Telegraph Road Bridge
The Telegraph Road Bridge (TRB) is a multi-girder composite steel bridge built in 1973 in Monroe, MI and designed to carry two lanes of I-275 northbound traffic. Monroe is located in the southeastern corner of Michigan and therefore sees extremely heavy truck loads on its roads and bridges due to the large manufacturing industry in the area and its close proximity to the Canadian border in Detroit. The total length of the TRB is approximately 224 ft. (68.28 m) and is designed with three main spans. The end spans are each 48 ft. (14.63 m) long and span from the bridge abutments to support piers with the their spans cantilevering 6 ft. (1.83 m) past the interior piers. The main span of 140 ft. (42.67 m) suspends from the cantilevered end using pin and hanger assemblies. The bridge has 7 steel girders in composite action with an 8 in. (20.32 cm) reinforced concrete deck. The bridge is also ske wed with a skew angle of 57 degrees. A plan and elevation view of the TRB is presented in Figure 1. TRB is owned and managed by the Michigan Department of Transportation (MDOT).
The TRB has shown signs of deterioration from its 40 year service life. Specifically, the deck was severely cracked with potholes located uniformly across its top deck surface. The bridge has also experienced fatigue cracks in the girder webs in a number of locations where the transverse cross frames are welded to the girders. In addition, some section loss has been witnessed in one girder on the lower flange of the girder immediately above its bearing. Finally, a section of the concrete abutment close to the support of a girder had spalled off requiring a temporary steel column to be installed to shore up the girder. In September 2011, the bridge underwent rehabilitation with a new deep concrete overlay on the deck installed, bridge expansion joints replaced, and abutment structures repaired. In addition, the structural steel was inspected with a new zinc-based coating applied to the steel areas showing wear. A number of deterioration processes are of primary concern during the decision making process for aging steel girder-concrete deck bridges like TRB. Pin and hanger assemblies are a major concern because they are fracture critical components that lack redundancy in their design. Fatigue life monitoring using strain measurements is a high priority for fracture critical bridge components . In addition to monitoring pin and hanger connections for fatigue, the integrity of the bridge deck is another major concern. Failing expansion joints can lead to extreme thermal stress and corresponding cracks in the concrete deck. Temperature and strain are key bridge response parameters to be measured to understand the behavior of the bridge deck. The degree of composite action between the bridge deck and steel girders is also of great interest to the bridge engineer. For example, loss of composite action can be a serious issue for the bridge performance leading to unsafe conditions. To assess the degree of composite action, strain measurements through the cross section of the bridge superstructure will be used to identify the neutral axis of the section. Neutral axes below their theoretical location (assuming full composite action) would indicate loss of composite action in the section. These deterioration processes will guide the installation of sensors on the TRB.
In addition, the bridge owner desires an accurate analytical model of the bridge. Towards this end, sensors will be installed to provide dynamic response data for the updating of finite element (FE) models . Initial FE models developed from structural drawings rarely exhibit the same dynamic behavior observed in the field ,. Hence, their calibration and updating is typically done using modal information extracted from field data to correct for model inaccuracies due to differences in boundary conditions, material properties, among other factors.
Architectural Overview of the SHM System
The proposed cyber-enabled wireless bridge monitoring system consists of two major tiers: the wireless sensor network deployed in the bridge is defined as the lower tier of the system while the host of cyberinfrastructure tools created to manage and process bridge response data populates the upper tier. As shown in Figure 2, the on-line cyberenvironment is based on a scalable database system (SenStore) accessible via the Internet. Sensor data collected by the wireless sensor network can be written to SenStore through a cellular access point to the Internet.
A network of Narada wireless sensor nodes and an on-site server manage the sensor system used to monitor the bridge. The Narada wireless sensor node is shown in Figure 3(c) and was developed at the University of Michigan. The Narada uses an Atmel Atmega 128 microprocessor with 128 kB of external SRAM for data storage and computation. Wireless communication is realized by the inclusion of the Chipcon CC2420 IEEE 802.15.4 wireless radios, making the unit exceedingly versatile for developing large-scale wireless sensor networks. The sensing nodes deployed on the TRB use an enhanced power-amplified version of the CC2420 that provides an increased communication range that exceeds 700 m. The Narada utilizes a four channel, 16-bit Texas Instruments ADS8341 analog to digital converter (ADC) for data acquisition and a two channel, 12-bit digital to analog converter (DAC) for issuing command signals. Prior to field deployment, the Narada node is housed into a water tight enclosure inclusive of a lead acid battery, sensor signal conditioning circuit, and power harvesting circuit (Figures 3(a) and 3(b)). Each sensor is powered by a small multi-crystalline solar panel continuously recharging the lead acid battery. Narada sensing nodes installed on the bridge are managed by a base station consisting of a small portable computer seen in Figure 3(d). The on-site base station houses a PC-104 single board computer, Chipcon CC2420 transceiver, SunSaver 10L Charge Controller, PowerSonic PS-12350NB 12V 35Ah sealed lead acid rechargeable battery, 12V-5V DC/DC converter, and Sprint 3G Sierra Wireless 250U USB cellular modem. The base station is fully powered by a UL-Solar 110W 12V multicrystalline solar panel. The server system, encased in a large water tight enclosure, is mounted to the south fascia of the bridge deck as seen in Figure 4(a). Sensor data is collected locally from the wireless sensor nodes by the PC-104 single board computer using the CC2420 receiver interfaced to one of its USB ports. Data collected is temporarily stored in the server memory until it is transmitted off-site to a SenStore server at the University of Michigan in Ann Arbor, MI using the Sprint 3G cellular modem.
