5.2 Modal Analysis Using The ANNSI Model
5.2.3 Modal Data of the Specimen Extracted from the FBG Sensors Measurements
Aforementioned, FBG sensors have much better immunity to electro-magnetic interference;
therefore, the noise effect when using FBG sensors is much smaller than when using RSGs. This has been first discussed in section 4.6, and will be further examined here. Following the same procedure when analyzing the RSGs measurements, the FBG sensors measurements are also analyzed to obtain the corresponding strain mode shapes information. Only the records from the FBG sensors on Channel 1 (i.e. FBG1 to FBG8) are used for modal analysis. The identified
results obtained from the FBG sensors measurements for each simulated damage case are shown in Tables 5.51 to 5.75.
Before discussing the identified results, certain important things should be noted in advance.
Although the rate for sampling the FBG sensors measurements is set to be 106Hz, the sampling rate did not stay constant during the test; it fluctuated around 106Hz. While the sampling rate for the input excitation is set to be constant 200Hz. The input excitations for each damage case are re-sampled with 106Hz by using linear interpolation method before they are used for modal analysis due to the inconsistent in sampling rates of the structural responses and input excitation.
Theoretically, no matter what measurements (such as structural displacement, velocity, acceleration, and strain) are used for modal analysis, the identified modal frequencies for the same structure should be identical to each other. Subject to the problems of fluctuant sampling rate and data re-sampling, however, the identified modal parameters extracted from the FBG sensors measurements could be different to those based on the RSGs measurements. According to the identifications, it can be concluded that:
(1) Generally, the signal noise increases the difficulty of system identification. More explicitly, since the signals from the FBG sensors are cleaner than those from the RSGs, the order needed for the ANNSI model when using the FBG sensors measurements is much lesser than when using the RSGs measurements. The number of order needed for identifying the lower modes is quite small. This feature is attractive in on-line system identification because smaller order implies quicker identification.
(2) Unlike the identification results obtained from the RSGs measurements, four modes in most cases can be successfully identified by using the FBG sensors measurements.
This feature is advantaged in the cases of higher modes are needed. For example, it has been seen that the changes in lower modes for slight damage scenarios are not distinct enough to indicate damage, while the changes in higher modes, though their accuracies are lower, are distinguishable to signify possible damage.
5.3 Damage Detection With The Monitoring Networks
The global and decentralized monitoring networks were preliminarily examined by either laboratory or numerical example, respectively, in previous sections. The results had shown their potentials for applying to the practical situations. In this section, they are further investigated by the experimental data obtained from the conducted shaking table tests on a four-story steel frame structure. Acceleration measurements as well as strain measurements from FBG sensors are used for investigations.
For health monitoring purpose, the MAN that had been trained by the measurements from an intact structure is employed to play the role of monitoring unit. The trained MAN should be capable of generating the system outputs from it within a tolerable error range if the structure does not change. On the contrary, if the structural characteristics of target structure changed significantly, the trained MAN for the intact structure will no more suitable for representing the current state of the structure; as a result, the generated outputs from the trained MAN will differ from the measured responses from the damaged structure.
The health monitoring approach of using global monitoring network is applied to the acceleration and strain measurements, respectively. Notably, the strain measurements used in this section are the ones observed from the FBG sensors (FBG1 to FBG8). Start from the acceleration measurements, each set of measurements of the 24 damage cases is fed into the MAN trained by the AAA_acc measurement. The relative changes in prediction error are shown in Figure 5.3.
Since the global monitoring network provides global view on structural condition, the prediction error is derived by calculating the average of MAEs of every DOF. It is seen from Figure 5.3 that the structural damage indeed increases the prediction error of the monitoring network. However, it is not easy to affirm structural damage from comparing any two of data, especially when the damage is not significant. Therefore, continuous monitoring on a structure is essential.
If the FBG sensors measurements are used for health monitoring by using global monitoring network, the prediction errors of the 24 damage cases are depicted in Figure 5.5. Likewise, the
results of the six cases for simulating the degradation development in a structure are shown in Figure 5.6. Compare the results of these two figures with those of Figures 5.3 and 5.4, the results show the similar trend while the structure was damaged though there were slight difference existed between them. Moreover, the increments in prediction error of strain measurements are larger than those of acceleration in serious damage cases. For examples, the relative increments in prediction error of acceleration and strain measurements for Dcase_NAA are within 100% and beyond 300%, respectively; the maximum values in Figures 5.3 and 5.5 are about 175% and 870%, respectively.
