Modal Analysis of the Yeondae Bridge using a Reconfigurable Wireless Monitoring System

In this study, a rapid-to-deploy wireless monitoring system is utilized to monitor the vibration response of the 180 meter long Yeondae Bridge, Korea. The bridge consists of a steel box girder section that continuously spans across five supports. Using 20 wireless sensors with MEMS accelerometers interfaced, the wireless monitoring system is installed to measure the bridge response to truck traffic. The elimination of wiring allows the wireless monitoring system to be reconfigured at will. The monitoring system sensors are installed using three different spatial configurations so that the acceleration response of the bridge can be measured at 50 sensor locations evenly distributed across the bridge length. Using an off-line outputonly modal identification technique, i.e. frequency domain decomposition (FDD), mode shapes were successfully obtained by combining acceleration data collected from the three system topologies. Mode shapes are compared to those predicted by a finite element model with excellent agreement found. SOURCE: Proceedings of the International Conference on Structural Safety and Reliability (ICOSSAR), Osaka, Japan, 2009. both commercial (e.g., Intel, Crossbow, Microstrain) and academic groups; interested readers are referred to a recent literature review of wireless structural monitoring by Lynch and Loh (2006). In this study, Narada, a low-cost wireless sensor designed explicitly for structural monitoring applications, is adopted for use in short-term dynamic testing of bridges (Swartz et al. 2005). The Narada wireless sensor (Figure 1) is assembled from commercial off-the-shelf components to form a low-cost, high-resolution wireless sensor node for installation in civil structures. The node can accommodate up to 4 sensors (e.g., accelerometers, strain gages, thermometers, among others) interfaced at one time; interfaced sensor outputs are locally digitized using a 16-bit analog-digital converter (ADC). Data collected is managed by an Atmel Atmega128 microcontroller. This specific microcontroller has a relatively large amount (128 kB) of embedded memory for the storage of software. To accommodate sensor data storage, an additional 128 kB of RAM memory is included in the wireless sensor design. For wireless communications, the Chipcon CC2420 transceiver is also integrated. This specific radio operates using the IEEE 802.15.4 protocol allowing it to interoperate with other IEEE 802.15.4-complient wireless sensors such as the Imote2 (Nagayama and Spencer 2007). Finally, a 2channel, 12-bit digital-to-analog converter (DAC) offers the wireless sensor a capability to operate actuators for structural control applications. Included in the Narada wireless sensor design is a fully functional embedded operating system that automates device operation (e.g., collect and communicate data) and data interrogation (e.g., modal analysis and damage detection). 3 MODAL ANALYSIS TECHNIQUES FOR CIVIL STRUCTURES 3.1 Output-only modal parameter estimation Classical modal analysis generally employs controlled excitation (i.e., inputs) of a test structure with responses (i.e., outputs) measured. Based on inputoutput data, modal parameters are estimated including modal frequencies, mode shapes and modal damping ratios (Ewins 2000). Such an approach is applicable to structures tested in the controlled laboratory environment. However, output-only modal parameter estimation has received considerable attention in the civil engineering domain because of the difficulties associated with exciting operational structures in a controlled manner. Furthermore, natural excitation of the structure is difficult to employ in input-output modal analysis due to the challenge inherent in precisely characterizing the input. Output-only modal analysis employs the underlying assumption that the input excitation of the system is broad-band white noise. Output-only modal parameter estimation techniques are categorized into two distinct groups depending on their analysis domains: frequency domain methods dealing with frequency response functions (FRFs) or output spectra; time domain methods using covariance of past and future outputs. In this study, a frequency domain technique called frequency domain decomposition (FDD) is adopted (Brinker et al. 2001). 3.2 Frequency domain decomposition (FDD) FDD is a sophisticated frequency domain modal identification method that is capable of accurately identifying the real and imaginary components of closely spaced modes. Since the early 1980s, the decomposition of output spectra using singular value decomposition (SVD) has been studied (Peeters and Ventura 2003). Shih et al. (1988) applied SVD to decompose system FRFs to indentify complexvalued modes using input-output data sets. Later, Brinker et al. (2001) reformulated their approach by using the power spectral density (PSD) functions of the system output; the approach was named frequency domain decomposition. The relationship between the system input, ) (t x , and the measured output, ) (t y , can be expressed in the continuous-time frequency domain as follows: ) ( ) ( ) ( ) (     j H j G j H j G H xx yy  (1) where ) (  j Gxx is the PSD matrix of the input, ) (  j Gyy is the PSD matrix of the output, ) (  j H is the FRF matrix, and ) (  j H H is its complex transpose conjugate of ) (  j H . If the system input, ) (t x , is white noise, ) (  j Gxx will simply be a constant matrix; hence, ) (  j Gyy is directly proportional to the product of FRFs, ) ( ) (   j H j H H . By applying SVD to ) (  j Gyy , it can be decomposed as ) ( ) ( ) ( ) (     j U j j U j G H yy   (2) where ) (  j U is a complex-valued matrix containing singular vectors as columns and ) (  j  is a realvalued diagonal matrix containing positive singular Figure 1. The Narada wireless sensing unit with its key hardware components highlighted.