Self-aware reliable monitoring

Abstract

Cyber-Physical Systems (CPSs) can be found in almost all technical areas where they constitute a key enabler for anticipated autonomous machines and devices. They are used in a wide range of applications such as autonomous driving, traffic control, manufacturing plants, telecommunication systems, smart grids, and portable health monitoring systems. CPSs are facing steadily increasing requirements such as autonomy, adaptability, reliability, robustness, efficiency, and performance.

A CPS necessitates comprehensive knowledge about itself and its environment to meet these requirements as well as make rational, well-informed decisions, manage its objectives in a sophisticated way, and adapt to a possibly changing environment. To gain such comprehensive knowledge, a CPS must monitor itself and its environment. However, the data obtained during this process comes from physical properties measured by sensors and may differ from the ground truth. Sensors are neither completely accurate nor precise. Even if they were, they could still be used incorrectly or break while operating. Besides, it is possible that not all characteristics of physical quantities in the environment are entirely known. Furthermore, some input data may be meaningless as long as they are not transferred to a domain understandable to the CPS. Regardless of the reason, whether erroneous data, incomplete knowledge or unintelligibility of data, such circumstances can result in a CPS that has an incomplete or inaccurate picture of itself and its environment, which can lead to wrong decisions with possible negative consequences.

Therefore, a CPS must know the obtained data’s reliability and may need to abstract information of it to fulfill its tasks. Besides, a CPS should base its decisions on a measure that reflects its confidence about certain circumstances. Computational Self-Awareness (CSA) is a promising solution for providing a CPS with a monitoring ability that is reliable and robust — even in the presence of erroneous data. This dissertation proves that CSA, especially the properties abstraction, data reliability, and confidence, can improve a system’s monitoring capabilities regarding its robustness and reliability. The extensive experiments conducted are based on two case studies from different fields: the health- and industrial sectors.