C HAPTER 6 C ONCLUDING R EMARKS
The main purpose of this work attempts to assemble a framework of a health monitoring system for smart structures based on ANN models. By investigating from analytical study to experimental study, the proposed framework for an ANN-based integrated system for structural monitoring and damage diagnosis is revealed adaptive and feasible. According to the study results shown in this research, they are summarized and discussed in the succeeding sections.
(1) The ANNSI model successfully identified the structural modal parameters of the specimen under various damage states from the measurements of the accelerometers, FBG sensors, and RSGs. The identified results show consistency between each of them.
(2) The induced damage can be reflected by the changes in structural modal parameters of the specimen. However, the modal parameters changes of the lower mode are not significant in the structure with slight damage.
(3) The FBG sensors do show their potentials in system identification and monitoring. The noise effect of the FBG sensors measurements is much smaller than that of the RSGs and accelerometers. This will make the identification easier when using the FBG sensors data. Furthermore, the distinguishing advantages of much less mass and great capacity of multiplexing a large number of sensors along a single fiber link make FBG sensors promising for health monitoring of practical structures.
(4) Compare with the CMS that based on the displacement mode shapes, the CSMS that based on the strain mode shapes is more sensitive to the structural damage. Moreover, the location of damage can be reflected by the sensing stations with larger value of CSMS. By using this approach, the damage location for the most simulated damage
cases can be identified.
(5) The damage detection strategy that based on the prediction errors from the monitoring networks is easy to implement without limitation on the number of sensors. The increasing prediction error from the global monitoring network in the simulation of degradation development signifies deterioration of the structural integrity. Moreover, the larger prediction errors from the decentralized neural networks indicate the locality of the structural damage.
(6) Although the damage detection method that based on the DLF and the UFN model failed to be applied to the experimental measurements due to the problem of without a suitable analytical model, the damage diagnosis of the structure can still be carried out by other proposed strategies. If a suitable analytical model is available, the damage diagnosis of the structure will be improved and enhanced.
(7) Since the methods and approaches involved in the system are mainly based on ANNs, the system is adaptive because ANNs are expected to improve their performance as they experience more episodes form the reality.
(8) The damage detection mechanism of the system was designed to integrate different diagnosis strategies to implement the similar tasks. In this way, even one of the diagnosis strategies fails to perform its duty, the system can still work properly.
(9) The system is independent of the methods used in each mechanism and is expandable.
Any effective or improved method can be added to the corresponding mechanism to enhance the performance of the whole system.
R EFERENCE
[1] Salawu, O. S., "Detection of structural damage through changes in frequency: a review,"
Engineering Structures, 19(9), pp. 718-723, 1997.
[2] Alampalli, S. and Fu, G. K., "Full scale dynamic monitoring of highway bridges,"
Structural Engineering in Natural Hazards Mitigation, 1(pp. 1602-1607, 1993.
[3] Salawu, O. S. and Williams, C., "Bridge assessment using forced-vibration testing,"
Journal of Structural Engineering, ASCE, 121(2), pp. 161-173, 1995.
[4] West, W. M., "Illustration of the use of modal assurance criterion to detect structural changes in an orbiter test specimen," Proc., 4th International Modal Analysis Conference, 1986.
[5] Lieven, N. A. J. and Ewins, D. J., "Spatial correlation of mode shapes, the coordinate modal assurance criterion (COMAC)," Proc., 5th International Modal Analysis Conference, 1988.
[6] Biswas, M., et al., "Diagnosis experiment spectral/modal analysis of highway bridges,"
The International Journal of Analytical and Experimental Modal Analysis, 5(1), pp. 33-42, 1990.
[7] Cawley, P. and Adams, R. D., "The location of defects in structures from measurements of natural frequencies," Journal of Strain Analysis, 14(2), pp. 49-57, 1979.
[8] Penny, J. E. T., et al., "Damage location in structures using vibration data," Proc., 11th Internatiional Modal Analysis Conference, Kissimee, FL., 1993.
