Seven minutes to collapse
Witness accounts, visible damage and a computer model offer insights into how a pool deck cave-in spread, resulting in the catastrophic failure at Champlain Towers.
The grainy security footage released hours after Champlain Towers South crashed to the ground left many assuming the tower went down without warning in mere seconds. But a recent Miami Herald investigation based on 10 key eyewitness accounts found the collapse began somewhere on the pool deck seven minutes before the northern wing of the residential tower fell.
The Herald partnered with University of Washington engineering professor Dawn Lehman to build a computer model and explore the following critical questions raised by their experiences: Where exactly could this collapse have started and how did it spread across the pool deck and into the tower to become one of the deadliest collapses in modern history?
The witnesses described the collapse sequence as a three-part failure, each with distinct sounds that engineers can use as clues when they try to piece together what happened. First, there was a series of intermittent but distinct — and increasingly loud — booms from just before 1 a.m. The loudest and final booms in the series came at 1:14 a.m. People heard the sounds on the first floor and in the basement but saw nothing, likely indicative of an initial rebar failure.
At 1:15 a.m. the western half of the pool deck and part of the valet parking area collapsed in one loud cascade of concrete. Witnesses described the collapsed region as initiating from the southern perimeter wall and extending to the northern edge of the pool deck, where a video shows the debris from the deck collapse at Column Line 9.1 near the southern edge of the 12-story tower. (That portion of the building was covered with debris after the tower fell and therefore the exact boundary is unknown.)
For seven minutes, the building creaked and groaned, but no one heard another major concrete failure until 1:22 a.m. when the northern wing of Champlain Towers South — the long leg of the “L-shaped” tower — suddenly collapsed and killed 98 people.
This timeline provided the basis of the Herald’s forensic analysis. And the team at UW began to look for possible collapse sequences that began with rebar failing in a sequence, followed by a rapid collapse of the deck.
Computer modeling involves a process of trial and error, testing and evaluating the feasibility of different scenarios.
The process broke down into three steps:
Establish a baseline for comparison by modeling the stresses on the level-one slab before the collapse and use the model to identify overstressed regions that might be susceptible to collapse.
- Identify possible points of origin through an analysis of witness testimony, structural drawings, inspection reports and photos of post-collapse damage, including broken, corroded rebar.
- Use computer modeling to test whether that origin could lead to the kind of progressive failure observed in post-collapse photos, specifically to check that the resulting damage would spread both all the way south to the pool deck perimeter AND into the tower itself.
First, the Herald/UW built a model of the structure as designed to provide a baseline for comparison. Using this model as a reference point, a single parameter was changed to investigate the impact on the damage pattern and progression. Each simulation mapped the resulting vertical deflections (downward movement of the slab), concrete cracking and steel stresses — conditions that could have been conducive to a collapse.
While it’s generally agreed that a model based on the actual conditions of the building prior to collapse would be more accurate in predicting the structure’s behavior, the Herald could not take that approach as the team did not have access to the site to conduct necessary tests. As such, the Herald reference model is based on the limited information available in the 1980 structural drawings, including the specified design strengths for the concrete and the reinforcing steel.
Federal investigators with the National Institute of Standards and Technology will be able to make a more accurate model based on actual conditions as they probe the collapse debris and other data, including concrete core and rebar samples, which would help in gauging water damage that occurred over time. For this and other reasons, the results of the Herald/UW model represent an important step in understanding a possible collapse scenario but not the definitive result.
Despite its limitations, the reference model based on the structural drawings appeared to predict the known conditions within the building prior to collapse. Areas where water was known to pool were deeply bowed in the model. Residents told stories about sticking doors and sagging bedroom floors that were consistent with the simulated conditions, suggesting the building had sagged and warped in the areas predicted by the model.
Once there was a working reference model, modifications were made to simulate various collapse scenarios. Two different models were built. The first version was a partial model of level one that focused on just the damaged region of the pool deck. The partial model was used to quickly assess the impact of many different parameters. Scenarios deemed viable were then tested on a full model of level one. (Note: the full model takes approximately 20 hours to run on a supercomputer at the UW, which is why the initial parameters were studied using the partial model.)
The first scenarios tested were based on a video taken by Adriana Sarmiento that showed debris covering the floor of the underground garage before the tower collapsed, prompting questions about whether a failure of either a slab or column in that area could have initiated the progressive failure.
The Herald/UW team simulated slab failures of various sizes in the northern part of the pool deck that would have been directly above the area shown covered in debris in the video. The team also tested a scenario raised by other publications as well as attorneys for the victims: one that assumes column M11.1 along the northern edge of the pool deck collapsed first, because it is not visible in Sarmiento’s video. (This scenario was simulated using the full model.)