For the TRB deployment, the system is configured to record acceleration and strain data from the bridge. Raw sensor data is collected periodically on 4 hour intervals, sampled at 200 Hz for 60 second durations for acceleration response data and 100 Hz for 20 seconds for strain response data. System end users though have the ability to remotely connect to the bridge base station to implement different data collection strategies.
The bridge wireless sensing system is part of a larger Internet-based SHM framework. The cyberenvironment is constructed in two tiers. The lower tier contains the wireless sensor network which is dedicated to collecting response data (e.g., acceleration, strain) and environmental data (e.g., temperature). The upper tier operates around the SenStore centralized data repository  which provides a specialized database architecture for managing bridge data as well as a standardized platform for data sharing among various data processing applications. As seen in the schematic of the cyberenvironment architecture in Figure 2, clients seeking access to system data include bridge inspectors seeking aid in observing structural behavior, researchers and bridge engineers interested in structural analysis of field data, and bridge managers responsible for decision making. Two way communications between the lower and upper tiers is managed through the SenStore system. Two way communications is a powerful characteristic of the cyberinfrastructure as it presents the opportunity for autonomous modification of the bridge sensing scheme based on the post-processing of existing data. For example, sensor data suggesting structural degradation in a specific region could trigger a redistribution of system resources for closer monitoring of the area of interest.
Field Deployment on TRB
Sensor deployment on the TRB started in September 2011. The Michigan Department of Transportation (MDOT) provided assistance in the sensor and server installations by providing equipment (e.g., lift trucks) and personnel for equipment operation and traffic control. In total, 18 accelerometers and 18 strain gauge sensors have thus far been installed on TRB as shown in Figure 1. All sensing units are contained in a polycarbonate weatherproof enclosure housing a Powersonic PS-832 8V sealed lead acid rechargeable battery and power harvesting board utilizing the Linear Technology LT3652 solar powered battery charger. Solar power is provided to each module for battery charging by UL Solar 10W 12V Monocrystalline Silicon PV panels removing the needed for site power or frequent physical intervention (e.g., battery replacement). Sensing modules and solar panels are magnetically mounted to the bottom flange of the steel girders for a well fixed but easily modified deployment, shown in Figure 4(b). The Narada takes advantage of the Atmega128 brown-out detection and sleep mode functions, necessary for sustained operation even with the use of solar power harvesting. The battery life for the acceleration and strain gage sensing modules are approximately 38 hours and 18 hours, respectively. Microprocessor sleep mode functions allow the unit to be partially shut down between sensing cycles, increasing the battery life by 2 to 3 times. An external MOSFET sensor power switch is triggered on or off by the Narada DAC before and after sampling, greatly reducing power consumption by the sensor during idle periods. These combined schemes used to reduce power consumption increase the battery life of the sensing module by 30 to 40 times.
Accelerometers are positioned around the outside perimeter of the main span for the purpose of identifying modal
parameters. Silicon Design 2012-002 uniaxial accelerometers are installed on girders #1 & #7, represented by
sensors #1 through #18 in Figure 1. The accelerometers have a range of +/- 2g with a sensitivity of 1V/g. Each
accelerometer requires one Narada wireless sensor node thereby requiring 18 Narada modules in total for this
effort. All acceleration signals are conditioned with a 25 Hz low pass filter and
5 times amplification. Frequency Domain Decomposition (FDD)  is used to
identify bridge modal frequencies and shapes. The modal information is fed
into a finite element model implemented in SAP for model calibration and updating.
Strain gages are positioned along the height of the steel girders to measure axial strain developed in the beams due to bending. Tokkyo Sokki FLA-6-11-3LT 6 mm 120 ohm metal foil strain gages are located over the piers and at mid span on girders #2 & #6 towards the effort of composite action assessment. Strain gages are numbered #19 through #36 in Figure 1, having three gages at each location. Each sensor node is capable of managing 3 strain gages, resulting in 6 Narada modules for this effort. Each strain gauge is configured into a quarter Wheatstone bridge with 100 times amplification. Strain gages are placed 3 in. (76 mm), 27 in. (686 mm), and 51 in. (1295 mm) from the bottom flange of the 54 in. (137.2 cm) web, as seen in Figure 5(a). Strain gage locations were prepared by grinding of the painted steel, followed by sanding and polishing, and finally cleaning with acetone before adhesion of the metal foil gage.