[9] Contursi, T., et al., "A multiple-damage location assurance criterion based on natural frequency changes," Journal of Vibration and Control, 4(5), pp. 619-663, 1998.
[10] Hearn, G. and Testa, R. B., "Modal analysis for damage detection in structures," Journal of Structural Engineering, ASCE, 117(10), pp. 3042-3063, 1991.
[11] Shi, Z. Y., et al., "Structural damage localization from modal strain energy change,"
Journal of Sound and Vibration, 218(5), pp. 825-844, 1998.
[12] Shi, Z. Y., et al., "Structural damage detection from modal strain energy change," Journal of Engineering Mechanics, ASCE, 126(12), pp. 1216-1223, 2000.
[13] Zimmermann, D. C. and Kaouk, M., "Structural damage detection using a subspace rotation algorithm," Proc. AIAA 33rd Structures, Structural Dynamics, and Materials Conference, Dallas, TX, 1992.
[14] Lim, T. W. and Kashangaki, T. A. L., "Structural damage detection of space truss structures using best achievable eigenvectors," AIAA Journal, 32(5), pp. 1049-1057, 1994.
[15] Yao, G. C., et al., "Damage diagnosis of steel frames using vibrational signature analysis,"
Journal of Engineering Mechanics, ASCE, 118(9), pp. 1949-1961, 1992.
[16] Stubbs, N. and Kim, J. T., "Damage localization in structures without baseline modal parameters," AIAA Journal, 34(8), pp. 1644-1649, 1996.
[17] Topole, K. G. and Stubbs, N., "Non-destructive damage evaluation of a structure from limited modal parameters," Earthquake Engineering and Structural Dynamics, 24(12), pp.
1427-1436, 1995.
[18] Messina, A., et al., "Damage detection and localisation using natural frequency changes,"
Proc., Identification in Engineering Systems, Swansea, Wales, 1996.
[19] Messina, A., et al., "Structural damage detection by a sensitivity and statistical-based method," Journal of Sound and Vibration, 216(5), pp. 791-808, 1998.
[20] Shi, Z. Y., et al., "Damage localization by directly using incomplete mode shapes,"
Journal of Engineering Mechanics, ASCE, 126(6), pp. 656-660, 2000.
[21] Lin, C. S., "Location of modeling errors using modal test data," AIAA Journal, 33(9), pp.
1650-1654, 1995.
[22] Pandey, A. K. and Biswas, M., "Damage diagnosis of truss structures by estimation of flexibity change," International Journal of Analytical and Experimental Modal Anal, 10(2), pp. 104-117, 1995.
[23] Pandey, A. K., et al., "Damage detection from chamges in curvature mode shapes,"
Journal of Sound and Vibration, 145(2), pp. 321-332, 1991.
[24] Cornwell, P., et al., "Application of the strain energy damage detection method to plate-like structures," Journal of Sound and Vibration, 224(2), pp. 359-374, 1999.
[25] Ratcliffe, C. P., "Damage detection using a modified Laplacian operator on mode shape data," Journal of Sound and Vibration, 204(3), pp. 505-517, 1997.
[26] Wahab, M. M. A. and Roeck, D. D., "Damage detection in bridges using modal curvatures:
application to a real damage scenario," Journal of Sound and Vibration, 226(2), pp.
217-235, 1999.
[27] Ricles, J. M. and Kosmatka, J. B., "Damage detection in elastic structures using vibratory residual forces and weighted sensitivity," AIAA Journal, 30(9), pp. 2310-2316, 1992.
[28] Kosmatka, J. B. and Ricles, J. M., "Damage detection in structures by modal vibration characterization," Journal of Structural Engineering, ASCE, 125(12), pp. 1384-1392,
1999.
[29] Zimmermann, D. C. and Kaouk, M., "Structural damage detection using a minimum rank update theory," Journal of Vibration and Acoustics, ASME, 116(pp. 222-231, 1994.
[30] Kaouk, M. and Zimmermann, D. C., "Structural damage assessment using a generalized minimum rank perturbation theory," AIAA Journal, 32(4), pp. 836-842, 1994.
[31] Shi, Z. Y., et al., "Optimum sensor placement for structural damage detection," Journal of Engineering Mechanics, ASCE, 126(11), pp. 1173-1179, 2000.