None of the scenarios based on Sarmiento’s video created a damage pattern consistent with what occurred at Champlain Towers, according to the Herald/UW model.
While losing column M11.1 increased cracking and sagging in the areas around the column, it did not seriously strain the construction joint on the pool deck or connections between the deck and the southern perimeter wall — areas where post-collapse photos show all steel connections failed either by snapping or pulling out.
Scenarios assuming the collapse began with a failure of concrete — either slab or column — on the north side of the pool deck were also inconsistent with a timeline compiled by the Herald.
A comparison with time-stamped emergency calls and witness testimony showed that Sarmiento’s video of debris on the garage floor was taken after the whole deck collapsed, and, given the rest of the timeline, was likely a result of that collapse rather than a precipitating event.
Around 1:14 a.m., when witnesses were already hearing the first booms indicating the building was beginning to fail, a couple drove through the area where debris was later documented by Sarmiento. The rubble wasn’t there at the time, nor did they report anything falling on their car. Although they reported hearing loud sounds, they saw nothing at all to explain the noise.
As the timeline became clearer and initial theories seemed increasingly unlikely, the investigation shifted south, to an area of the first floor where heavy planters separated the southwestern pool deck from the valet parking area, and where two witnesses described seeing an initial cave-in. (Those planters held hundreds of cubic feet of wet earth and were not supported by beams, unlike those on the north side of the deck. Although original plans called for beams under the planters to the south, they were left out of the final versions.)
Based on the accounts of six witnesses who first heard a series of booms culminating at 1:14 a.m. the investigation began to focus on an initial failure of internal rebar, since none of those witnesses saw any evidence of concrete failure. Those sounds came in intermittent intervals for at least 15 minutes prior to the collapse, suggesting to engineers there were multiple pieces of rebar failing one after another.
Post-collapse photos show rebar placed every 12 inches or so, snapped or pulled out piece by piece along the southern perimeter wall during the collapse.
Although it would seem unusual for a progressive collapse to begin with a failure in a connection more than 100 feet away from the tower that ultimately fell, Champlain South was a continuous structure making that progression possible, if somewhat less likely than other scenarios on face value. The slab-wall connection was continuous at the southern perimeter of the building site, and this continuity provided more strength and stiffness than the individual column connections to the slab. Therefore, from a structural engineering viewpoint, loss of such a stiff connection would change the demands on the nearby slab column connections. If those connections fail in turn, the damage could progress north and into the tower.
And photos from after the collapse showed pre-existing problems large enough to justify building a model to test whether those problems would have been enough to cause a progressive collapse across the pool deck.
Photos not only indicated corroded rebar but suggested that some connecting rebar called for in structural plans was not actually present in structural members. This was factored into the analysis, but conservatively.
While inspection reports indicated that the pool deck had suffered potentially debilitating water intrusion, concrete degradation was not factored into the model. The concrete was modeled at full design strength.
A simulation of the missing connecting rebar — something more like the condition of the building prior to the collapse — showed cracking spread into the tower and caused further deflection around a beam located at the southern edge of the residential tower. Encouraged by the initial results, the team at UW tested that same scenario on the full level-one model.
The results: Corrosion, plus the removal of some connections between the pool deck and southern perimeter wall seemed to set the stage for progressive failure across the western half of the pool deck and into the tower through a vulnerability behind the elevator shaft, in the crook of the “L-shaped” building.
How we built the full level-one model
The Herald/UW team built the non-linear, finite element model using the program LS Dyna, based on peer-reviewed methods published by Lehman and co-authors Mu-Zi Zhao and Charles W. Rooeder. Published in February 2021 in Engineering Structures, a journal focused on structural engineering and structural mechanics, their work emphasized the software’s ability to accurately model complex reinforced-concrete connections — a primary reason this method was used by the team modeling Champlain South. (Finite element models built using LS-Dyna used solid elements to represent the concrete and circular beam elements to model the rebar.)
The full model included the entire first level of the structure, extending to the perimeter walls and including connections with columns and walls. The geometry of the floor and vertical supports follows the 1980 structural drawings, with the exception of the approximate locations and geometries of the Jacuzzi and pool, which were only specified in other, non-structural drawings. The Jacuzzi walls provided a boundary condition to the damaged region and therefore needed to be included in the model.
Two different modeling approaches were used. In the region of interest extending from the south perimeter wall to Column Line 8, around the area where the pool deck collapsed, which extended E-W from Column Line G to a construction joint just east of the Jacuzzi, a refined mesh of approximately two-inch solid elements was used to model the concrete with embedded beam elements to model the reinforcement in the slabs, columns, beams and walls.
