Progressive Collapse Testing and Analysis of a Steel Frame Building

Transcription

1 Progressive Collapse Testing and Analysis of a Steel Frame Building Halil Sezen 1, Brian I. Song 2, and Kevin A. Giriunas 3 Abstract A steel frame building was tested by physically removing four first story columns from one of the perimeter frames prior to building s scheduled demolition. The purpose of the field experiment was to simulate sudden column loss in buildings that may cause progressive collapse. Another objective was to investigate the load redistribution within the building after each column removal. The measured experimental data and observed performance of the building was valuable because it is very difficult and cost-prohibitive to build and test three-dimensional fullscale building specimens in the laboratory. Generally, the design code requirements prescribe simplified analysis procedures involving instantaneous removal of certain critical columns in a building. Design methodologies and simplified analysis procedures recommended in the design guidelines were also evaluated using the experimental data. In this study, two and threedimensional models of the building were developed and analyzed to simulate the progressive collapse response. The effectiveness of the analysis procedures were evaluated by comparing with the experimental data. Introduction Progressive collapse is a chain reaction of failures initiated by instantaneous loss of one or a few vertical load carrying elements. Once the vertical structural element fails, the structure should enable an alternative load-carrying path and transfer the loads carried by that element to neighboring elements. Dynamic internal forces in adjoining members increase as a result of release of internal energy due to a member loss. After the load is redistributed throughout the structure, each structural component supports different loads including the additional internal forces. If any redistributed load exceeds the capacities of surrounding undamaged members, it 1 Associate Professor, Department of Civil and Environmental Engineering and Geodetic Science, The Ohio State University, Columbus, Ohio, sezen.1@osu.edu (corresponding author) 2 Structural Engineer, P.E., M.S.C.E., URS Corporation, Warrenville, Illinois, Brian.Song@urs.com 3 Graduate Student, Department of Civil and Environmental Engineering and Geodetic Science, The Ohio State University, Columbus, Ohio, giriunas.1@osu.edu 1

2 can cause another local failure. Such sequential failures can spread from element to element, eventually leading to the entire or a disproportionately large part of the structure. In general, such progressive collapse happens in a matter of seconds. The definition of progressive collapse may incorporate the concept of disproportionate collapse, meaning that the extent of final failure is not proportional to the initial triggering events. For example, the American Society of Civil Engineer (ASCE) Standard 7 defines progressive collapse as "the spread of an initial local failure from element to element resulting eventually in the collapse of an entire structure or a disproportionately large part of it" (ASCE 7, 2005). A similar definition of progressive collapse is provided in the General Services Administration (GSA) guidelines (2003): a situation where local failure of a primary structural component leads to the collapse of adjoining members, and hence, the total damage is disproportionate to the original cause. The Ronan Point apartment tower collapse on May 16, 1968 is the first well-known case of disproportionate progressive collapse (Griffiths et al., 1968). The building was a 22-story precast concrete bearing wall system, located in Newham, England. The collapse was initiated by a gas stove leak in a corner kitchen on the 18th floor. The Ronan Point collapse prompted interest and concern in the structural engineering community all around the world. In particular, this collapse led to significant changes in building codes in England and Canada to prevent progressive collapse. Since the collapse of the World Trade Center (WTC) towers due to terrorist attacks on September 11, 2001, interest in progressive collapse has further increased (NIST, 2005). The collapse of twin WTC structures was caused by a very large impact force and subsequent fire. It was a progressive collapse and not a disproportionate collapse (Dusenberry et al., 2004). This collapse showed that well-designed and robust modern buildings can also be susceptible to progressive collapse. Failure of one or more columns in a building and the resulting progressive collapse may be a result of a variety of events with different loading rates, pressures or magnitudes. The magnitude and probability of natural and man-made hazards are usually difficult to predict. Therefore, current progressive collapse design guidelines are generally threat-independent and do not intend to prevent such local damage. Rather, their purpose is to provide a level of resistance against disproportionate collapse and to increase the overall structural integrity. Design guidelines typically require minimum level of redundancy, strength, ductility and 2