Data Processing Clients
Figure 1 shows the location of the 18 accelerometers located on girder #1 and girder #7. The FDD method, developed by Brincker et al., is used for modal analysis of acceleration response data as it is known to identify close modes with higher accuracy than the classical peak picking approach, and is suited well for automated processing of output only data. Using the FDD procedure, three modes are easily identified. Shown in Figure 6 are the associated modes, displayed as mode shape with corresponding frequency. Specifically, 1stmode bending is observed at 2.44 Hz, 1st mode torsional at 2.78 Hz, and 2nd mode bending at 10.16 Hz. The modal parameters identified can now contribute to the calibration of the initial finite element model as well as the updating of the model in the future.
Strain gages instrumented at six locations around TRB provide information about the neutral axis of the composite
section, a valuable piece of information towards the assessment of composite action. As seen in Figure 5(a), three
metal foil strain gages were installed along the height of the girder web. To increase the resolution of the strain
profile, the assumption of linear strain is made. A best fit line from the bottom flange to the top flange, in the least
squares sense, is determined from the three strain gauge measurements. The computed linear strain profile is
discretized for 1 in. spacing along the girder height. The change in strain over incremental periods of time at each point along the height of the girder is computed and stored over the entire time history available for analysis. The
position along the height of the girder that shows the least amount of variance in strain over the entire time history is
chosen as the neutral axis location. The identified
neutral axis is considered an average as the neutral axis is observed to move during vehicle loading.
Strain gages are also used to determine the rate of fatigue damage. The fatigue algorithms are embedded for decentralized monitoring, having the data processed and stored long term on-board the Narada wireless sensing unit. A streaming version of the rainflow counting  method is embedded on the Narada and used in conjunction with Palmgren-Miner linear damage accumulation [#10] rules to sum damages. Stress cycles obtained through rainflow counting procedure are stored long term onboard the Narada in a bivariate histogram architecture (amplitude and mean) . The system manager can query the sensing nodes at any time to retrieve the most recent cycle history.
The steel girder-concrete deck bridge already has shown signs of wear as it was built in 1973 and experiences harsh weather conditions during the Michigan winter rendering it an ideal demonstration project for bridge SHM. Accelerometers and strain gages are currently deployed for the periodic collection of bridge response data under normal traffic loadings. Data collected by the wireless sensor nodes are communicated to the bridge b ase station which forwards all bridge data via a cellular modem to a SenStore server housed on the campus of the University of Michigan. Acceleration data stored on the server is currently being used for FE model calibration of Telegraph Road Bridge. Opportunities for structural analysis and other post processing tools continue to be explored as the database of bridge response data grows. Furthermore, the authors are working on multiple data interrogation schemes centered on the estimation of the reliab ility index of bridge components as a means of enhancing the bridge owner’s decision making. Towards this end, a decision making toolbox is under development to visualize both raw sensor data and health assessment results.
 ASCE, "2009 Report Card for America's Infrastructure," American Society of Civil Engineers, Washington, D.C.2009.
 R. A. Swartz, D. Jung, J. P. Lynch, Y. Wang, D. Shi, and M. P. Flynn, "Design of a wireless sensor for scalable distributed in-network computation in a structural health monitoring system," in 5th International Workshop on Structural Health Monitoring, Stanford, CA, 2005.
 Y. Zhang, M. Kurata, J. P. Lynch, G. van der Linden, H. Sederat, and A. Prakash, "Distributed cyberinfrastructure tools for automated data processing of structural monitoring data," San Diego, California, USA, 2012, pp. 83471Y-8.
 S. O’Connor, J. Kim, J. P. Lynch, K. H. Law, and L. Salvino, "Fatigue Life Monitoring of Metallic Structures by Decentralized Rainflow Counting Embedded in a Wireless Sensor Network," 2010.
 A. A. Mosavi, H. Sedarat, S. M. O'Connor, A. Emami-Naeini, V. Jacob, A. Krimotat, and J. Lynch, "Finite Element model updating of a skewed highway bridge using a multi-variable sensitivity-based optimization approach," San Diego, California, USA, 2012, pp. 834727-12.
 J. M. W. Brownjohn and P. Q. Xia, "Dynamic assessment of curved cable-stayed bridge by model updating," Journal of structural engineering, vol. 126, pp. 252-260, 2000.
 J. M. W. Brownjohn, P. Moyo, P. Omenzetter, and Y. Lu, "Assessment of highway bridge upgrading by dynamic testing and finite-element model updating," Journal of Bridge Engineering, vol. 8, p. 162, 2003.
 R. Brincker, L. Zhang, and P. Andersen, "Modal identification from ambient responses using frequency domain decomposition," 2000, pp. 625-630.
 S. D. Downing and D. F. Socie, "Simple rainflow counting algorithms," International Journal of Fatigue, vol. 4, pp. 31-40, 1982.
 M. A. Miner, "Cumulative damage in fatigue," Journal of Applied Mechanics, vol. 12, pp. A159-A164, 1945.