Low-Cost Modal Identification Sensors for Bridge Field Testing

Background This project seeks to provide a framework for experimental load rating of bridges via inclusion of low-cost vibration sensors and dynamic tests. Currently 25% of bridges in Nebraska are posted for live load. According to National Bridge Inventory in 2012, of all posted bridges in the US, 93% were posted using analytical load ratings, 7% were posted using field evaluation and engineering judgement, and only 1% were posted using experimental load rating methods. Instrumentation costs and traffic interruptions can be problematic when load testing is necessary to accurately assess in-situ bridge live load capacity. Recent advances in (i) sensing technology and (ii) numerical methods used to process load test data permit more cost-effective data­ enabled decision making. According to the AASHTO Manual for Bridge Evaluation (MBE), dynamic tests can be used for calibration of bridge numerical models that would enhance the value of a diagnostic test. This study aims to develop a procedure for selection and use of inexpensive, off the shelf vibration sensors for dynamic testing of typical bridges in Nebraska. The sensors could include smartphones; as previous research has demonstrated their efficacy for modal identification. Matarazzo et al. used mobile phone accelerometers for structural health monitoring. Matarazzo and Pakzad then developed a structural identification framework using an expectation maximization algorithm for compatibility with a truncated physical model to enable scalable, output-only modal identification using smartphones as acceleration/vibration sensors. The dynamic sensor network data class is an adaptable and efficient technique for storing measurements from a large number of sensing nodes, which is the case for mobile sensor networks. In a similar vein, Ndong et al. used acceleration measurements from smartphones to load rate a concrete bridge in Virginia. Their method involved modal identification of the bridge using ambient vibrations and finite element (FE) model updating using vibration characteristics. The updated FE model was used for load rating. A total of nine "highly precise" accelerometers were installed on the bridge to collect acceleration data for 15 min at a sampling rate of 500 Hz. The modal properties of the bridge were determined using an enhanced frequency domain decomposition technique. The initial finite element model of the bridge was updated such that modal properties of the bridge matched field measurements. The load effects and capacity of the bridge were determined and used to calculate its rating factor. Experimental and analytical load ratings were compared, and it was concluded that the proposed method is viable and can reveal reserve capacity. Identification of modal properties of bridges has been extensively studied. It is generally understood that, in a vibration test, identification of vibration frequencies has less uncertainty when compared with vibration modes. Therefore, when merely the first few vibration frequencies of a bridge are needed, a low-cost acceleration sensor might be sufficient for precisely identifying them. Results from ongoing project by the Research Team focused on structural health monitoring, system identification, model development and updating of a railway bridge in central Nebraska has revealed that first 3 vibration frequencies are sufficient for accurate model calibration. Acceleration Measurement Using Smartphone, Bluetooth, and MEMS Sensors In this study, a comprehensive literature survey will be conducted to identify suitable low-cost acceleration sensors for bridge load testing. It is anticipated that, at a minimum, sensor categories that will be included will encompass smartphones and Bluetooth enabled acceleration sensors. All smartphones and smartwatches in the market feature an accelerometer and acceleration data can be measured, stored, and transmitted. For the purpose of this study, as a first example of a low-cost sensing device, triaxial accelerometers from an iOS-based smartphone will be used for modal identification. A commercial iOS app, VibSensor, is available in the App Store and enables measurement, collection, storage and transmission of vibration data. It is noteworthy that VibSensor also supports Android. VibSensor features: Live Display for visualization of the measured data in real-time; - Acquisition via timed or triggered recording of the measured signals with a sampling frequency of up to 100 Hz; Storage on the device and in the cloud in the format of MATLAB data matrices and .csv files (comma separated values); - Analysis using power spectral densities of the raw signals that can be displayed; and - Transmission of measured data via email. As a part of the proposed research project, low-cost Android smartwatches and smartphones will also be identified. One of striking features of smartwatches is that multiple smartwatches, both Android- and iOS-based, can simultaneously connect to a smartphone. The latter could potentially make it easier to use a network of sensors. Bluetooth-enabled and/or USS-enabled acceleration sensors will also be considered. As an example, a Serial 6-axis accelerometer/gyroscope lnvenSense MPU6050, which is packaged by WIT MOTION, will be considered as a potential modal identification sensor. This sensor unit connects easily to a laptop using USB ports. Data acquisition, recording, and storage is managed using software provided by the manufacturer. The MPU6050 takes advantage of data fusion techniques including Kalman filtering for coupling inclinometer, gyroscope and accelerometer