BLIND TEST ON DAMAGE DETECTION OF A STEEL FRAME STRUCTURE

BLIND TEST ON DAMAGE DETECTION OF A STEEL FRAME STRUCTURE C.J. Black< 1 >,C.E. Ventura(2) <1> Graduate Student, < 2 > Associate Professor University of British Columbia Department of Civil Engineering 2324 Main Mall Vancouver, BC Canada V6T 1 Z4 ABSTRACT At the last meeting on civil engineering structures during the 15th International Modal Analysis Conference (IMAC XV) in Orlando, Florida, the participants agreed to initiate a series of blind tests to investigate the applicability of the different techniques for damage assessment. This paper presents the information provided to the potential participants in the blind test on damage detection for a steel frame, as well as the solution. For this test, the data was generated from a finite element model of a one-third scale steel frame structure. Removing members or changing member properties in a localized region simulated the damage. Data files containing damaged and undamaged response records were available to groups interested in participating. Those who chose to participate in the blind test exercise were asked to determine the extent of damage, and if possible, the location where the damage had occurred. INTRODUCTION The blind test on damage detection of a steel frame was organized by the authors with the assistance of Profs. R. Brincker and Poul Kirkegaard of Aalborg University, Denmark. The objective of the test was to investigate how the different techniques currently being researched in the international modal analysis community for damage detection of civil engineering structures would detect the prescribed damage. The blind test aspect allows for an impartial measure of the merits of each technique with emphasis on their future development and application to damage detection. Damage detection is an important aspect of civil engineering. It has particular significance to assessment of structures damaged or suspected of being damaged during extreme natural events such as earthquakes and hurricanes. As most structural systems are hidden by architectural features of the structure there is usually no way of determining damage in a structure unless it is clearly visible. A structure may suffer considerably damage without showing signs of this damage. If damage extent and location could be estimated using modal analysis techniques the cost and time required to repair a structure could be significantly reduced. This paper is organized in three sections. The first section describes the procedure used to generate the data for the blind test, the second presents the information that was supplied to the participants in order to carry out the blind test and the third section presents the solution to the test. PREPARATION OF THE DATA The blind test data was generated using a finite element model of an actual steel frame used for shake-table testing at the University of British Columbia. This model was constructed with the finite element program SAP90 (Habibullah and Wilson, 1990). Uncorrelated shaker inputs were created using the Mathcad (MathSoft, 1995) whitenoise function. The maximum amplitude of the input motions was set to 6675 N. The input (in they or X direction) was placed on the model at the roof level. SAP90 calculated the response to the input using the first 50 modes of the system. Modal damping was set to 2% for the undamaged case and 5% for both of the damaged cases. The resulting accelerations at the nodes were obtained using SAPTIME, a post analysis program included with SAP90. To simulate actual measurement conditions, noise was added to each of the signals obtained from SAPTIME. The 623