3 element continuity. In general, the design code requirements prescribe simplified analysis procedures involving instantaneous removal of certain critical columns in a building. Among a number of building codes, standards, and design guidelines for progressive collapse, General Services Administration (GSA, 2003) and Department of Defense (DOD, 2005) address progressive collapse mitigation explicitly. They provide quantifiable and enforceable procedures to resist progressive collapse. This paper investigates the effectiveness of such commonly used progressive collapse evaluation and design methodologies through numerical simulation and experimental testing of a building. A large number of analytical studies have been conducted to evaluate the effectiveness and consistency of the current progressive collapse design guidelines. However, very limited experimental research has been performed to validate the results of these computational studies and to verify the methodologies prescribed in the guidelines. This is mainly because it is difficult to construct and test full scale building specimens because such large scale testing is discouragingly expensive. In this study, several first-story columns were physically removed from an existing steel frame building scheduled for demolition. The building was instrumented and experiment was conducted prior to its demolition. Two and three-dimensional models of the building were analyzed following the requirements of the current progressive collapse evaluation and design guidelines, such as ASCE 7 (2005) and GSA (2003). Progressive Collapse Design Approaches Indirect and direct methods are the two approaches typically used for providing resistance against progressive collapse (Ellingwood and Leyendecker, 1978). The indirect design approach attempts to prevent progressive collapse through the provision of minimum levels of strength, continuity, and ductility (ASCE 7, 2005). The examples of this approach are to improve connection resistance by special detailing, to improve redundancy, and to provide more ductility to a structure. The indirect design approach can easily be integrated into building codes or standards because it can create a redundant structure that will perform well under various extreme loading conditions and improve overall structural response. However, this method is not recommended if the specific design goal is to prevent progressive collapse because of no special consideration of specific loads or removal of critical vertical load carrying members. The 3

4 goal of indirect design method or inclusion of general structural integrity requirements in design codes or guidelines is to improve the overall structural performance of the building, not specifically the progressive collapse resistance. The direct design approach explicitly considers resistance of a building to progressive collapse during the design process (ASCE 7, 2005). There are two direct design methods: specific local resistance method and alternate load path method. The specific local resistance method seeks to provide strength to be able to resist progressive collapse. The alternate load path method seeks to provide alternative load paths to absorb localized damage and resist progressive collapse. The specific local resistance method requires that a critical structural element be able to resist abnormal loading. Regardless of the magnitude of the load, the structural element should remain intact because of its robustness. For this method, a sufficient strength and ductility must be determined for the element during design against progressive collapse. Critical load carrying elements can be designed to have additional strength and toughness to resist the abnormal loading, simply by increasing the design load factors. In the alternate path method, the design allows local failure to occur, but seeks to prevent major collapse by providing alternate load paths. Failure in a structural member dramatically changes load path by transferring loads to the members adjacent to the failed member. If the adjacent members have sufficient capacity and ductility, the structural system develops alternate load paths. Using this method, a building is analyzed for the potential of progressive collapse by instantly removing one or several load-bearing elements from the building, and by evaluating the capability of the remaining structure to prevent subsequent damage. The advantage of this method is that it is independent of the initiating load, so that the solution may be valid for any type of the hazard causing member loss. The alternate load path method is primarily recommended in the current building design codes and standards in the U.S., including General Services Administration (GSA, 2003) and the Department of Defense (DOD, 2005) guidelines. Test Building and Experiment The Bankers Life and Casualty Company (BLLC) insurance building was a three story structure located in Northbrook, Illinois. The building was constructed in 1968 and was demolished in August 2008 immediately after the experiment was conducted. The sixth edition 4