data for noise removal. Sensor costs and specifications are: - Sensor and board costs less than $50; - Sampling frequency of up to 100 Hz for acceleration time-histories ; - Measurement range of +Sg; and - Absolute accuracy of +0.001 m/s 2 . Concerning Bluetooth sensors, MONNIT wireless accelerometer/tilt sensor will be considered. This wireless sensor can be used in a variety of applications where knowing impact, vibration, inclination, etc. is required. It features: - Each sensor costs less than $70; - Connects to a laptop, tablet, or smartphone using Bluetooth, - Can measure inclination and velocity at up to 500 Hz; - Measurement accuracy of + 2.5 %; - It has an operating range of up to 250 ft; and is powered using various battery types including a replaceable 3.0V Coin Cell Battery. Uncertainties associated with using inexpensive sensors - including those mentioned above - to determine dynamic properties under normal traffic need to be quantified. This requires establishing a minimum number of modal parameters needed for optimal, automated identification of bridge behavior. Ndong et al. compared modal identification results for a single-span reinforced concrete bridge from a smartphone to those from conventional accelerometers. Acceleration time histories and associated power spectral densities were compared using the Enhanced Frequency Doman Decomposition method for modal identification and it was shown that smartphone accelerometers could accurately identify the first 3 vibration frequencies. Preliminary Work: Modal Identification and Model Calibration of a Truss Bridge The Pl and co-Pl have developed and validated a framework for structural health monitoring, model updating, and parameter identification for a double track, riveted steel, through girder railway bridge in central Nebraska. The railway bridge studied by the authors was instrumented using highly precise accelerometers to identify its dynamic characteristics and highly precise strain transducers were deployed for measuring strain time histories. This framework can be utilized for the current study to perform modal identification and model calibration for experimental load rating of Nebraska road bridges. The studied bridge is a simply-supported, through-plate girder that spans 22.0 m. This plate girder segment is comprised of 7 panels with floor beams spaced longitudinally at 3.14 m. The span contains riveted, built-up plate girders and floor beams. The stringers and bottom laterals are rolled, steel sections. Two types of built-up, I-sections are used for the floor beams, each having differing numbers and sizes of web plates, angles, and cover plates. Stringers are rolled S24x80 and bottom laterals are single angles of varying dimensions. Accelerations were recorded under ambient vibrations using 12 uniaxial accelerometers. The accelerometers were installed at 6 locations, with each location instrumented using 2 accelerometers, one aligned vertically and one horizontally. Strain sensors locations included: (i) south main girder midspan (1 location); (ii) stringer at midspan and floorbeam connections (7 locations); and (iv) a single floor beam at midspan and close to stringers #1 and #4 (3 locations). At this point it should be highlighted that strain measurements were used to validate numerical models that calibrated using either (i) strain or (ii) acceleration output. In this study, 12 accelerometers were used to ensure accurate estimates of mode shapes will be possible. For identifying vibration frequency, only a few accelerometers would suffice. Vibration frequencies associated with stable physical modes are identified using a stabilization diagram, which is produced using ARTeMIS Modal Operational Modal Analysis (OMA) software. To automatically calibrate the FE model of the bridge, mathematical optimization tools were used to identify important parameters and minimize discrepancies between field estimated and model predicted values of modal parameters (w and 0). The corresponding objective function that was minimized, parameter selection and updating procedures and initial model configuration are described below. To build the objective function, vibration frequencies and vibration modes were extracted from the model at locations 12 to 17. Vectors denoting vibration frequencies calculated from experimental data and model simulations are denoted by Wexpand Wsim: Where, exp refers to experiments and sim stands for simulation. Vibration values were extracted in 3 directions and arranged in a single vector for each mode. In the field of vibration-based FE model updating, the most frequently used objective function consists of summation of the residuals of vibration frequencies and modes with an applied weighing W vector used to adjust the importance of each mode. Residuals were calculated for each mode separately using the difference between normalized frequency values and mode-shapes. In this study, in addition to the conventional objective function Rw,r/J , an objective function using only frequency residuals Rw was also used to determine the minimal set of modal parameters needed for precise estimation of model parameters. Results based on minimization of this objective function were compared against conventional results to ensure excluding vibration modes from the objective function will not compromise the accuracy of the parameter estimation framework. To match predicted frequencies from the initial analytical model to those estimated from field measurements, which were based on OMA, selecting an optimal set of structural parameters for the model updating process was necessary. A trial-and-error approach