noise was added using another whitenoise function for each signal. At each time step, the original function was multiplied by 1 plus a whitenoise function with maximum amplitude of 0.05 This essentially added 5% noise to the output signals. THE BLIND TEST -INFORMATION The following information was made available to the potential participants on a world wide web page managed by Prof. Kirkegaard and in the package of data provided to the participants. It is reproduced verbatim below for documentation purposes. Frame Description The frame is a two by two bay, four storey, rectangular modular steel structure built at approximately one third scale (see figure 1). The model stands 3.6m tall with a total width of 2.5m. All the members are made of hot rolled grade 300W steel (nominal yield stress 300 MPa). Members The sections chosen are B100x9 {a specially made light section) for the columns and S75x11 for the beams. The frame is braced with equal angle braces L25x25x3 in a chevron configuration (see figure 2). The properties of these members are given in Tables 1 and 2. Added Mass Each floor has added mass in the form of plate elements. Figure 3 shows the location and quantity of the added mass. Each floor is divided in four quadrants and the total mass of each quadrant is indicated in the figure. Modeling Assumptions There are a few modeling assumptions that have been made for the purpose of the blind test. These assumptions are: 1. The floors are rigid in the horizontal plane. 2. The braces are axial members (pinned) and therefore cannot transfer bending moments. 3. With the exception of the braces all beam-column connections are rigid. Sensor Layout The frame has been instrumented with 16 accelerometers four at each floor. From figure 4, the odd numbered sensor~ measure acceleration in the x direction while the even number sensors measure accelerations in the y direction. Location of Channels The input was from a shaker which is located at the top floor {channel17, in Figure 4). The shaker was oriented to provide random input along either the x or they axis, thus giving two sets of data for each configuration. The two input signals (inputx and inputy) are uncorrelated. Cases Examined Three cases were developed for the blind test. Each case has two parts corresponding to the direction of the input. They are as follows: Case One - No Damage Damage: None Input: Random input from shaker in either the x or y direction {no noise) Output: 16 acceleration histories (with low-level noise) Case Two - Known Damage Type Damage: Slight- 2 braces removed Input: Random input from Shaker in either the x or y direction {no noise) Output: 16 acceleration histories (with low-level noise) Case Three - Unknown Damage Type Damage: Moderate Input: Random input from Shaker in either the x or y direction {no noise) Output: 16 acceleration histories (with low-level noise) Data Provided The data from the three cases listed above is contained in six files (two files for each case). The names of the files are: case1x, case1y, case2x, case2y, case3x, case3y. The files are tab delimited text files, which have been compressed in a self-extracting exe file. The total amount of disk space required to expand the files is 18 megabytes. Each file contains 7 lines of header followed by columns of data. The format of the header is as follows: line1: title line2: case number line3: description of damage line4: description of input lines: number of points line6: time increment (uniform) line?: column titles There will be 17 columns of data for each file. The first column contains the input signal from the shaker (either inputx or inputy depending on file) and the remaining contain the signals from each of the sixteen signals. The units for the shaker data are in Newtons and the acceleration data is expressed as a fraction of gravity (g's). Results of Analysis The results of the blind test should include the following (as a minimum): a) Description of methodology used for damage detection. b) Results of modal analysis for undamaged case {Case 1). c) Estimated damage locations for Case 2. d) Estimated damage type and location for Case 3. Further Information Further information can be obtained by contacting Dr. Carlos Ventura (ventura@civil.ubc.ca) or Cameron Black (black@civil.ubc.ca) at the University of British Columbia. 624

Prooertv units B100x9 S75x11 25x25x3 ' mm 1.97x10" 1.22x10 6 lv mm 6.46><10 2.49x10 5 r. mm 41.7 29.2 fv mm 23.9 13.2 Cw mm" 1.41x10" 2.99x10" J mm 4 8.01x10 3 3.82x10 4 Area mm 2 1.13x10' 1.43x10 3 141 b mm 100 64 25 d mm 99 76 25 t mm 4.76 6.6 3 w mm 3.23 8.9 Table 1: Member Properties Modal Analysis Using the data provided in the test the participants should have been able to construct mode shapes and estimate natural frequencies and modal damping. The damping in the structure was set at 2% of critical for the undamaged case and 5% of critical for the damaged cases. The shaker amplitude was sufficient to excite at least 10 modes. The frequencies corresponding to the first 1 0 modes in each of the three cases are presented in table 3. Frequency (Hz) MODE Case 1 Case 2 Case 3 1 7.80 (Y1) 7.74 (X1) 7.68 (Y1) 2 8.42 (X1) 7.80 (Y1) 8.17 (X1) 3 11.11 (T1) 10.68 (T1) 10.84 (T1) 1. Momert of inertia in the x direction (strong) ly Momert of inertia in the y direction (weak) r. Radius of gyration in the x direction rv Radius of gyration in they direction Cw Warping torsional constant J St. Venant torsion constant Area Cross-sectional area b Width of flange d Depth of member (including flanges) t Thickness of flange w Thickness of web Table 2: Legend of Symbols THE BLIND TEST -SOLUTION This section will discuss the location of the damage for case 2 and the location and type of damage for case 3. The frequencies and mode shapes of the three cases are presented to allow for comparison with the participants' estimated values. Damage State 1 (Case Two - Known Damage Type) Two of the frames braces were removed to simulate damage. In order to mimic actual damage that my occur in an earthquake the braces at the base of one bay and the bay immediately above it were removed. These braces were diagonally across from the eccentric mass in the X direction. Figure 5 shows the location. Damage State 2 (Case Three - Unknown Damage Type) Another common damage state which results from an earthquake is that the base columns form plastic hinges at the base and below the first floor which act like a pinned connection. The three base members in the X direction on the eccentric mass side were made axial members by releasing the moment stiffness. A figure showing the location of the pinned members is given in figure 6. 4 21.41 (Y2) 21.41 (Y2) 20.96 (Y2) 5 23.99 (X2) 23.37 (X2) 22.92 (X2) 6 31.07 (T2) 30.48 (T2) 30.01 (T2) 7 33.68 (Y3) 33.68 (Y3) 33.23 (Y3) 8 39.38 (X3) 37.84 (X3) 38.04 (X3) 9 43.34 (Y4) 43.34 (Y4) 43.17 (Y4) 10 48.46 (T3) 47.18 (T3) 47.39 (T3) Table 3: Natural Frequencies for the Three Cases The first and second modes of case 2 are opposite to the undamaged case. This is a result of removing the braces in the X direction and thus decreasing the stiffness below that of they direction. In both of the damage cases stiffness was removed from the structure. One would expect that there would be an associated reduction in the frequencies. The frequencies in the Y direction of case 2 however, are the same (for this level of accuracy) as the undamaged case. This shows that even with eccentric mass in the structure there was very little coupling between the modes in case 2. This is not true for case 3 as the Y direction modes are affected by the reduction in the X direction stiffness. Figure 7 shows the first 8 modes for the three cases. The mode shapes were not significantly affected by the damage other than the first and second modes of case two being exchanged. The expected effect of the damage would be to increase the torsional response. This response can be seen in the rotation of the floors in the damaged cases relative to the undamaged case. The location of the major differences are indicated in figure 7. 625

