity to apply machine learning in a physical and realistic environment, to test and evaluate the efficacy and innovativeness of their Al-driven solution methodologies in detecting anomalies within a condensed timeframe.

Participants in the Challenge will be tasked with:

• Developing Advanced AI Solutions:
  Crafting Al algorithms and models capable of identifying and analysing potential cyber threats in real-time, thereby improving the robustness of OT and ICS environments.
• Enhancing Anomaly Detection:
  Creating predictive and reactive Al systems that can effectively detect and respond to unusual behaviour or potential threats, reducing the risk of cyber incidents and ensuring system integrity.
• Improving Threat Mitigation:
  Designing intelligent systems to detect threats that can help guide strategies to mitigate these risks, enhancing the overall security posture of critical infrastructure.
• Contributing to Cybersecurity Best Practices:
  Through their innovative solutions, participants will contribute to the body of knowledge in cybersecurity, helping to establish new benchmarks and best practices for Al applications in OT and ICS.