Edge Intelligence for Safety Assurance of Time Critical Cyber Physical Systems

Cyber-Physical-Systems (CPS) harbor the potential for vast economic and societal impact in domains such as mobility, home automation, energy, agriculture, and delivery of health. The open and cooperative nature of CPS allows multiple heterogeneous systems to collaborate with each other to achieve some common goals. One thing many modern autonomous CPSs have in common is that they use real-time artificial intelligence decision feedback, e.g., by using machine learning, to achieve their operational goals. Such systems continuously learn from their operations and adapt their behaviour accordingly; as a result, their emergent behaviours are difficult to predict from the simple superposition of the behaviour of the individual system elements.

As many CPSs are safety-critical in nature, their failure may cause great harm to people and lead to the temporary collapse of important infrastructures with catastrophic results for industry and society. Therefore, ensuring their dependability (e.g. safety, reliability, availability) is the key to unlocking their full potential. Traditional safety assurance is defined as design/development time activities that are performed to support the overall claim that the system will be safe to operate. Design-time assurance of non-functional system properties such as safety and reliability would no longer be valid for the autonomous CPSs since it is based on outdated, static assumptions about constantly evolving behaviour. To provide continuous safety assurance of the adaptive CPSs, researches have been performed to provide runtime safety guarantees, where evidences are collected at runtime from various sources and processed to provide runtime safety guarantees.

The provision of runtime safety guarantee requires continuous real-time artificial intelligence decision feedback, which is often a time critical task for safety critical applications. Neither the on-board machine learning algorithms nor the cloud-based solutions can meet the strict time constraint and reliability requirements of the decision making process. The emergence of relatively newer cloud-based technologies such as edge and fog computing has the potential to address these challenges by bringing the computation and communication capabilities closer to the CPSs, where some of the workload can be offloaded to the edge devices.

The primary goal of this project is to develop an edge/fog computing-enabled framework to provide continuous runtime safety assurance for the CPSs by meeting the stringent latency and service reliability requirements. The framework will combine advanced software engineering techniques with emerging artificial intelligence methods, which include “models@runtime”, machine learning, bio-inspired metaheuristics, and data analytics. This will facilitate the dependability analysis of

distributed or dynamically reconfigurable cyber-physical systems, e.g. convoys of self-driving vehicles that have to talk to each other but the configurations cannot be predicted in advance.