: : NCASSR : Projects

Model-Based Data Aggregation for Large-Scale Sensor Networks

Advances in sensor networks will allow the use of cyber infrastructure in protecting civil infrastructure. For example, sensors attached to a building can monitor the health of the structure; actuators attached to nodes can then be used to dampen vibrations caused by an earthquake or an explosion, increasing the likelihood of the structure withstanding the damage. In order to realize such applications, a key problem that needs to be addressed is data management: a sensor network generates large amounts of data, which has be collected and analyzed collectively. Transmitting such data to a central node is prohibitively expensive, both in the energy required for communication and the time it would take because of the limited bandwidth. In particular, network traffic near the gateway nodes becomes disproportionately heavy, resulting in a bottleneck. This track pattern reduces the lifetime of the nodes near the gateways and also hinders the scalability of the network.

We will develop model-based data aggregation as a way to address the problem. The essential insight is that we can often build a model of the behavior of a sensing domain that can predict the evolution of the sensor readings over time. For example, applications may expect constant measurements, a linear evolution, observations that are a function of a stochastic variable, or periodicity. By increasing the priority given to the sensor data showing significant deviation from the model, we can reduce the network traffic problem. Moreover, the models can be dynamic--for example, allowing intelligent sensor nodes to locally estimate future values based on a model with evolving parameters and communicate only the deviations from the model rather than raw data. Such networks will have more balanced lifetime and will be scalable. We propose to investigate particular applications of this methodology to facilitate the use sensor networks to improve the safety of the civil infrastructure.

We will to use the structural health monitoring application to drive the development of our model-based data aggregation service. This involves developing application-specific distributed data models and aggregation functions which make use of our service, and generalizing them to classes of problems with similar characteristics. Application demands and experimental performance will help us better understand the requirements of the data aggregation service in general. In addition to application-specific data aggregation, we will explore various application-independent schemes for computing the priority of the aggregate messages. This will also be useful as a foundation for applying model-based data aggregation to future sensor network applications.

Studies of node failures and mobility in data aggregation routing trees will lead to a more robust data aggregation algorithm. In particular, we aim to minimizes recovery latency and the impact of faults on sensor data quality, in real application terms. We will focus on stable methods to find new routes for data in the event of failure. Key factors include power and network capacity and the ability of nodes to efficiently aggregate data along the new route. We will compare the impact of failures on the aggregation service's performance under a variety of reconfiguration approaches and application scenarios.

Through simulation and experiments we will assess the power, communication, and processing load incurred by the data aggregation service. This information will help us build a more efficient and power-aware system. We intend to combine prior work on energy efficiency, latency, and fault tolerance in aggregation trees with application-specific properties. For example, the application-specific priorities assigned to data may be used to influence the structure of the aggregation tree. The demands of our data aggregation service on individual nodes will also help us assess its efficiency for the network as a whole.

Consider a wireless sensor network (WSN) deployed on a bridge, with sensor nodes located at joints and along beams. Under normal conditions, the bridge has a low-amplitude ambient vibration pattern. Sensor nodes may measure their acceleration and calculate the Fourier transform of the data to determine the dominant modes of vibration. If this data does not change, the sensors do not need to communicate. However, wear and tear over time may cause the vibration frequency profile to change in sections of the bridge. In this case, the sensors detecting a change need to inform maintenance personnel so that the structure can be repaired before it degrades significantly further.

Project Leads

Gul Agha, UIUC CS Deptartment

Return to Projects list

SELS 0.7 released

Secure Email List Services (SELS) is an open source software for creating and developing secure email list services among user communities.

Strong community engagement strengthens cybersecurity research and development

NCASSR-supported exploratory research at NCSA and elsewhere has sparked additional external funding and development opportunities as well as successful deployment and adoption by users ranging from the defense sector to state law enforcement to the utilities industry.

NCASSR Collaborator Goes To Washington

Carl Gunter, a professor in the University of Illinois Department of Computer Science and a project lead on NCASSR-supported work involving adaptive, secure messaging, recently spoke to an audience of congressional staffers and lobbyists on Capitol Hill regarding ways to address a variety of critical cybersecurity issues in areas such as healthcare and energy distribution.