Abstract
As sensor and actuator networks mature, they are becoming a core utility of smart homes like electricity and water. As electricity enables running many electrical appliances in the home, the sensor and actuator networks enable the running of many Cyber-Physical Systems (CPSs) from different domains including energy, health, security, and entertainment. It has been hypothesized that integrating these CPSs in a smart home will offer innumerable advantages including achievement of the positive synergistic effects of the CPSs and avoidance of the negative consequences. However, integrating these CPSs is very challenging. Because, each individual system has its own assumption and strategy to control the physical world entities without much knowledge of the other systems. As a result, when these systems are integrated in a home setting without careful consideration, they raise many systems of systems interdependency problems.
In this thesis, we propose a utility sensing and actuation infrastructure for smart homes called DepSys that integrates many CPSs in a home by treating each system as an app. From the application layer to the link layer, DepSys provides the most comprehensive strategies to address a spectrum of dependencies including dependencies on sensor and actuator level control, sensor reliability status, involving human-in-the-loop, and on real-time constraints for delivering packets within latency bounds over wireless networks. We are the first to demonstrate that dependency metadata that focuses on the effect on the environment enables us to detect conflicts across devices that isn’t possible by monitoring actuations of individual devices as performed by the state of the art solutions. We also specify novel metadata that helps understanding context and resolving actuator level conflicts more accurately that couldn’t be achieved through existing priority based solutions. We develop a novel sensor failure detection scheme called FailureSense that addresses not only fail-stop failures, but importantly, obstructed-view and moved-location failures; that are realistic and common in smart homes and barely addressed to date in the literature or in real deployments. Our solution is the first one to detect these failures without requiring sensor redundancy and with minimal training effort. To address human-in-the-loop dependency, we develop EyePhy that uses a simulator to model the complex interactions of the human physiology using over 7800 variables. Using the simulator, EyePhy determines the potential conflicts among the human-in-the-loop apps’ interventions. It provides the most comprehensive dependency analysis to date by taking into account a wide range of physiological parameters and allows the dependency analysis to be personalized. Some smart home applications may have dependencies on real-time constraints, where packets need to be delivered within latency bounds reliably over wireless networks. To address this need, we perform a 21 days long empirical study, where we transmit 3,600,000 packets over every link of a 802.15.4 testbed to characterize wireless links. Based on the findings, we design a static network-wide stream scheduling algorithm that uses a novel least-burst-route to produce latency bounds of real-time periodic streams by taking into account link bursts and interference. In addition to addressing these various types of dependencies, DepSys handles the case when app developers fail to specify dependencies. Although DepSys is designed for smart homes, some of the its principles can be applied to other CPS application domains, e.g., industrial process control.