[32] Koh, C. G., et al., "Damage detection of buildings: numerical and experimental studies,"
Journal of Structural Engineering, ASCE, 121(8), pp. 1155-1160, 1995.
[33] Shi, Z. Y., et al., "Improved damage quantification from elemental modal strain energy change," Journal of Engineering Mechanics, ASCE, 128(5), pp. 521-529, 2002.
[34] Golden, R. M., Mathematical methods for neural network analysis and design. MIT Press, Cambridge, MA., 1996.
[35] Adeli, H. and Hung, S. L., "A fuzzy neural network learning model for image recognition," Integrated Computer-Aided Engineering, 1(1), pp. 43-55, 1993.
[36] Adeli, H. and Hung, S. L., Machine Learning- Neural Networks, Genetic Algorithms, and Fuzzy Systems. John Wiley & Sons, New York, NY, 1995.
[37] Adeli, H. and Park, H. S., Neurocomputing in design automation. CRC Press, Boca Raton, FL, 1998.
[38] Haykin, S., Neural networks: a comprehensive foundation. Prentice-Hall, Englewood Cliffs, NJ., 1999.
[39] Ghaboussi, J., et al., "Knowledge-based modeling of material behavior with neural networks," Journal of Engineering Mechanics, ASCE, 117(1), pp. 132-153, 1991.
[40] Wu, X., et al., "Use of neural networks in detection of structural damage," Computers and Structures, 42(4), pp. 649-659, 1992.
[41] Elkordy, M. F., et al., "Neural networks trained by analytically simulated damage states,"
Journal of Computing In Civil Engineering., ASCE, 7(2), pp. 130-145, 1993.
[42] Szewezyk, P. and Hajela, P., "Damage detection in structures based on feature-sensitive neural network," Journal of Computing In Civil Engineering., ASCE, 8(2), pp. 163-178, 1994.
[43] Pandey, P. C. and Barai, S. V., "Multilayer perceptron in damage detection of bridge structures," Computers and Structures, 54(4), pp. 597-608, 1995.
[44] Pandey, P. C. and Barai, S. V., "Time-delay neural networks in damage detection of
railway bridges," Advance in Engineering Software, 28(1), pp. 1-10, 1997.
[45] Zhao, J., et al., "Structural damage detection using artificial neural networks," Journal of Infrastructure Systems, ASCE, 4(2), pp. 93-101, 1998.
[46] Masri, S. F., et al., "Application of neural networks for detection of changes in nonlinear systems," Journal of Engineering Mechanics, ASCE, 126(7), pp. 666-676, 2000.
[47] Zapico, J. L., et al., "Vibration-based damage assessment in steel frames using neural networks," Smart Materials and Structures, 10(3), pp. 553-559, 2001.
[48] Sahin, M. and Shenoi, R. A., "Quantification and localisation of damage in beam-like structures by using artificial neural networks wth experimental validation," Engineering Structures, 25(14), pp. 1785-1802, 2003.
[49] Kim, S. H., et al., "Structural monitoring system based on sensitivity analysis and a neural network," Computer-Aided Civil and Infrastructure Engineering, 15(4), pp. 309-318, 2000.
[50] Marwala, T., "Damage identification using committee of neural networks," Journal of Engineering Mechanics, ASCE, 126(1), pp. 43-50, 2000.
[51] Yun, C. B., et al., "Joint damage assessment of framed structures using a neural networks technique," Engineering Structures, 23(5), pp. 425-435, 2001.
[52] Ni, Y. Q., et al., "Constructing input vectors to nueral networks for structiral damage identification," Smart Material and Structures, 11(pp. 825-833, 2002.
[53] Tsai, C. H. and Hsu, D. S., "Diagnosis of reinforced concrete structural damage based on displacement time history using the back-propagation neural network technique," Journal of Computing In Civil Engineering., ASCE, 16(1), pp. 49-58, 2002.
[54] Wu, Z., et al., "Decentralized parameteric damage detection based on neural networks,"
Computer-Aided Civil and Infrastructure Engineering, 17(pp. 175-184, 2002.
[55] McCulloch, W. and Pitts, W., "A logical calculus of ideas imminent in nervous activity,"
Bulletin of Mathematical Biophysics, 5(pp. 115-133, 1943.
