Modular Design of Monitors for Cyber-Physical Systems from Formal Specifications
thesisposted on 21.10.2015 by Ruggero Balteri
In order to distinguish essays and pre-prints from academic theses, we have a separate category. These are often much longer text based documents than a paper.
Robotics is a field that is subject to rapid evolution, with its ground-breaking innovations becoming increasingly more present in our lives. In the past, automation was designed to operate in a predetermined and completely controlled environment (e.g. assembly lines). The control system was sufficiently easy to model in an exhaustive way and reasonably less prone to unpredicted failures. Nowadays, the attention of the international community is shifting towards a new implementation of systems that can operate in an unstructured environment, inherently unpredictable and where everything is subject to change. In this framework, a complex system may have countless number of states and checking its correct behaviour against all the possible inputs has been proven, in general, theoretically undecidable. The probabilistic nature of the system and of the environment need to be considered. To monitor the correct behaviour of a system, an on-line algorithm that takes into account a probability of transition (not only a deterministic transition) of the system model has to be employed in real-time. The main goal of the entire research was to implement and test, on a realistic system, a valid monitoring methodology that could take into account the limits provided by inaccurate models and use that retrieved data to design better and more reliable systems. Particular emphasis has been paid to implementing a physical platform that could be used as initial prototype. Once a working robot has been correctly engineered, several monitor implementations have been designed and tested on it. All the retrieved data has been subsequently imported into Matlab and then analysed. In this way, it was possible to perform a number of tests to investigate several problems and, among the most relevant ones, the influence of inaccurate system models inside the monitor played a central role. In parallel, special attention was also given to the influence in performance and accuracy of alternative ways in implementing monitors.