Wireless sensor network technology is being considered for many real world applications that require high reliability and long lifetimes. For example, new, low cost wireless sensor networks (WSN) can be embedded into large city skyscrapers to support fire detection and reaction. Such systems must reliably detect a fire on any floor, activate alarms, notify fire stations, and announce and illuminate egress routes. These buildings are passively monitored for hazards and are largely unattended. However, such systems require high confidence in their operation and must also be able to demonstrate that they are operational on a periodic inspection basis (at a minimum). This project determines how to specify and support, at runtime, a collection of solutions that enable embedded systems to improve confidence and demonstrate application operability.
The project is novel in several ways. First, it develops a requirements language that permits designers to specify, via a combination of declarative statements, invariants, and rules, the runtime assurances required for high confidence. The language addresses application semantics, the statistical nature of WSN, costs, future predictions on system performance, and monitoring needs for various mechanisms. It also permits automatic code generation. Second, a runtime assurance methodology and framework is developed that supports specific demonstrations of a system?s key functional capabilities on demand and offers a well defined set of diagnosis capabilities including data mining when the system fails to meet its assurances. Third, various runtime mechanisms are created and used in novel ways including virtual event generation and real event replay. Fourth, as a system evolves solutions for understanding the system model as applied to controller design are developed. Fifth, an implementation and evaluation in an application domain is undertaken. Broad impact of the work is possible because similar issues exist for WSN applications in industrial plants, home and assisted living health care, and transportation. A set of course modules is developed and incorporated into two current course offerings at the University of Virginia: Wireless Sensor Networks and Cyber Physical Systems. The corresponding teaching materials (slides and labs) are available for use at other Universities via the Web. The School of Engineering Office of Minority Affairs is used to match minority students with this research.
With the proliferation of inexpensive sensors, sensors are increasingly being used in smart homes for medical and energy savings applications. Most previous work has focused on short term deployments under many assumptions about the environment and how people use the system. Our objectives are to develop more robust and fault tolerant smart home wireless sensing systems for practical long term and unattended deployments. Three main software systems, primarily in the home health care medical domain, have resulted from our work: Empath, Holmes, and Vocal-Diary. Home health care sensing systems are projected to streamline the efficiency of the practice of medicine by decreasing the costs of senior care and by providing preventative care to keep people out of hospitals and nursing homes. Most current sensing systems are not yet flexible enough nor handle enough realism to easily handle widely different applications, or add new modalities into the system, or to leverage the cloud’s horizontal scalability for storage, analysis, and display of this data. Our system, called Empath, provides a flexible three layer framework that uses the Cloud and can easily be instantiated for different home health care applications. To demonstrate the flexibility of the framework, Empath was instantiated for three different purposes: monitoring sleep behaviors and stress for those who suffer from severe epilepsy, supporting a study that assesses the relationship of incontinence with sleep agitation for those suffering from dementia, and monitoring for depression. These applications of our platform were performed with the University of Virginia School of Nursing. Empath’s potential impact is that it has the capability to be used in many home medical studies and as a monitoring platform for the elderly, thereby increasing medical knowledge and lowering costs. Many home health monitoring systems attempt to discover models of normal behaviors and then detect anomalies. In most cases realisms are not address well enough and the systems generate too many false alarms. We developed Holmes, a comprehensive anomaly detection system for daily in-home activities. Holmes accurately learns a resident’s regular behavior by considering normal variability in daily activities based on specific days of the week, combining activity instances of a day / multiple days together to find the features that best represent regularity, and by detecting temporal and causal correlations among multiple activities. Having accurate models for what is normal behavior is a critical first step. As an example, we found that people need 10-15 different models to characterize normal behavior, not just one model often used in many systems. Then, based on resident and expert feedback, Holmes learns semantic rules that explain specific variations in daily activities in specific scenarios. Together the comprehensive modeling of regular behavior and the set of semantic rules help Holmes to reduce false alarms which is essential for the reliability of any anomaly detection system. The Holmes solution supports a robust and useable home health care system. In evaluating technology for real deployments, a universal problem is obtaining ground truth. To solve this problem we developed Vocal-Diary, a voice command based ground truth collection system that uses grammar based commands from residents to log start and end of activities. Vocal-Diary ensures robustness in the presence of sounds from different environmental noise and day-to-day conversation by using two-way acknowledgement and integrating speaker recognition in the pipeline. Vocal-Diary also utilizes the sensor data produced by the underlying activity recognition system to query residents periodically to check if they forgot to log any activity. Evaluation shows that Vocal-Diary increases precision by at least 40% and recall by at least 10% compared to a system that uses voice command recognition without any acknowledgement and speaker recognition. Vocal-Diary can have significant impact by providing a ground truth system instead of having each research group develop their own.