[56] Rosenblatt, F., "The perceptron: a probabilitistic model for information storage and organization in the brain," Psychological Review, 65(pp. 386-408, 1958.
[57] Rumelhart, D. E., et al., "Learning international representation by error propagation," in Parallel Distributed Processing, R. J. Williams, Ed. Cambridge, MA.: The MIT Press, 1986, pp. 318-362.
[58] Hecht-Nielsen, R., "Theory of the back propagation neural network," Proceedins of International Joint Conference on Neural Network, IEEE, 1989.
[59] Hung, S. L. and Lin, Y. L., "Application of an L-BFGS Neural Network Learning
Algorithm in Engineering Analysis and Design," Proc., The 2nd National Conf. on Struct.
Engrg., Taiwan, R.O.C., 1994.
[60] Adeli, H., "Neural network in civil engineering: 1989-2000," Compter-Aided Civil and Infrastructure Engineering, 16(2), pp. 126-142, 2001.
[61] Hung, S. L. and Jan, J. C., "Machine learning in engineering design: an unsupervised fuzzy neural network learning model," Proc. of Intelligent Information Systems, IEEE Computer Society, California, 1997.
[62] Hung, S. L. and Jan, J. C., "Machine learning in engineering analysis and design: an integrated fuzzy neural network learning model," Computer-Aided Civil and Infrastructure Engineering, 14(pp. 207-219, 1999.
[63] Hung, S. L. and Jan, J. C., "Augmented IFN Learning Model," Journal of Computing in Civil Engineering, ASCE, 14(1), pp. 15-22, 2000.
[64] Hung, S. L. and Lai, C. M., "Unsupervised fuzzy neural network structural active pulse controller," Earthquake Engineering and Structural Dynamics, 30(4), pp. 465-484, 2001.
[65] Wen, C. M. and Hung, S. L., "Unsupervised fuzzy neural network for the damage detection of structures," (Submitted to Earthquake Engineering and Structural Dynamics), 2004.
[66] Maia, N. M. M., et al., Theoretical and experimental modal analysis. Research Studies Press, Ltd., Baldock, Hertfordshire, England, 1997.
[67] He, J. and Fu, Z. F., Modal analysis. Butterworth-Heinemann, Woburn, MA, 2001.
[68] Yun, C. B. and Bahng, E. Y., "Substructural identification using neural networks,"
Computers and Structures, 77(1), pp. 41-52, 2000.
[69] Huang, C. S., "Structural identification from ambient vibration measurement using the multivariate AR model," Journal of Sound and Vibration, 241(3), pp. 337-359, 2001.
[70] Huang, C. S., "A study on techniques for analyzing ambient vibration measurement (II) - time series methods," National Center for Research on Earthquake Engineering, Taiwan, R.O.C NCREE Report No. NCREE-99-018, 1999.
[71] Law, S. S., et al., "Efficient numerical model for the damage detection of large scale engineering structures," Engineering Structures, 23(5), pp. 436-451, 2001.
Table 4.1 Specifications of the shaking table in NCTU
Item Value
Table size (m ) 2 3×3
Weight of table (kg) 5000 Max. specimen weight (kg) 10,000 Max. displacement (cm) ± 12.5
Max. velocity (cm/sec) ± 60 Max. acceleration (g) ± 1
Table 4.2 The characterizations of the experimental specimen
Item Value
Plane size (m2) 2× 2
Story height (m) 1.6
Weight (kg) ≈ 5000
Cross section of the column (mm) 125×60×6×8 Cross section of the beam (mm) 125×60×6×8 Cross section of the girder (mm) 100×50×5×7 Size of the mass block (mm) 1360×1360×32 Mass at 4th floor (kg−s2/m) 117.06 Mass at 3rd floor (kg−s2/m) 121.21 Mass at 2nd floor (kg−s2/m) 121.21 Mass at 1st floor (kg−s2/m) 121.54
Table 4.3 Analytical modal parameters of the test model in the transverse direction
Mode 1 2 3 4
Frequency (Hz) 1.18 3.48 5.45 6.80
Damping ratio (%) 5 5 5 5
4F 1.000 1.000 0.664 0.380
3F 0.879 0.011 -0.846 -0.903
2F 0.648 -0.998 -0.348 1.000
Mode shape
1F 0.332 -0.976 1.000 -0.659
Table 4.4 Specifications of the accelerometers
TYPE Axes Span (g)
A4 CrossBow CXL02LF1 X ± 2 A3 CrossBow CXL02LF1 X ± 2 A2 CrossBow CXL01LF1 X ± 1 A1 CrossBow CXL01LF1 X ± 1 Abase CrossBow CXL01LF1 X ± 1
Table 4.5 Specifications of the FBG-SLI
Optical
Number of Optical Channels 4
Maximum Number of FBG Sensors/Channel 64 (256 total across 4 channels)
Wavelength Range 1525 - 1565 nm
Absolute Accuracy +/- 5 pm (~4.2 µ) typ, +/- 10 pm max Repeatability +/- 2 pm (~1.7 µ) typ, +/- 5 pm max
Optical Power/Channel -10 dBm approx.
