Building Distributed Data Models in a Performance-Optimized, Goal-Oriented Optimization Framework for Cyber-Physical Systems

Abstract

Cyber-physical systems (CPS) are large, distributed embedded systems that integrate sensing processing, networking and actuation. Developing CPS applications is currently challenging due to the sheer complexity of the related functionality as well as the broad set of constraints and unknowns that must be tackled during operation. Building accurate data representations that model the behavior of the physical environment by establishing important data correlations and capturing physical laws of the monitored entities is critical for dependable decision making under performance and resource constraints. The goal of this thesis is to produce reliable data models starting from raw sensor data under tight resource constraints of the execution platform, while satisfying the timing constraints of the application. This objective was achieved through adaptation policy designs that optimally compute the utilization rates of the available network resources to satisfy the performance requirements of the application while tracking physical entities that can be quasi-static or dynamic in nature. The performance requirements are specified using a declarative, high-level specification notation that correspond to timing, precision and resource constraints of the application. Data model parameters are generated by solving differential equations using data sampled over time and modeling errors occur due to missed data correlations and distributed data lumping of the model parameters.