CONCLUSION At I MAC XV it was decided that a series of blind test sessions on civil engineering structures would be developed for IMAC XVI. One of these tests was a blind test on a steel frame structure. The participants were given data generated from a finite element model of a one-third scale steel frame excited by a shaker at the roof level. The participants were asked to determine the location of the damage for a case with 2 braces removed and the location and type of damage for a case with unknown damage. This paper presented the information provided to the participants of the blind test and the solution to the test. A discussion of the mode shapes and frequencies of the actual data was presented. At the time of writing this paper the results of the participants were not available to summarize. The results of the test will be presented at the blind test session on civil engineering structures at the 16th International Modal Analysis conference in Santa Barbara, California. To obtain a summary of the results please contact the second author. ACKNOWLEDGEMENTS The Authors would like to thank Prof. R. Brincker and Prof. Paul Kirkegaard of Aalborg University, Denmark for the assistance with the organization and promotion of the blind test. The authors would like to thank the participants of the blind test as without their participation the blind test would not have been successful. REFERENCES Habibullah, A. and Wilson, E.L. (1990) SAP90 A series of Computer Programs for the Static and Dynamic Finite Element Analysis of Structures, Computers and Structures Inc., Berkeley, California. MathSoft 1995. Math cad PLUS 6.0 (1 995) Professional Edition, MathSoft Inc., Cambridge, Mass. 1.25m 1.25m Figure 1: Steel Frame Figure 2: Schematic Diagram of Frame 626

Figure 3: Location and Quantity of Added Mass (Kg) Figure 4: Location of Sensors X ~y Figure 5: Locations of Removed Braces (Case Two) Figure 6: Locations of Pinned Connections (Case Three) 627

Case 1 Case2 Case 3 Mode Y1 (7.80 Hz) Mode X1 (7.74 Hz) Mode Y1 (7.68 Hz) ~ <;==J Mode X1 (8.42 Hz) Mode Y1 (7.80 Hz) =J Mode X1 (8.17 Hz) Mode T1 (11.11 Hz) Mode T1 (10.68 Hz) Mode T1 (10.84 Hz) Mode Y2 (21.41 Hz) Mode Y2 (21.41 Hz) Mode Y2 (20.96 Hz) Mode X2 (23.99 Hz) Mode X2 (23.37 Hz) Mode X2 (22.92 Hz) Figure 7: First 8 Mode Shapes of Cases 1, 2 and 3 628

Case 1 Case2 Case 3 Mode T2 (31.07 Hz) Mode T2 (30.48 Hz) Mode T2 (30.01 Hz) Mode Y3 (33.68 Hz) Mode Y3 (33.68 Hz) ~ Mode Y3 (33.23 Hz) Mode X3 (39.38 Hz) Mode X3 (37.84 Hz) Mode X3 (38.04 Hz) Figure 7: First 8 Mode Shapes of the Cases 1, 2 and 3 (cont.) 629