OVERVIEW

Structural Health Monitoring (SHM) focuses on developing technologies and systems that assess integrity of structures. Most existing SHM implementations use wired data acquisition systems to collect vibration data from various locations in the structure induced by ambient sources for analysis. Installing a large scale wired data acquisition system may sometimes take several weeks and may often be turn out to be prohibitively expensive. Moreover, for old or damaged structures, instrumenting a large scale data acquisition system may not be possible for safety reasons. Our goal is to develop a wireless sensor network based data acquisition system which promises enormous benefits such as ease and flexibility of deployment and low maintenance and deployment costs.

We have deployed our system on the Four Seasons building. The Four Seasons building is a four-story reinforced concrete office building located in Sherman Oaks, California, which was damaged in the 1994 Northridge Earthquake and hence declared unsafe and scheduled for demolition. A series of forced-vibration experiments has been conducted by the UCLA before its demolition. Four Seasons building provided us with a unique opportunity to develop and test a wireless seismic monitoring array, through which we have explored the system issues behind the design of seismic monitoring and structural health monitoring system.

APPROACHes

We have developed Wisden, a self-configuring wireless sensor network system for structure response data acquisition. Wisden allows rapid, reliable and time-synchronized delivery of three-dimensional structural response to the base station. The hardware platform of the system is the Crossbow Mote (MicaZ/Mica2), together with a 3-channel, 16-bit vibration card (MDA400) and an accelerometer both by Crossbow Inc. We modified the vibration card firmware to support continuous sampling at 200Hz and resolve the conflict between the vibration card and the EEPROM on the Mica motes. In addition, Wisden consists of several software pieces:

1. Self-configuring multi-hop routing: Ease of deployment is a major advantage over wired approach. We used existing software component, Blast-1.0 multi-hop module by Berkeley, to build the communication tree with base station as the root.
2. Data compression: To reduce the data-rate and relieve the bandwidth limitations of the motes, Wisden uses run-length encoding to do silence period suppression. Also, Wisden uses onset-detection technique to accurately detect the active/silent period which allows the system to preserve the fidelity of the structural frequency response while effectively suppressing the inactive period.
3. Reliability: In a data acquisition system, reliability is of ultimate importance. Hence, one of our design goals is 100% reliability in recovering from wireless losses. Wisden implements a NACK-based hybrid hop-by-hop and end-to-end reliability scheme; the former is a necessary performance optimization in the lossy wireless environment. For the hop-by-hop reliability, nodes infer loss through a gap in the sequence number of sent packets. Nodes overhear transmissions and repair losses from a cache of recently forwarded packets. End-to-end reliability is required since hop-by-hop reliability scheme cannot recover losses when topology changes or when the packet cache overflows. For the end-to-end recovery scheme, a copy of every generated packet is also stored in the source node’s EEPROM for re-transmission in case of packet loss. Wisden’s base station keeps track of missing packets from all nodes. The base station initiates an end-to-end recovery by using the same mechanism as hop-by-hop recovery, per-hop NACKs.
4. Rate-limited transmission: The seismic traffic is bursty and has high data rate requirements. We used the 512K EEPROM to store compressed sample data, and a bandwidth allocation scheme to send out data at a limited rate, so as not to saturate the link toward the base station.
5. Time-synchronization: As opposed to traditional approaches for synchronizing the nodes to a global clock, we take a different light-weighted approach: synchronizing samples at the base station. The idea is to calculate the total residence time spent in the network by each packet and use this information to synchronize them. This approach incurs low overhead, yet is sufficient to align sample data in the base station.

SYSTEMS / EXPERIMENTS

Initial version of Wisden was built on Mica2 motes, using a 4-channel vibration board with 16bit ADC and an accelerometer, both from Crossbow. It supported acquisition of tri-axial data at sampling frequency of 50Hz from maximum of 10 nodes, and it used run-length encoding for data compression. But the series of real deployment experiences revealed the need for higher sampling frequency and loss-less compression. So, in the second version of Wisden, we have upgraded the system 1) to use MicaZ platform, 2) adopt a new onset-detection technique, and 3) support higher sampling frequency. The newer version allows data acquisition at sampling frequency of 200Hz from 15 nodes, without loss of data integrity during active period. And it also provides higher throughput and lower latency in data retrieval.

We first deployed Wisden in the Four Season building during the six force-vibration tests from July to August 2004. From the results of these tests, we re-designed the system to meet the real-world requirements and gained some satisfying data. We validated our result of the tests on July.22 and August.2 using the measurements taken from wired instrumentations by UCLA civil engineers. The experiment at the Four Seasons building (shown in figure 1) was conducted with 10 motes deployed in a 90x150 feet area on the fourth floor of that building. Four nodes were co-located with the wired instruments and others were carefully placed to achieve the goal of both data collection and multi-hop forwarding. The depth of the multi-hop tree established by our self-configuring system varied from 2~4 hops.

Figure 1. Four Seasons Building

Figure 2. A Wisden node co-located with wired accelerometer

Figure 3. Results from Wisden (Plot of vibration and the Frequency spectrum)

Figure 4. Results from wired infrastructure (Plot of vibration and the Frequency spectrum)

From this test, we were able to collect 7 minutes of forced vibration data within an hour with approximately 84% reliability. The collected data was reasonably accurate (shown in Figure 3) compared to the data collected by the wired infrastructure (shown in Figure 4). The loss of reliability in this harsh communication environment was mainly due to a small bug in our system which prevented the nodes from rapidly recovering the lost packets. As observed in the experiment, packet loss inside a structure tends to be very high. In some links it is common to have a loss rate of more than 30%. Topology changes are also frequent. This observation, together with the memory constraint on motes, implies that end-to-end recovery mechanism is necessary for the current hardware platform.

Based on the experiences gained from the Four Seasons building deployment, we were able to refine the system and made some improvements in Wisden. In addition to this, we have also upgraded Wisden onto MicaZ platform with new onset-detection technique.

For the second experiment, we deployed 15 motes on the Seismic Test Structure at USC civil engineering department to gain some experiences from real world deployment. The test structure is a platform for conducting seismic experiments on a full-scale realistic imitation of a 28’ × 48’ hospital ceiling. We were able to collect forced vibration data identical to that of a wired system (Figure 7, 8 and 9), with 100% reliability achieved from the network with 1~3 hop depth.

Figure 5. The Seismic Test Ceiling Structure

Figure 6. A Wisden node mounted on the structure

Figure 7. Structural response from the wired node

Figure 8. Structural response from Wisden

Figure 9. Frequency spectrum of the structural response from Wisden and the wired node

ACCOMPLISHMENTS

Software: Wisden, a wireless sensor network system for structure response data acquisition.

FUTURE DIRECTIONS

In the near future, we plan to deploy Wisden in the Factor building. This deployment will reveal the feasibility for deploying this network on a large structure. Also, we plan to extend Wisden to support multiple clusters for scalability.

PEOPLE

FACULTY

Prof. Deborah Estrin
Prof. Ramesh Govindan

GRADUATE STUDENTS

Jeongyeup Paek
Krishna Chintalapudi
Sumit Rangwala
Nupur Kothari
Ning Xu