5 of the AISC Steel Construction Manual (1963) was used to design the structure. The longitudinal perimeter frame located on the north side of the BLCC building was tested and used in this paper (Figure 1). The tested part of the BLCC building was a two-story steel frame structure. As shown in Figure 2, the heights of first and second floors were 20 ft-6 in. and 14 ft-8 in., respectively. The building had a 10 ft-6 in. tall reinforced concrete framed basement. Although the experiment was performed in the first story, the computer models included basement of the building, which may have had very limited or no effect on the response of the upper two floors. Steel columns with yield strength of 36,000 psi were rigidly connected to the concrete columns at the ground level. Steel girders and beams had a specified yield strength of 42,000 psi. Test frame geometry and designation of columns and beams of the BLCC building are shown in Table 1 and Figure 2. A large loading dock bay area was located at the far Northwest corner of the structure. In addition, there was a mezzanine level between the first and second floors at the Southeast and Southwest end of the structure. For this research, the docking bay and mezzanine level were not considered in computer modeling because neither would affect the experiment. Also, the plans available do not include a recent addition of two stairwells at the south end of the building. This addition does not have any effect on the experiment. Detailed description of the building, details of instrumentation and testing, and building plans can be found in Giriunas (2009). Modeling assumptions and details of structural analysis and results are provided in Song (2010) and Song et al. (2010). Four of the ten first-story columns were removed from the perimeter frame in the following order: (1) two columns near the middle of the frame, (2) column in the building corner, and (3) column next to the corner column (Figure 1). As shown in Figure 3a, a 3-ft long section of the test columns was first torched at approximately 6 to 9 ft above the base of the column. Only a very small portion of the flange was left intact when the two column cross sections were cut. The column segment between the torched sections was then pulled out using a steel chain (Figure 3b). No significant damage was observed in the building during or after the columns were removed. The demolition team first exposed the columns and beams by partially removing the exterior brick walls. Strain gauges were attached on the columns numbered 5, 8, 11, 17 and 20 in the frame model shown in Figure 4a. Strain gauges were typically installed on columns 5

6 approximately 6 ft from ground. Universal general purpose strain gauges with a resistance of 120±0.3% Ohms were used. The major objective of strain gauge instrumentation was to monitor the redistribution of loads during the removal of columns. A scanner and portable data acquisition system were connected to a laptop computer to record the strains during testing. Figure 5 shows the history of measured strain data for strain gauge 7, attached on column 11, during the torching and removal of columns. As shown in Figure 5, the strain values dropped most after each torching or poking process. It is clear from the recorded data that there was a sudden compressive (negative) strain increase of approximately 20 to 40x10-6 and 75 to 105x10-6 in column 11 near the end of torching of neighboring columns 14 and 17, respectively (see Figure 4a). This indicates that part of the axial loads from these columns was transferred to column 11. Details of the measured test data are reported by Giriunas (2009). Modeling Assumptions and Structural Models Several assumptions were made to simplify and clearly demonstrate the steps for progressive collapse analysis of the two-dimensional (2-D) and three-dimensional (3-D) building frame models. The assumptions are: 1) the building was modeled as a special moment resistant frame with connections stronger than beams. Thus, the model allowed plastic hinges to form in the beams, not in the connections or columns; 2) connections at the foundations were modeled as pinned connections; 3) secondary members (e.g., transverse joist beams and braces) were disregarded. Other than transferring the initial gravity loads, they did not directly contribute to the progressive collapse resistance; 4) effect of large deflections was not considered. This is a reasonable assumption in this study because very large deflections or collapse was not observed in the test building; and 5) live load was assumed to be zero because non-structural loads were removed from the building prior to its demolition. At the time of testing, the frames carried only dead loads due to weight of walls, slabs, beams and columns. The weight and properties of frame members were obtained from the original design notes and structural drawings of the building. The weight of roof including corrugated steel plates, membranes and roof joists was assumed to be 25 lb/ft 2. The 12 in. thick wall contained glass, brick, and concrete masonry units. To calculate the dead load of the walls, the densities of glass, reinforced concrete masonry blocks, and exterior bricks were assumed to be 160 lb/ft 3, 135 lb/ft 3, and 120 lb/ft 3, respectively. 6

