Semantic Networks for Hybrid Processes.

Abstract

Simulation models are often used in parallel with a physical system to facilitate control, diagnosis and monitoring. Model based methods for control, diagnosis and monitoring form the basis for the popular sobriquets `intelligent', `smart' or `cyber-physical'.

We refer to a configuration where a model and a physical system are run in parallel as a emph{hybrid process}. Discrepancies between the model and the process may be caused by a fault in the process or an error in the model. In this work we focus on correcting modeling errors and provide methods to correct or update the model when a discrepancy is observed between a model and process operating in parallel. We then show that some of the methods developed for model adaptation and diagnosis can be used for control systems design.

There are five main contributions.

The first contribution is an analysis of the practical considerations and limitations of a networked implementation of a hybrid process. The analysis considers both the delay and jitter in a packet switching network as well as limits on the accuracy of clocks used to synchronize the model and process.

The second contribution is a semantic representation of hybrid processes which enables improvements to the accuracy and scope of algorithms used to update the model. We demonstrate how model uncertainty can be balanced against signal uncertainty and how the structure of interconnections between model components can be automatically reconfigured if needed.

The third contribution is a diagnostic approach to isolate model components responsible for a discrepancy between model and process, for a structure preserving realization of a system of ODEs.

The fourth contribution is an extension of the diagnostic strategy to include larger graphs with cycles, model uncertainty and measurement noise. The method uses graph theoretic tools to simplify the graph and make the problem more tractable and robust to noise.

The fifth contribution is a simulation of a distributed control system to illustrate our contributions. Using a coordinated network of electric vehicle charging stations as an example, a consensus based decentralized charging policy is implemented using semantic modeling and declarative descriptions of the interconnection network.