A Perfect Match: How Secure AI Can Shape the Future of Manufacturing
The advent of next-generation smart connectivity systems has enabled the collection of rich datasets, as part of the manufacturing process, which has emerged as a key enabler for a wide gamut of safety-critical applications ranging from improving predictive maintenance to enhancing cyber-security to enabling remote machine calibration and human-robot collaboration. A key enabler in this direction is the emergence of Artificial Intelligence technologies that can make manufacturing safer and more efficient.
Despite the clear advantages that such unprecedented quantity of data brings forth, it is also subject to inherent data trustworthiness challenges due to factors such as malevolent input and faulty sensors. With constricted demand for manufactured goods and more reliance on remote solutions, preempting machine failure and reinforcing security has never been more important. One of the main hurdles to actively gauging factory assets’ health is on the validation of the received data correctness; especially considering the vast amount of available machine data towards predicting potential failures. Industrial control systems also remain vulnerable to cyber attacks.
Promising advances in artificial intelligence and machine learning are enhancing the way manufacturers prevent asset failure, block cyber threats, and autonomously calibrate machines. There has been a plethora of proposed solutions, based on the use of traditional machine learning algorithms, towards assessing and sifting faulty data without any assumption on the trustworthiness of their source. However, there are still a number of open issues: how to cope with the presence of strong, colluding adversaries while at the same time efficiently managing this high influx of incoming machine data? Designing such trust anchors is not an easy task: it requires fusing (contradictory) data, originating from untrustworthy sources describing dynamic and uncertain phenomena evolving over space and time.
Particularly with respect to safety and security, system components (managing machine data assets) must be enabled to make and prove statements about their state and actions so that other components can align their actions appropriately and an overall system state can be assessed and security policies can be evaluated and enforced. This goes substantially beyond simple authorisation schemes telling who may access whom but will require understanding of semantics of requests and chains of effects throughout the system and an analysis both statically at design-time and dynamically during runtime. The latter will then even allow to conduct dynamic Risk Assessment (RA) and decide at runtime if an entity is still safe to be used (even if some components are compromised and fail) or needs to be shut-down in a failsafe state.
Compounding this issue, the STAR ICT-38-2020 project focuses on leading the design of AI-enabled data verification frameworks with wide applicability in manufacturing environments. It will employ deep learning techniques in the form of neural networks to address the issue of possible data poisoning and evasion attacks in smart manufacturing environments while demonstrating high accuracy and scalability.
With this, we claim that manufacturing supply chains, in general, can withstand even a prolonged siege by a pre-determined attacker with known or unknown capabilities as the system can dynamically adapt to its security and safety state. This is substantially more flexible than traditional security mechanisms that often try to maintain and enforce pre-defined policies using rather static security mechanisms. Even more, STAR’s intelligent multi-layered framework (including Explainable AI, Active Learning, Simulated Reality, Human-Centric Digital Twins for AI) allows a very high degree of automation, something that is definitely required in manufacturing scenarios where the mere number of devices will prohibit human intervention for security management.