To improve run time, the remainder of the level–one modeling was simplified using shell elements and smeared reinforcement. Because those regions were not critical to understanding the progression of damage into the tower (in much of this portion, the slab was not damaged), nonlinear action areas could be less accurate than those in the structure to the south.
The majority of models did not account for bond strength in the concrete surrounding the connecting rebar. The connection between was assumed not to affect the connection strength, which depends highly on the length of the rebar away from the point of maximum stress. As this is not an issue with rigid contacts, 90-degree hooks were used, which adequately simulated 180-degree hooks called for in the plans. These were the specifications:
▪ Slab to core wall connection – No. 4 bars at 10” on center – 90 degree hooks with 18” leg with additional No. 4 bars at 26” on center as specified on plan.
▪ Slab to perimeter wall connection – No. 5 bars at 12” on center, 90 degree hooks with 18” leg.
In both connections, the connector bars were placed at the middle of the slab.
The model evaluated the state of structure subjected to the vertical load at the time of the collapse, as listed below. Load factors were not applied, nor did engineers apply live loads to the slab.
The following provides a summary of the applied loads:
▪ Slab self-weight of 150 pounds per cubic foot (pcf). (Assumes 9.5” slab everywhere, as detailed in 1980 structural drawings.)
▪ Topping on the pool deck and surface parking with a depth of 3.5 inches assuming a self-weight of 140 pcf. (A report dated Oct. 13, 2020, indicated a topping of more than 3.5 inches in some regions, but 3.5 was used in the calculation to be conservative.)
▪ Planters. The loads included soil weights at planter locations, with the weights specified in ASCE 7, a national building code that specifies loads. It was based on a planter height of three feet, as gleaned from limited planter details in the 1980 building plans.
▪ Superimposed dead load of 10 pounds per square foot (psf) to account for mechanical, electrical and plumbing and architectural finishes. In addition, a dead load of 15 psf was applied to the slab in the gym to account for exercise equipment.
▪ Vehicles were added to the surface-level parking area consistent with those observed in post-collapse photos.
Southern Wall Damage Simulation Results
Simulation 1: Reference (as designed)
Cracking: The concrete material model simulates cracking through a tensile damage parameter. Tensile damage was mapped both on the top and bottom of the slab.
These crack locations are approximate and are not intended to “match” the cracks in the garage since the actual location of cracks depends on discontinuities, shrinkage cracking and other aspects of construction.
Deflection: The deflected shape and contours are indicated by the depth of displacements (sag) in inches.
Steel stress: The output of the model indicates stresses for the steel in the slab as well as the steel connecting the slab to the vertical members. The steel that was specified for this building was ASTM A 615 Grade 60 reinforcement. Grade 60 steel has a nominal yield strength of 60,000 pounds per square inch (psi).
Simulation 2: Removing rebar connections to wall at Column Line (CL) K
Based on the photographs of the southern perimeter wall, it appears that some of the bars called for in the plans to connect the slab to the wall were missing, or did not extend into the wall.
Design drawings call for bent L-shaped No. 5 (5/8 inch diameter) steel reinforcing bars spaced at 12 inches to connect the southern perimeter wall to the slab. Photographs of the wall were inspected and the team noted that in some places, specifically east of column line L, the bars could be seen in the slab and had fractured. They appeared rusty. However, in regions around and west of CL K, the bars could not be seen and the slab appeared to have pulled off the wall rather than shearing vertically as it had in areas to the east.
Those bars around CL K were removed from the model in order to simulate this condition.
NOTE: This simulates what the state of the structure would be if those bars were not included from the start (i.e., during the original construction). Therefore this model only investigates the state of the slab-wall connection without these bars hooking from the slab into the wall, not what the state of the slab would be if the bars were removed after construction. Nor does this model account for the added factor of potential degradation or damage to the concrete itself.
Cracking: Removing the bars near CL K changes the crack pattern in the bay from the southern perimeter wall to CL 14.1, where the cracking is less significant. More importantly, the cracking propagates north into the building at the gym.
Deflection: The most significant changes in the deflected shape of the building are near the CL 9.1 perimeter beam, which suggests that changes in the restraint at the southern perimeter wall impacted the deflection of the slab both in the pool deck area and the level one of the tower.
Steel stress: A comparison of steel stresses of the as-designed model with this version shows an increase in stresses at the connection of the slab and the southern perimeter wall. In addition, the steel stresses in the slab supporting the gym increased.
