Statistical Approach to Structural Monitoring Using Scalable Wireless Sensor Networks.

A wireless sensor network (WSN) for structural monitoring applications was designed, developed, and deployed to investigate scalability of the network and the effect of dense spatial and temporal sampling on statistical characterization of structural vibration modes. A sensor board based on micro-electro-mechanical-systems (MEMS) technology with dual accelerometers in two directions was developed to provide a wide range of sensitivity with acceptable resolution. The sensor board integrates a MICAz mote micro-controller for control and communication to form a node in a WSN. In the software architecture, the requirement of high-frequency sampling for structural monitoring is the critical challenge for temporal scalability of the network. TinyOS, the widely used operating system for MEMS-based devices, was modified significantly, in coordinated research, to impose an acceptable threshold on the time-synchronization error (jitter). Using a recently developed communication protocol for reliable data collection and command dissemination, a new data pipelining technique was implemented to reuse the spatial bandwidth of a multi-hop wireless network. For initial analysis of the WSN, the integrated hardware and software system was deployed and evaluated on a pedestrian bridge. ARMA models with ambient data were used to identify modal properties of the bridge and calibrate a finite element model of the structure. Large-scale deployment of the WSN on the Golden Gate Bridge consisted of sixty-four nodes on the main-span and south tower. The three-month deployment, the largest of its kind to date, provided ambient vibration data that was used to identify modal properties of the bridge. The superior performance of the network was demonstrated by maintaining a bandwidth of 550 bytes/sec throughout forty-five wireless hops. Using the spatially dense network, sixty-seven vibration modes of the bridge were identified. Multiple data sets were used to estimate statistical properties of the identified modes as histograms and confidence intervals. Comparisons are made between modes identified by alternative system realization methods (spectral methods), ambient data from a previous instrumentation of the bridge, and finite elements models. The advantage of dense sensor networks is further demonstrated through a novel algorithm for local damage detection, which monitors the influence coefficients between signals collected by the sensors around critical points of a structure, and identifies damage using a Bayesian statistic with the desired confidence level.