Statistical Methods for a Sensor Rich Building
Advisor: Mark Hansen
Our goal in this dissertation is to study the data coming from a sensor network. In particular, we examine a network of seismic sensors in a building. The Factor building has been outfitted with a network of 72 accelerometers to records its movements, and this seismic array in the building generates data at a ferocious rate. This illustrates the volume of data that can be expected in sensing applications, and the kinds of rich stochastic structures that might be extracted from the sequences of measurements.
We have performed a set of analyses and experiments to explore the signal coming from this sensored environment. First, preliminary statistical analyses are performed on the data streams generated by this network. Various methods are emplyed to understand the earthquake and ambient vibration data in the civil engineering context. In this stage, we evaluated what we learned from the data to decide how to improve data collection and analysis. This phas of our work was exploratory in nature.
Next, we install a secondary sensor system to add values to the existing seismic sensors, to make use of the elevators as a previously unexamined source of ambient vibrations. From a structural health monitoring perspective, while this excitation source is repeatable, its spectral characteristics are not known. We propose that this signal could offer us an opportunity to detect structural changes in a building in real time, a task that is often infeasible using artificial (known) excitations. We have studied both elevator counterweight vibration and structural motion to formulate and learn statistical models related to structural characteristics of the building.