Seismic structural displacement measurement using a high-speed line-scan camera: experimental validation

Transcription

1 Seismic structural displacement measurement using a high-speed line-scan camera: experimental validation M. Nayyerloo, X.-Q. Chen & J.G. Chase Department of Mechanical Engineering, University of Canterbury, Christchurch A. Malherbe National Advanced Engineering School of Limoges (ENSIL), Limoges University, Limoges, France 2 NZSEE Conference G.A. MacRae Department of Civil and Natural Resources Engineering, University of Canterbury, Christchurch ABSTRACT: This paper examines high speed line-scan cameras as a robust and high speed displacement sensor for various seismic monitoring applications. They have the additional benefit of requiring no invasive mechanisms or added processing to provide a high resolution output measure, and do not interfere architecturally. Following the method proposed by Lim et al. for measuring foundation pile movements, multiple displacements and motions of any structure can be determined in real-time at rates over khz using only one high speed line scan camera and a special pattern. This resolution is more than sufficient for structural monitoring and control problems. Versatility of the proposed measurement technique is examined through both harmonic and random vibration experiments with a suite of different input motions applied to a computer controlled cart. Comparing the input and the measured motions confirms that visionbased structural displacement measurement utilizing a high speed line-scan camera offers a high resolution and low-cost means of measuring structural vibration-displacements. INTRODUCTION Due to a variety of practical constraints and the high cost of implementation, direct high speed measurement of structural displacement and velocity is difficult. Estimation of velocity and displacement by integration of measured acceleration is subject to drift and error, which needs to be corrected using independent displacement data (Li et al. 24). Displacement sensors can be divided into two main categories: contact and non-contact. In contact sensors, such as resistive potentiometers, linear variable differential transformers (LVDTs), or piezoelectric transducers, part of the sensor should be in contact with the specimen. Contact sensors also require extensive networking as they can only measure displacements in one direction. Moreover, they are not always non-invasive, and, in some cases, interfere architecturally. These problems can be resolved by removing the physical contact by means of mechanical or electro-magnetic waves travelling between a non-contact sensor and the specimen. Eddy-current transducers, ultrasonic transducers, laser vibrometer, global positioning system (GPS), and computer vision-based techniques are some examples of non-contact devices for motion detection. Although some of these sensors, such as GPS and vision-based systems, can provide displacement data in multiple directions, their limitations have restricted their use to specific applications. For instance, Eddy-current transducers are only suitable for measuring movements of metallic objects in close proximity to the transducer, and GPS-based measurements are low rate and have coverage problems. Due to these problems, non-contact high-speed structural displacement measurement remains a challenging task. Owing to low cost and flexibility, computer vision-based solutions are widely used for applications in Paper Number

2 many fields. Light-based motion tracking of buildings subjected to earthquake motions or equipment inside buildings (Hutchinson et al. 2; Hutchinson et al. 26) is one application. Threedimensional (3D) structural displacement measurement with multiple digital cameras is a second (Chang and Ji 27). Others include, non-target stereo-vision spatiotemporal response measurement of line-like structures (Ji and Chang 28), digital image correlation based stereovision for 3D displacement measurement (Orteu 29), and applying edge detection technique with sub-pixel accuracy for structural displacement measurement (Fu and Moosa 22). In digital image-based methods, resolution and accuracy of the measurement data directly depend on the frame rate and resolution of the imaging system. Higher resolutions and frame rates in image acquisition result in a large volume of data to be processed at each measurement time step. This large volume of data slows down the image acquisition and processing and makes it unsuitable for real-time applications. Using line-scan cameras dramatically decreases the volume of the acquired frames and makes the image processing possible without using very expensive real-time hardware. 2 LINE-SCAN DISPLACEMENT MEASUREMENT 2. Method To measure structural displacement, the method proposed in Lim and Lim (28) was used. This method was originally proposed for foundation pile penetration measurement. In Lim et al. method, only one high speed line-scan camera and a special patterned target is used to capture the vertical, horizontal, and rotational movements of any point on a given structure. The printed black and white triangles pattern is posted on the structure, and the camera is pointed at it so that the scan line intersects the pattern lines, as shown in Figure. Coordinates of the intersection points, P nh and P ns, between the scan line, y = ax+b, and the normal and slanting lines of the pattern, y nh and y ns, in {T} coordinate system, which has been rotated by 4 with respect to the original coordinate system, {M}, can be written in terms of the known pattern dimensions, W and H, and the unknown scan line parameters, a and b. Centre of the scan line, or in other words centre of field of view (FOV) of the camera, can then be written in terms of a and b using coordinates of adjacent intersection points to the centre point. Measuring the Euclidian distance between any two consecutive intersection points through image processing, and calculating ratios of two following white to black (or black to white) distance pairs in the pattern yield two independent equations that can be solved for the scan line parameters. Calculating these parameters for each frame yields the centre point coordinates at each sample time, and thus the displacement of the centre point can be readily captured. Figure. Special pattern enables vertical, horizontal, and rotational displacement measurement with only one line-scan camera (Lim and Lim 28). 2