Simulation 3: Simulation 2 plus Bar Corrosion
Research has established that corrosion not only reduces rebar area, but also changes the material response of the steel. This so-called stress-strain relationship does not depend on the bar size. One of the most important aspects of the steel material that changes is a reduction in the “stretch” or deformability, also called strain, of the steel material. Using prior research results, the team at UW investigated the impact of a reduced strain capacity on the damage in the pool deck — that is the strain at which the steel fractures.
In this simulation, the corroded steel material model was applied to the Simulation 2 model for the bars susceptible to corrosion at important connections that exhibited fracture during the collapse (using photographic evidence), including the steel connecting the southern perimeter wall and the slab as well as at the construction joint to the east of the Jacuzzi. While the Herald has never had access to the site to examine these areas, photographs of the southern perimeter wall and the construction joint indicate that these fractured rebar appear to be corroded.
Cracking: There is significant cracking in the slab to the south of the gym (both top and bottom) as well as to the wall-slab connection (top). Cracking extends into the building at the gym. Cracking is more extensive in the pool deck around the columns.
Deflection: The most significant changes in the deflected shape and the deflection contours are at the gym and the connection to the core wall (the core wall frames the elevator and stairwell; this remained standing after the collapse but the connection to the slab was fractured). It should be noted that most of the materials do not fracture and, as such, this version is focused on the demand on and damage to the connections and the slab without full fracture modeling for all.
There are significant increases in the deflected shape in three regions: the slab adjacent to the southern perimeter wall resulting from fracture of the steel at that connection, in the area of the slab between CL 14.1 and 14, and in the gym, particularly adjacent to the core wall.
Steel stress: All of the reinforcement that was modeled as corroded fractured. Damage to these bars results in larger stresses in bars connecting the slab to the core wall, indicating these bars are susceptible to fracture. Larger stresses are also found in the bars at the column lines in the pool deck, suggesting the collapse vulnerability of those connections is increased.
This model indicates that corrosion of and damage to the southern perimeter wall-to-slab connection results in increased demand in critical regions in the building. It also points to the gym area (both north and south of the perimeter beam) in the tower as the vulnerable region.
Finally it shows it is unlikely that the collapse of the building initiated under Unit 111, as has been suggested, since the additional structure there protected that area of the slab; it is more likely that the damage to the level-one slab within the tower propagated north and east from the gym and the slab-to-core wall connection.
Although the slab-column connections do not fully lose connectivity in this version, there is significant deformation at these locations due, in part, to loss of clamping stresses in the slab, which degrade with the deterioration of the slab-wall connection. Simulating full failure of the concrete around the rebar such that the bars pull out of the slab-column connections requires an accurate estimate of the concrete strength and a constitute model for the bond capable of simulating this complex failure.
BEHIND OUR REPORTING
The project model and approach in support of the Miami Herald’s forensic investigation into the collapse of Champlain Towers South were designed and supervised by the Herald’s consulting structural engineer, Dawn Lehman.
Lehman is a professor of civil and environmental engineering at the University of Washington. The professor has extensive experience in nonlinear analysis of concrete structures and has published this experimentally validated modeling approach for reinforced concrete members and connections. She is an expert on earthquake damage.
The Herald/UW model was built by Nicolette Lewis and Ray Yu under Professor Lehman’s supervision. Lewis is a PhD student at the University of Washington, where she also works as a graduate research assistant focusing on fluid-structure interaction. Yu is a recent graduate from the University of Washington whose master’s research focused on minimum design requirements for insulated concrete form systems.
Dawn Lehman, professor of structural engineering at the University of Washington, was hired by the Miami Herald as a lead consultant for its ongoing forensic analysis of the Champlain Towers South collapse.
Nine other engineers rerviewed the Herald’s findings for this story:
▪ Dan Abrams is professor emeritus at the University of Illinois’ Grainger College of Engineering. He specializes in structural engineering.
▪ Abieyuwa Aghayere is a professor at Drexel University’s College of Engineering. He researches structural design, including the analysis of structural failures.
▪ Atorod Azizinamini is the director of the Moss School of Construction, Infrastructure and Sustainability at Florida International University. He specializes in the design and engineering of bridges.
▪ David Lange is professor emeritus at the University of Illinois’s Grainger College of Engineering. His primary research focuses on construction materials.
▪ Khalid Mosalam is the director of the Pacific Earthquake Engineering Research Center at the University of California Berkeley. He researches the assessment and rehabilitation of essential facilities such as bridges and electrical substations, and fields related to building energy efficiency and sustainability.
▪ Shankar Nair is a structural engineer with more than 50 years of professional experience and a member of the National Academy of Engineering. He is the former chairman of the Council on Tall Buildings and Urban Habitat.
▪ Three others spoke on background.
This story was originally published December 1, 2021 10:54 AM.