Towards Robust Models of Cyber-Physical Systems

Cyber-Physical Systems (CPS) combine software with the physical world. For this purpose, CPS must model physical behavior in software. However, a software-based model cannot always accurately reflect the physical world. Often the model is a simplification of complex physical processes, or it suffers from measurement errors, or the physical side is subject to modifications and parameter drift, or the model is simply subject to misconceptions. It is an open research challenge how we can verify that physics and software-based model fit together. However, to rely on CPS in real-world scenarios we must ensure that physics and model are aligned. We propose a model formalism based on hidden Markov models that considers uncertainty and unknown phenomena and is robust enough to allow the analysis of CPS when working with error-prone data. More specifically, given observation data and an instance of the proposed model for a CPS (both of which may be flawed) the proposed formalism allows us to quantify the suitability between physics and model. If, however, a given model instance is deemed correct, the formalism enables methods which identify and smooth corrupt observation data as well as compute the most likely sequence of events for a given set of observations. Additionally, the formalism enables the learning of a suitable model according to given observation data. The model formalism will be tested with a simulation and a case study of an overhead traveling cargo crane system.