3 Transferring the measured movements from {T} to a new coordinate system, with the scan line as one of the axes, represents the linear movements of the pattern perpendicular and parallel to the scan line. In addition, rotational movements of the pattern can be calculated as the difference between the inverse tangent of a for the two following frames captured. The major advantage of the proposed displacement measurement algorithm by Lim et al., apart from the benefit of being non-contact, is that there is no need for camera calibration since only ratios of parameters are used. Furthermore, high speed line-scan cameras with line rates as high as 23 khz and resolutions in the order of 2k pixels are now commercially available (Dalsa 2). Such cameras provide high resolution and high speed at the same time, which is of great importance in real-time applications. Moreover, the line-scan displacement sensing is low cost because having only one row of pixels at each frame compared to area scan cameras significantly reduces the volume of image data to be processed in real-time and consequently reduces the cost of the total image processing system. 2.2 Hardware design Main components of the image acquisition and processing system are introduced in Table. The main component in the displacement measurement chain is the line-scan camera. Line-scan cameras scan a single pixel width digital imaging sensor at very high speed. Therefore, the images are only pixel wide. Consider a black dot moving against a white background on a linear path in the FOV of the camera, the line-scan camera sees a moving darker pixel over a series of images. This is as illustrated in Figure 2. Calibrating these pixels to the FOV in meters offers a displacement measurement. Table. Specifications of the measurement set-up Item Description Camera Lens Frame grabber board Light source Host and target PCs Image acquisition and processing software DALSA P2-23-8K4 Maximum line rate (khz) 9.3 Pixel size (µm) 7 7 Resolution (pixel) 892 Schneider Componon-S 4./8 with focusing mount and accessories National Instruments PCIe-43 Philips MASTERLine halogen reflector lamp, 6W 2V 8D Intel Core 2 Duo CPU 2.66 GHz 3.24 GB RAM GB HDD LabVIEW real-time operating system compatible LAN card LabVIEW 8. and LabVIEW real-time operating system (LabVIEW RTOS) Figure 3 shows the data flow in the displacement measurement system. The image acquisition and processing code is written in LabVIEW on a host PC for real-time execution. The target PC runs under LabVIEW real-time operating system (RTOS) to make it dedicated to the image acquisition and processing tasks in order to achieve high sampling rates. The line-scan camera is connected to a frame grabber board installed on the target PC. This card enables the direct transferring of a high volume of image data to the target PC hard disk drive (HDD). As the image processing is performed in real-time, results can be retrieved from the target to the host PC for further off-line processes either immediately after the acquisition process or even while it is being performed. Communication through the server makes remote measurement possible. 3

4 (a) (b) (c) Figure 2. a) Line scan camera used in this study, b) a linear mass-spring-damper system, and c) what the line-scan camera sees over several frames. line-scan camera Frame grabber board Target computer server host computer Figure 3. Data flow chart of the displacement measurement system (pictures and icons from and ). 2.3 Software design The image processing is to provide the time-varing coordinates of the intersection points between the scan line and the pattern lines. This simply means that edges from the white to black and black to white regions can be detected. This detection can be done using a relative intensity threshold level to detect location of large changes in the intensity of pixels at each acquired grayscale image from the camera. Following the edge detection, edges in the first frame can be tracked to detect relative displacements of the target with respect to its initial position. As the image acquisition is very fast compared to movements of the target, the next position of each edge falls within a small distance to its previous location. Therefore, the next position of each edge can be sought in a small bounded area around its previous position. This bounded area should not be less than maximum estimated movement of the target at each measurement time step on either side of the edge (Lim and Lim 28). This edge tracking algorithm requires a sufficiently wide FOV depending on the maximum likely displacements. 2.4 Measurement resolution and speed System settings, such as the number of active camera pixels, FOV, lens magnification, geometrical configurations, and frame rate can be changed using Equations a-c: FOV N d a ) = R, b) m =, and N FOV mp c ) f = () m + where FOV is the field of view of the camera, N is the number of active pixels of the camera, R is the resolution of the measurement process, d is the pixel size of the camera, m is the lens magnification, f is the focal length of the lens, and p is the distance between the camera and the moving target. For example, to measure displacements of a structure with maximum likely displacement of 2 mm, a 4