7 The computer program SAP2000 (2010) was used to evaluate the progressive collapse performance of the test building. Figure 4a shows 2-D model of the longitudinal perimeter frame of the BLCC building. Four circled columns were sequentially removed in the SAP2000 analysis, in the same order as the field torching and removal process. Figure 4b shows the 3-D model of the building. The north side of the BLCC building, mainly considered in this study, had nine bays in the longitudinal direction and eight bays in the transverse direction. To simplify 3-D models, insignificant six bays in the back side were neglected. As shown in Figure 4b, the 3-D model includes only front two bays that were most impacted by column removals. Numerical Analysis and Results Static or dynamic analysis methods with varying complexities, for example, including the effect of geometric or material nonlinearities, can be used to analyze a structure. Researchers investigated the advantages and shortcomings of different analysis procedures for progressive collapse analysis (Marjanishvili 2004, Marjanishvili and Agnew 2006, Powell 2005). A complex analysis is desired to obtain more realistic results representing the actual nonlinear and dynamic collapse response of the structure. However, linear static analysis is presented in the GSA guidelines (2003) as the primary method of analysis because it is cost-effective and easy to perform. Nevertheless, it is difficult to predict accurate behavior in a structure due to the absence of dynamic effects and material nonlinearity by sudden loss of one or more members (Kaewkulchai and Williamson, 2003). The analysis is run under the assumptions that the structure only undergoes small deformations and that the materials respond in a linear elastic fashion. Linear static procedure, therefore, is limited to simple and low- to medium-rise structures (i.e., less than ten stories) with predictable behavior (GSA, 2003). In the linear static analysis performed in this research, dead loads were multiplied by 2.0 as recommended in the GSA guidelines (2003). The amplification factor of 2.0 is used to account for dynamic effects, such as damping and inertia, when static analysis procedure is used. Results of linear static analysis are evaluated here by comparing the demand-to-capacity ratios (DCR) based on the recommendations of GSA guidelines. DCR for a structural component is defined as the ratio of the maximum demand, e.g., moment, M max of the beam or column to its expected capacity, e.g, ultimate moment capacity M p, which is calculated as the product of plastic section modulus and yield strength. In M p calculations for columns, the effect of the axial load is 7

8 neglected because the column axial loads were relatively small and did not significantly affect the moment capacity of the cross section. DCR M M max = (1) p If a DCR value is greater than 1.0, theoretically the member has exceeded its ultimate capacity at that location. However, this alone does not signify failure of the structure as long as other members are capable of carrying the forces redistributed after the initial plastic hinge formation or failure. According to GSA (2003), if DCR values for steel columns and beams in the BLLC building frame exceed 2.0 and 3.0, respectively, the members are to be considered failed members, resulting in severe damage or potential partial or total collapse of the structure. The GSA (2003) acceptance criteria is applied for DCR values calculated from static analysis, when only one first floor corner column or only one first floor column at or near the middle of one of the perimeter frames is removed from the computer model. Table 2 and Figure 6 show DCR values for all members of the original 2-D frame and after each of the four columns are removed. Frame member numbers up to 26 are columns, and beams are numbered from 27 to 49 (Figure 4a). Figure 6 also shows the DCR limits for frame members specified in the GSA (2003) for critical column removal scenarios. As shown in the figure, calculated DCR values were quite large and exceeded the acceptance criteria. It should be noted, however, that GSA criteria is applicable for removal of a single column. As indicated by red inverted triangles in Figure 6, after the loss of the first column, only one member (column 18 with a DCR of 2.10) exceeded the GSA acceptance criteria. After two or more columns were removed, many columns and beams, mostly above or next to the first two removed columns, exceeded the DCR criteria. The DCR values for all members remarkably increased after the last two columns were lost. The maximum calculated DCR values for columns and beams were 6.50 and 8.03, respectively. When all four columns were removed, the DCR values of almost all structural members were much higher than the specified limits. Again, the GSA limits are specified for one column removal scenario. In addition, in the computer model, the effect of infill walls was not considered. However, as shown in Figure 1, the infill walls were mostly intact, probably carrying a significantly large portion of the load and reducing the actual demand on the steel frame members. Also, DCR values were 8

