Coliseum failure

The case study involves the January 18, 1978 collapse of the roof of the Hartford Coliseum in Hartford, Connecticut.¹

The analysis and design of the space truss of the roof was an early application of computer simulation. The roof of this three-year-old structure collapsed at 4:15 AM on January 18, 1978 during a freezing rainstorm after a period of snow. Fortunately, there were no injuries sustained in the collapse. The night before, there were over 5,000 people in the Coliseum attending an event. Through several investigations of the collapse it was determined that this was an instance primarily of inadequate structural design. This CROSS-US report contains findings of three investigating firms, called herein Investigator 1, Investigator 2, and Investigator 3.

A triangular lattice steel space grid, supported on four reinforced concrete pylons giving spans of 270 ft (82m) and 210 ft (64m), was used to support the roof. Investigator 1 concluded that the interaction of top chord compression members and their bracing played an important role in the redistribution of load and the eventual collapse. They noted that certain compression members were braced against buckling only in one plane. As loads increased, these members buckled out of plane and redistributed the loads to other members. Over time, more chords buckled, and fewer and fewer members carried the load. This situation worsened until the remaining members were unable to withstand the added stress due to the loads present that night, and the final sudden collapse occurred.

Background²

To save money the Coliseum’s designer proposed an innovative design for the 300 ft by 360 ft (91m by 110m) space frame roof over the arena. ‘The proposed roof consisted of two main layers arranged in 30 by 30-ft [9.1m by 9.1m] grids composed of horizontal steel bars 21 ft [6.4] apart. 30-ft [9.1m] diagonal bars connected the nodes of the upper and lower layers, and in turn, were braced by a middle layer of horizontal bars. The 30 ft [9.1m] bars in the top layer were also braced at their midpoint by intermediate diagonal bars. The design departed from standard space frame roof designing procedures in five ways:

1. The configuration of the four steel angles did not provide good resistance to buckling. The cross-shaped built up section has a much smaller radius of gyration than either an I-section or a tube section.

2. The axes of the top horizontal bars and the diagonal bars did not intersect at a common working point. This induced bending in the members, making the roof especially susceptible to buckling.

3. The top layer of the roof did not support the roofing panels; the short posts on the nodes of the top layer did. Not only were these posts meant to eliminate bending stresses on the top layer bars, but their varied heights also allowed for positive drainage.

4. Four pylon legs positioned 45-ft [13.7m] inside of the edges of the roof supported it instead of boundary columns or walls.

5. The space frame was not cambered. Computer analysis predicted a downward deflection of 13-in. [33cm] at the midpoint of the roof and an upward deflection of 6-in [15cm] at the corners.

‘[Investigator 2] discovered that the roof began failing as soon as it was completed due to design deficiencies. A photograph taken during construction showed obvious bowing in two of the members in the top layer. Three major design errors coupled with the underestimation of the dead load by 20% (estimated frame weight = 18 psf [0.86kPa], actual frame weight = 23 psf[1.10kPa]) allowed the weight of the accumulated snow to collapse the roof (ENR [Engineering News Record], April 6, 1978). The load on the day of collapse was 66-73 psf [3.2-3.5kPa], while the arena should have had a design capacity of at least 140 psf [6.7kPa] (ENR, June 22, 1978). The three design errors responsible for the collapse are listed below.

• The top layer's exterior compression members on the east and the west faces were overloaded by 852%.

• The top layer's exterior compression members on the north and the south faces were overloaded by 213%.

• The top layer's interior compression members in the east-west direction were overloaded by 72%.’

‘The most overstressed members in the top layer buckled under the added weight of the snow, causing the other members to buckle. This changed the forces acting on the lower layer from tension to compression, causing them to buckle also. Two major folds formed initiating the collapse (ENR, April 6, 1978).’

‘Excessive deflections apparent during construction were brought to the engineer's attention multiple times. The engineer, confident in his design and the computer analysis which confirmed it, ignored these warnings and did not take the time to re-check their work. A conscientious engineer would pay close attention to unexpected deformations and investigate their causes. Excessive deflections often indicate structural deficiencies and should be investigated and corrected immediately.  Unexpected deformations provide a clear signal that the structural behavior is different from that anticipated by the designer.’

The joint of the truss members was incorrectly modeled in the computer as having no eccentricity. As a result of this inaccuracy, bending moments developed in the built structure, causing additional stresses in members. In post-failure investigation a nonlinear analysis was performed using correct joint modeling, and the analysis predicted collapse at the loading at which it actually occurred.

[Investigator 3] agreed with [Investigator 2] ‘that gross design errors were responsible for the progressive collapse of the roof, beginning the day that it was completed. They, however, believed that torsional buckling of the compression members, rather than the lateral buckling of top chords, instigated the collapse. Using computer analysis, [Investigator 3] found that the top truss members and the compression diagonals near the four support pylons were approaching their torsional buckling capacity the day before the collapse. An estimated 12 to 15 psf [0.57 to 0.72kPa] of live load would cause the roof to fail. The snow from the night before the collapse comprised a live load of 14 to 19 psf [0.67 to 0.91kPa]. Because torsional buckling is so uncommon, it is often an overlooked mode of failure (ENR, June 24, 1979).’

The engineers for the Hartford Arena depended on computer analysis to assess the safety of their design. However, computer programs have a tendency to provide engineers with a false sense of security. Investigators found that the roof design was extremely susceptible to buckling, which was a mode of failure not considered in that particular computer analysis and, therefore, left undiscovered (The Education Committee of the Technical Council on Forensic Engineering of the American Society of Civil Engineers’ Failures in Civil Engineering: Structural, Foundation and Geoenvironmental Case Studies). A more conventional roof design would have been much stronger. Instead of the cruciform shape of the diagonal members, a tube or I-member configuration would have been much more stable and less vulnerable to bending and twisting.

Also, if the horizontal and diagonal members intersected at the same location, it would have reduced the bending stresses in these members. In their well-known book published years after the failure and investigations, Levy and Salvadori opined that the failure of a few members would not have triggered such a catastrophic collapse if the structure had been designed and built with more redundancy (Why Buildings Fall Down:  How Structures Fail).

The Hartford Department of Licenses and Inspection did not require the design of the project to be peer-reviewed, which it usually did for projects of this magnitude. If a second opinion had been required, the design deficiencies responsible for the Coliseum’s collapse may have been discovered.