5 mm long FOV is needed. For the best resolution with the camera used in this study (maximum 892 active pixels), this FOV results in 6 µm resolution. By knowing the image size (N d) and FOV, magnification, m, can be calculated from Equation 2, and Equation 3 then can be used to find a suitable lens with the required focal length based on a desired or allowed distance between the target and the camera. The resolution from Equation, can then be improved using sub-pixel interpolation techniques if required. For this paper, full camera resolution at frame rates up to frames per second were used and other settings were chosen to have less than.3 mm resolution in measurements. 3 EXPERIMENTAL VALIDATION 3. Case study structure and selection of input ground motions To assess the performance of the system in capturing seismic structural displacements, a singledegree-of-freedom (SDOF) case study structure with an un-damped natural period of. seconds was considered. Displacements of the structure under different earthquakes, shown in Table 2, were simulated in MATLAB with % constant damping. More details about the selected records can be found in (Christopoulos et al. 22). This suite has been widely used by earthquake engineers for structural dynamic analysis in different studies. Table 2. Selected ground motions EQ Event Year Station R-Distance (km) Soil Type Duration (s) Scaling Factor PGA (g) Peak Displacement (cm) EQ Cape Mendocino 992 Fortuna - Fortuna Blvd B EQ2 Rio Dell Overpass - FF 8. B EQ3 Desert Hot Springs 23.2 B EQ4 Landers 992 Yermo Fire Station 24.9 C EQ Capitola 4. C EQ6 Gilroy Array #3 4.4 C EQ7 Loma Prieta 989 Gilroy Array #4 6. C EQ8 Gilroy Array # C EQ9 Hollister Diff. Array EQ Anderson Dam 2.4 B EQ Beverly Hills 44 Mulhol 2.8 B EQ2 Canoga Park - Topanga Can.8 C EQ3 Glendale - Las Palmas 2.4 C EQ4 Northridge 994 LA - Hollywood Stor FF 2. C EQ LA - N Faring Rd 23.9 C EQ6 N. Hollywood - Coldwater 4.6 B EQ7 Sunland - Mt Gleason Ave. 7.7 B EQ8 Brawley 8.2 C EQ9 Superstition Hills 987 El Centro Imp. Co. Cent. 3.9 C EQ2 Plaster City. 2. C The. second natural period was chosen to ensure the presence of higher frequencies in the displacement data to examine how well the line-scan measurement system captures fast motions. FFT of the derived displacement suites for simulated structures with higher and lower natural periods, as shown in Figure 4, confirms that there is no effective frequency in the displacement data greater than

6 Hz. Therefore, sampling at khz is far beyond the frequencies involved in the structure s motion and guarantees that all the motions are captured..4 7 Tn =.2 (s) Tn =. (s).2 Tn = (s) Spectral Amplitude (m) Tn =. (s) 6 Spectral Amplitude (m) 2 Spectral Amplitude (m) 4 3 Spectral Amplitude (m) Frequency (Hz) Frequency (Hz) Frequency (Hz) 2 Frequency (Hz) Figure 4. FFT analysis of displacement suites derived from records in Table 2 for case study structures with different natural periods A computer controlled cart, controlled by a dspace system, was used to generate the simulated displacement records. Encoder measurements of the actual position of the cart were used for comparison with results from the imaging system. The cart movements represent motions of the centre of mass of a real linear structure with the same natural period simulated. Figure a shows the experimental set-up. In addition, to show the capability of the proposed method to capture even higher frequencies and smaller motions, the dynamic material testing machine (MTS 8), shown in Figure b, was used to generate high frequency sinusoidal motions as small as ± mm at Hz. (a) (b) Figure. Experimental set-up used for a) random (. Line-scan camera, 2. Pattern, 3. Cart, 4. dspace,. Light, and 6. Data acquisition computer) and b) harmonic (. Line-scan camera, 2. Pattern, 3. MTS machine, 4. Computer to control the MTS machine,. Light, 6. Camera and light source power supplies, 7. Target computer, 8. Host computer, and 9. Moving head of the MTS machine) displacement measurement tests. 4 RESULTS Figure 6a shows variations of the difference between norms of the measured and actual displacement signals over the norm of the actual signal for the 2 different displacement records derived from the earthquake ground accelerations described in Table 2 for the case study structure. This ratio is less than 3% for the entire suite, and the measured and actual signals are almost identical. Figure 6b shows the results for record 2 in Table 2 as an example. 6