9 calculated from static analysis of the building frame with dead loads multiplied by 2.0, which follows the GSA guidelines (2003). DCR values reported in Table 2 and Figure 6 show that the beams were more impacted than columns by the loss of columns. DCR values observed in beams were larger than in columns for all column removal cases. This was probably due to large span lengths and large slab tributary areas for the beams. The centerline distance between the transverse bays was 47 ft in the first and second floors of the BLCC building. After a column is lost, the demand on the beam bridging over the removed column significantly increases because the new beam spanning over two bays has a much longer span length. Figure 7 compares the DCR values for moments calculated from 2-D and 3-D models after all four columns were removed. It was observed that DCR values calculated from the 3-D model were smaller than those from 2-D model for almost all frame members. This simply could be due to contribution of transverse beams, which enable the structure to have more stiffness, as well as additional loads to be transferred to the columns in the transverse direction, leading to decreased demands on the perimeter frame. Thus, it can be concluded that the 2-D model may lead to overestimated demands. Figures 8 and 9 show the 2-D and 3-D models of the BLCC building with corresponding DCR values, respectively, after the loss of four columns. Since no significant visible damage was observed during the field testing, apparently the actual demands were not as large as those predicted by the SAP2000 elastic static analysis. Strains calculated from static analysis of the 2-D and 3-D models are compared with the average strain measured by strain gauge 7 attached on Column 11 after the removal of each of the four columns. Strains were calculated by considering the combination of axial loading and a bending moment generated from 2-D and 3-D SAP2000 analyses. Details of strain calculations from SAP2000 models were described in Giriunas (2009) and Song (2010). Figure 10 shows comparison between calculated strains and strain data recorded in the field. 2-D SAP2000 model significantly overestimated the measured response of the structure. The percent error was calculated for each column removal case, indicating that the results obtained from 3-D model were in close agreement with experimental results than 2-D model. Average 30% difference between the 3-D model and the field data was observed while the average difference between the 2-D model and the field data was 84%. 9

10 The loss of columns significantly affects the adjacent members, causing deformation of the structure, especially in the area where columns are lost. Vertical displacements of the joints right above the removed columns of the BLCC building were calculated. Figures 11 shows changes in the joint displacements calculated from the 3-D analysis, during the entire column removal process. Joints 1, 2, 3, and 4 designate the joints above the first, second, third, and fourth removed columns, respectively (columns 14, 17, 5 and 2 in Figure 4a). Joints 1 and 2 above the first two removed columns had the largest displacements. This result was consistent with the DCR values and 3-D deformed shape of the BLCC building shown in Figure 9. Conclusions Progressive collapse performance of an existing steel frame building was evaluated by physically removing four first-story columns from the BLLC building and by analyzing the 2-D and 3-D models of the building. The following conclusions were reached during this study based on the evaluation of experimental data and structural analysis of the test building. Most structural members of BLCC building exceeded the DCR limits once the second column was removed. However, the building did not experience a collapse during the field test, even after four columns were removed. Failure to consider the infill walls effect and the amplification factor of 2 for the dead load in linear static analysis may have led to conservative analysis results. Lower DCR values were observed in both columns and beams from the 3-D model, compared with 2-D model. Force demand on each member of 3-D model is generally smaller mainly because the additional loads caused by the loss of load-bearing columns were distributed to more structural members in the 3-D model. Strain data obtained from each field test was compared with the strain values calculated from 2-D and 3-D models. The strain values from the 3-D model were closer to experimental results. It was indicated that 3-D computer models were more accurate to simulate response of buildings to removal of columns because 3-D models can account for 3-D effects including the contribution of transverse members resulting in more conservative solutions. Acknowledgements This research was partially funded by the American Institute of Steel Construction. This support and Research Director, Tom Schlafly s feedback are gratefully acknowledged. The 10