Dynamic Range (4 software-controlled gain settings) 30 dB
Resolution <1 pm (~0.8µ )
Scan Frequency 108 Hz max
Minimum FBG Spacing 0.5 nm
Optical Connector FC/APC
Hardware and Software
Computer Interface Card PCI or PC CARD (PCMCIA)
Interface Cable Included
FBG-IS Software for Windows
95, 98, 2000, NT and XP Included Electrical
Power Supply 95-135 VAC or 190-265 VAC, 15W Uncalibrated Analog Output - BNC Connectors Test, sync and scan
Mechanical
Operating Temperature 10o – 40oC
Dimensions 69 x 277 x 267 mm
Weight 4.1 kg
Options
Test Processor Laptop computer/data management system Custom Optical Connectors FC/SPC
Custom Computer Interface Cards ISA
Table 4.6 Center wavelength of the FBG sensors along Channel 1
FBG1 FBG2 FBG3 FBG4 FBG5 FBG6 FBG7 FBG8 Wavelength
(nm) 1542 1545 1548 1551 1554 1557 1560 1563
Table 4.7 Center wavelength of the FBG sensors along Channel 2 FBG9 FBG10 FBG11 FBG12 Wavelength
(nm) 1530 1533 1539 1536
Table 4.8 Dimension of the SC Cross section
(mm)
Area (cm2)
Mass (kg/m)
Ix (cm4)
SC-A 100×50×20×2.3 5.14 4.06 80.7 SC-B 75×45×15×2.3 4.14 3.25 37.1
Table 4.9 Characterizations of the simulated damage cases
No. Notation of the damage case
SC arrangement (1F-2F-3F)
Notation of the damage class
1 AAA A-A-A Intact
2 Dcase_BAA B-A-A Dclass_k1
3 Dcase_NAA N-A-A Dclass_k1
4 Dcase_ABA A-B-A Dclass_k2
5 Dcase_ANA A-N-A Dclass_k2
6 Dcase_AAB A-A-B Dclass_k3
7 Dcase_AAN A-A-N Dclass_k3
8 Dcase_BBA B-B-A Dclass_k1&k2
9 Dcase_BNA B-N-A Dclass_k1&k2
10 Dcase_NBA N-B-A Dclass_k1&k2
11 Dcase_NNA N-N-A Dclass_k1&k2 12 Dcase_BAB B-A-B Dclass_k1&k3
13 Dcase_BAN B-A-N Dclass_k1&k3
14 Dcase_NAB N-A-B Dclass_k1&k3
15 Dcase_NAN N-A-N Dclass_k1&k3 16 Dcase_ABB A-B-B Dclass_k2&k3 17 Dcase_ABN A-B-N Dclass_k2&k3
18 Dcase_ANB A-N-B Dclass_k2&k3
19 Dcase_ANN A-N-N Dclass_k2&k3 20 Dcase_BBB B-B-B Dclass_k1&k2&k3 21 Dcase_BBN B-B-N Dclass_k1&k2&k3
22 Dcase_NBB N-B-B Dclass_k1&k2&k3
23 Dcase_BNN B-N-N Dclass_k1&k2&k3
24 Dcase_NNB N-N-B Dclass_k1&k2&k3 25 Dcase_NNN N-N-N Dclass_k1&k2&k3
Table 4.10 Operation sequence of the shaking table tests
Case Save acc. data as Save RSG data as Save FBG data as
Case Save acc. data as Save RSG data as Save FBG data as