was used to evaluate the influence of changing certain parameters on predicted frequencies. When the varied parameter had an influence on the predicted frequencies, it would be selected for model updating. Initial, predicted vibration frequencies from the model were 5.58 Hz (lateral mode), 12.58 Hz (vertical mode) and 16.93 Hz (torsional mode) with corresponding nondimensionalized ratios (model/field) being 1.08, 1.34 and 1.17, respectively. Parameters that were examined directly influenced member stiffness and end fixity conditions. They included: (i) modulus of elasticity (E), which, can vary about 6% to 21% from its nominal value for steel [12]; (ii) floor beam weak axis bending moment of inertia (FB Iv) that accounts for the effect of stringer connections into floor beam webs; (iii) through girder strong axis bending moment of inertia (PG Ix) that accounts for slip between built-up plate girder elements; (iv) stringer end connection out-of-plane bending stiffness ( STR Kv) that accounts for compatibility between stringers and floor beams; and (v) stringer end connection in-plane bending stiffness (STR Kx) that accounts for unknown fixity between the stringers and floor beams [13]. Selected parameters are listed in Table 2. It should be highlighted that vibration frequencies and modes of the bridge were less sensitive to variations in the STR Kx and STR Ky. Simulated experiments were used to determine the minimal set of important parameters for precise model calibration. A set of acceleration time histories were generated from an initial numerical analysis (Model #1) and an optimization procedure was performed to calibrate a separate model, Model #2, against modal properties extracted from Model #1. It was concluded that modal frequencies contained enough information for precise calibration of model parameters. The optimization process was repeated 4 times using different initial guesses to ensure robustness. Results from the simulated experiments and associated parameter estimations are reported. Both engineering judgement and the Levenberg Marquardt (LM) [14] method were adopted to assist with optimally choosing parameter values. The estimated parameter values are denoted by P and the true values of the parameters are denoted by P1. Engineering judgement was used to establish their lower (LB) and upper bounds (UB). It was observed that, for both objective functions, parameters were identified with reasonable accuracy. Identification was repeated 4 times with initialization points being slightly perturbed to ensure adequate parameter convergence. E, FB Iv, and PG Ix were consistently identified with less than 1% error for each of the 4 identification runs. Estimates of STR Kx and STR Ky had an error less than 20% largely due to lower sensitivity of vibration characteristics to variations in those parameters. Next, the framework was applied to experimental ambient bridge vibration data. Optimizations were run 10 times, each time with a different initial guess for the value of the parameter to ensure methodology robustness when field data is utilized. The mean value of each parameter, denoted by Pmean, was calculated, and results were normalized with respect to the mean. It was observed that, similar to the simulated experiments estimates of E, FB Iv, and PG Ix converged to mean values, which testifies the robustness of the method for estimation of those parameters. Again, similar to the simulated experiments, estimates of STR Kx and STR Kv featured a larger standard deviation, largely due to lower model sensitivity to variations in those parameters. Dynamic Test Validation: Response Prediction Using Calibrated Bridge Model The final portion of the study compared measured and predicted strains from the numerical model. This comparison was completed to determine how accurately a model calibrated using dynamic data predicted local and global strains. Results show that numerical models calibrated using strain measurements and those calibrated using modal properties extracted from acceleration measurements accurately predicted strains, with the model calibrated using accelerations being less conservative. These comparisons help to substantiate the proposed approach for the current study. Objective This project has one overarching objective: to provide a framework for experimental load rating of bridges via inclusion of low-cost vibration sensors and dynamic tests. More specifically, this project aims to: • examine and select cost-effective dynamic sensors for use during field tests; • develop cost effective procedures for modal identification of bridges in Nebraska that will make experimental load rating more viable for owners; and • develop protocols for performing bridge load tests that will potentially require limited traffic disruption.


    • English


    • Status: Active
    • Funding: $ $142,519.00
    • Sponsor Organizations:

      Nebraska Department of Transportation

      1500 Nebraska 2
      Lincoln, Nebraska  United States  68502
    • Project Managers:

      Halsey, Lieska

    • Performing Organizations:

      University of Nebraska, Lincoln

      1400 R Street
      Lincoln, NE  United States  68588
    • Principal Investigators:

      Linzell, Daniel

    • Start Date: 20190701
    • Expected Completion Date: 20201231
    • Actual Completion Date: 0
    • USDOT Program: Transportation, Planning, Research, and Development

    Subject/Index Terms

    Filing Info

    • Accession Number: 01705858
    • Record Type: Research project
    • Source Agency: Nebraska Department of Transportation
    • Files: RiP, STATEDOT
    • Created Date: May 24 2019 11:12AM