11 authors would like to thank the Environmental Cleansing Corporation of Markham, Illinois for their help with the experiment. References AISC (1969), Manual of Steel Construction. 6th Edition, American Institute of Steel Construction, Chicago, IL. ASCE (2005), Minimum Design Loads for Buildings and Other Structures. American Society of Civil Engineers (ASCE), Reston, VA. DOD (2005), Design of Buildings to Resist Progressive Collapse. Unified Facilities Criteria (UFC) , Department of Defense (DOD), Washington, D.C. Dusenberry, D., Cagley, J. and Aquino, W. (2004), Case Studies. Multihazard Mitigation Council National Workshop on Best Practices Guidelines for the Mitigation of Progressive Collapse of buildings. National Institute of Building Sciences (NIBS). Washington, D.C. Ellingwood, B. and Leyendecker, E. (1978), Approaches for Design against Progressive Collapse, Journal of Structural Division, ASCE. Vol. 104, No. 3, pp Giriunas, K. (2009), Progressive collapse analysis of an existing building. Undergraduate Honors Thesis. Department of Civil and Environmental Engineering and Geodetic Science. The Ohio State University, Columbus, OH. Griffiths, H., Pugsley, A. and Saunders, O. (1968), Report of inquiry into the collapse of flats at Ronan Point, Canning Town. Ministry of Housing and Local Government. London, United Kingdom. GSA (2003), Progressive Collapse Analysis and Design Guidelines for New Federal Office Buildings and Major Modernization Projects. General Services Administration (GSA), Washington, D.C. Kaewkulchai, G. and Williamson, E.B. (2003), Dynamic Behavior of Planar Frames during Progressive Collapse, 16th ASCE Engineering Mechanics Conference, July 16-18, University of Washington, Seattle. Marjanishvili, S.M. (2004), Progressive Analysis Procedure for Progressive Collapse, Journal of Performance of Constructed Facilities, ASCE, Vol. 18, No. 2, pp Marjanishvili, S. and Agnew, E. (2006), Comparison of Various Procedures for Progressive Collapse Analysis, Journal of Performance of Constructed Facilities, ASCE. Vol. 20, No. 4, pp NIST (2005), The Collapse of the World Trade Center Towers, Final Report. National Institute of Standards and Technology (NIST), Gaithersburg, MD. Powell, G. (2005), Progressive Collapse: Case Study Using Nonlinear Analysis. ASCE Structures Congress and Forensic Engineering Symposium, April 20-24, New York, NY. SAP 2000 (2010), SAP 2000 Advanced Structural Analysis Program, Version 12. Computers and Structures, Inc. (CSI). Berkeley, CA 11

12 Song, B.I. (2010), Experimental and Analytical Assessment on the Progressive Potential of Existing Buildings. Master s Thesis. The Ohio State University. pp. 125 Song B., Sezen H., and Giriunas K. May 12-15, Experimental and Analytical Assessment on Progressive Collapse Potential of Actual Steel Frame Buildings. ASCE Structures Conference and North American Steel Construction Conference, American Society of Civil Engineers, Orlanda, Florida Table 1. Column and beam sections of the BLLC building Column section Beam section Column number Column type Beam number Beam type c1 concrete b1 RC flat slab c2 10 WF 49 b2 24 I 79.9 c3 10 WF 72 b3 21 WF 62 c4 8 WF 31 b4 18 WF 45 Note: RC and WF are reinforced concrete and wide-flange shaped steel I-section, respectively. The first and last numbers are the depth (inch) and nominal weight (lb per linear ft) of the steel column or beam, respectively. 12

13 Table 2. DCR values calculated from 2-D model for steel frame members. Frame member No. Before removal 1 column removed 2 columns removed 3 columns removed 4 columns removed Removed Removed Removed Removed Removed Removed Removed Removed Removed Removed

14 Figure 1. Four circled columns were removed the BLCC building during the experiment. 14

15 Figure 2. Longitudinal end frame elevation of BLLC building, including beam and columns sections (see Table 1). (a) (b) (c) Figure 3. (a) Torching of column 16C, (b) column ready to be pulled out, and (c) removal of column 16C. 15

16 (a) (b) Figure 4. (a) two-dimensional SAP2000 model with frame member numbers, and (b) threedimensional SAP2000 model (circled columns are removed in the order shown). 16

17 Column 14 torching Columns poking Column 17 torching Column 5 torching Column 5 torching Column 2 torching Columns removal Figure 5. Strain data from strain gauge 7 during testing of the BLCC building. 17

18 Figure 6. Change in DCR values of each frame member in the 2-D model for all column removal cases. Figure 7. Comparison of DCR values determined from 2-D and 3-D models after four columns removal. 18

19 Figure 8.: Moment diagram of the 2-D model with corresponding DCR values after loss of four columns. Figure 9. Deflected 3-D model with corresponding DCR values after loss of four columns. 19

20 Figure 10. Comparison of measured and calculated strain values. Figure 11. Maximum joint displacements for the 3-D model after each column removal. 20