7 Since sampling rate and resolution of the displacement measurement set-up is sufficient to capture movements of the cart, the error is mainly due to inaccurate cart position data from the encoder. These encoder errors are caused by the backlash in a pinion coupled to the encoder. This pinion is moved on a rack by the cart movements. Therefore, output of the encoder does not change when the cart movements are smaller than the backlash. Results for the harmonic test are shown in Figure 7. Once more, error between the actual and measured displacements at peak points is less than 3%, and the actual and measured displacement signals are almost identical. Moreover, norm of the error for the entire 2-second measurement period over norm of the actual displacement signal from interior LVDT of the MTS machine is 4.64%. These values demonstrate the capability of the proposed method for high-frequency low-amplitude vibration measurement, which are typically the most difficult to measure. Difference between the norms over norm of the actual displacement (%) Displacement (mm) Cart displacement measurement results for record No. 2 in Table (fps) Actual Measured with +2 mm shift Record No. (a) time (s) Figure 6. a) Absolute value of the difference between norms of the actual and measured displacement signals over the norm of the actual displacement signal in percent for the records in Table 2, and b) measurement results for record 2 in Table 2 as an example. (b) 3 Error in peak points measurement for a +/- mm harmonic Hz images (fps). Actual and measured displacements for a harmonic Hz images (fps) Actual Measured 2. Error (%) 2. Vertical displacements (mm) Time (s) (a) time (s) (b) Figure 7. a) Error in peak points measurement and b) actual and measured displacements for the lower head of the MTS machine, travelling ± mm harmonically at Hz. FUTURE POTENTIAL APPLICATIONS Displacements and rotations of any given structure in a plane parallel to the scan line could be measured using only one line-scan camera. For instance, for the joint in Figure 8, movements in the XY-plane as well as rotations about the z-axis can be captured using a line-scan camera positioned at A. An additional line-scan camera at position B could provide movements and rotations of the joint in 7

8 the YZ-plane and make full 3D movements and rotations available. Modifying the method to account for the effect of projection at an angle provides the opportunity of using only one line-scan camera for full 3D displacement and rotation measurement. As Figure 8 shows, only one line-scan camera could be placed at an angle to both sides of a joint (position C), and the FOV could be divided into two parts to be processed based on an extended method. This novel cost-effective method, which will be pursued at later stages of this study, makes non-contact full 3D structural displacement and rotation measurement possible using only one line-scan camera. Figure 8. Extension of Lim et al. method for full 3D displacement and rotation measurement (3D model of the camera from 6 CONCLUSIONS Efficacy of the displacement measurement method initially proposed by Lim et al. for foundation pile movement measurement was empirically assessed for seismic structural vibration measurement through different random and harmonic tests. An extension to this method was proposed that makes 3D structural displacement and rotation measurement possible using only one line-scan camera. Comparison of input and measured motions for the case-study structure confirm that the proposed vision-based structural displacement measurement offers a high-speed, high-resolution, and low-cost means of measuring structural vibrations. REFERENCES: Chang, C. C., and Ji, Y. F. (27). "Flexible Videogrammetric Technique for Three-Dimensional Structural Vibration Measurement." Journal of Engineering Mechanics, 33(6), Christopoulos, C., Filiatrault, A., and Folz, B. (22). "Seismic response of self-centring hysteretic SDOF systems." Earthquake Engineering & Structural Dynamics, 3(), 3-. Dalsa. (2). "Piranha 3 P3-8x-xxx4 Line Scan Cameras-Datasheet." Dalsa Corporation. Fu, G., and Moosa, A. G. (22). "An Optical Approach to Structural Displacement Measurement and Its Application." Journal of Engineering Mechanics, 28(), -2. Hutchinson, T. C., Chaudhuri, S. R., Kuester, F., and Auduong, S. (2). "Light-Based Motion Tracking of Equipment Subjected to Earthquake Motions." Journal of Computing in Civil Engineering, 9(3), Hutchinson, T. C., Kuester, F., Doerr, K. U., and Lim, D. (26). "Optimal hardware and software design of an image-based system for capturing dynamic movements." Instrumentation and Measurement, IEEE Transactions on, (), Ji, Y. F., and Chang, C. C. (28). "Nontarget Stereo Vision Technique for Spatiotemporal Response Measurement of Line-Like Structures." Journal of Engineering Mechanics, 34(6),

9 Li, X., Peng, G. D., Rizos, C., Ge, L., Tamura, Y., and Yoshida, A. "Integration of GPS, accelerometer and optical fibre sensors for structural deformation monitoring." the 7th International Technical Meeting of the Satellite Division of the Institute of Navigation ION GNSS, Long Beach, California, Lim, M.-S., and Lim, J. (28). "Visual measurement of pile movements for the foundation work using a high-speed line-scan camera." Pattern Recognition, 4(6), Orteu, J.-J. (29). "3-D computer vision in experimental mechanics." Optics and Lasers in Engineering, 47(3-4), Poudel, U. P., Fu, G., and Ye, J. (2). "Structural damage detection using digital video imaging technique and wavelet transformation." Journal of Sound and Vibration, 286(4-),