SPECIAL REPORT The underlying causes of the trouble at San Francisco’s 4.5-block-long Salesforce Transit Center are coming into focus. A combination of low fracture toughness deep inside thick steel plates, cracks present as a consequence of normal steel fabrication and stress levels from loads, which are a function of design, apparently caused brittle fractures in the bottom flanges of the center's twin built-up plate girders that span 80 ft across Fremont Street. 

The problematic twin tapered girders forced the closing of the transit hub in late September, less than six weeks after it opened in mid-August.

All three ingredients were needed for the brittle fractures. “Take away any one and there is no brittle fracture because the other two compensate,” says Michael D. Engelhardt, a professor of engineering at the University of Texas, Austin, and the chair of the independent peer review panel formed in October by the area’s Metropolitan Transportation Commission. 

One ingredient is a material’s low fracture toughness, which indicates how intrinsically resistant the steel is to brittle fracture, says Engelhardt.  Another is an initiating discontinuity, known informally as cracking, often present as a consequence of normal steel fabrication. A crack, as in a pane of glass, causes a stress concentration, explains Engelhardt. It is benign if the other ingredients are not present.

The third ingredient is a high-enough level of stress, which is a function of design. Service loads, in this case live loads such as buses, generate the stress.

The troubles at Salesforce Transit Center open a window on some of the process, quality-control and responsibility challenges that occur when designing, detailing, fabricating and building long-span steel structures with large or very thick sections. Among the questions raised are whether codes and standards should be changed.

The MTC panel, in addition to its charge of technical review and guarding public safety, expects to make recommendations for change in standard practice, and possibly codes and procedures.

“There are a lot of things we can learn from this,” says Engelhardt. “People who write codes and standards should take this seriously,” he adds.

Closing of the Hub

The problematic twin tapered girders forced the closing of the transit hub in late September, less than six weeks after it opened in mid-August. The spans had passed all design reviews and inspections, met all required specifications and were designed and built by some of the most highly regarded firms in the buildings sector.

The 1.2-million-sq-ft transit center, designed by Pelli Clarke Pelli Architects with engineer-of-record Thornton Tomasetti Inc. (TT), was built by the Webcor/Obayashi Joint Venture. Skanska USA Civil West held the $189.1-million subcontract to furnish the building’s 23,000 tons of structural steel and erect the entire system, including the building’s exotic exoskeleton.

Fremont and First streets have identical twin parallel girders. First Street bottom flanges did not fracture because the notch was cut in after butt-welding was complete. Photo Courtesy TJPA

The transit center has an identical bridge-like system of parallel built-up plate girders spanning First Street.

“First Street has the same conditions but no fractures,” says Engelhardt. The reason, he adds, is a different sequence of thermal cutting and welding during fabrication.

The Herrick Corp. shop-fabricated all four girders from steel flange plates supplied by ArcelorMittal and thinner web plates supplied by Nucor Corp.

Because there was no laydown area at the site, the girders were picked in prefabricated 80-ft-long assemblies, directly from trucks, and set on end columns on either side of the street. Many elements of the structure were “dry” fit offsite to ensure there were no fit-up issues and enable installation, says the public owner, the Transbay Joint Powers Authority (TJPA).

Before the closure, the facility was functioning as a bus depot, with three levels above grade and a 5.4-acre rooftop public park. The third-floor girders support the park directly above and the second-floor bus level—via a welded midspan hanger connection—directly below the girder level. The problem girders span east to west in the longitudinal direction of the building, over the 50-ft-wide Fremont Street. Fremont crosses under the eastern section of the building.

Bottom Flange Fractures

The bottom flange fractures are near the 8-ft-deep midspan of each shop-welded girder, at 2 in. x 4 in. cuts in the flange plates, on each side of the web. The cuts were made where welds come together, near where the hanger plate slots through the flange and the two 40-ft-long flange plates, each 4 in. thick, were joined using a full-penetration weld. The 18-in.-wide hanger plate is 4 in. thick at its middle and tapers to 1.5 in. thick at its ends.  

The hanger slot is a bit more than 18 in. in the longitudinal direction of the flange and 4 in. in flange’s transverse direction, following the contour of the hanger plate but a bit bigger to let the hanger plate through. There are ¾-in. fillet shop welds except at the connection of the ends of the flange plates.

The thick hanger plate goes a couple of feet below the girder, where it is bolted to a wide-flange member in tension. That member picks up the second level bus deck.

Subsequent to the problems surfacing, the involved parties and their expert consultants have been trying to piece together the conditions and actions that led to the fractures. “Nothing has been ruled out, including design, fabrication or installation,” says Dennis Turchon, TJPA’s senior construction manager.

In addition to the material and the fabrication, the MTC panel is carefully scrutinizing the hanger assemblies. “We have to analyze that connection detail to say it’s safe,” says Engelhardt.

Material samples will be taken up high in the girder, where the hangers are in compression. There will be a detailed analysis of the state of stress.

 “We will do a detailed stress analysis of how the hanger connects to the girder as it meets the girder web,” says Engelhardt. “If we see any stress concentrations, we will smooth by grinding and polishing or enlarging the web plate where it meets the thickened plate,” he adds.

The panel also will look at issues of fatigue loading, to determine to what degree fatigue was considered in the original design. “We have barely started that,” says Engelhardt. “Analysis needs to be done first. If uncomfortable with the risk of brittle fracture, we can put in a redundant load path,” he says. 

Many Questions Remain

Many questions are in the air. Should there be stricter tests required of steel mills for material quality, including fracture toughness? Should there be extra procedures in place for the review of nonstandard details?

There are still more questions.

Should there be different procedures required of the steel detailer and steel fabricator when the engineer of record introduces cuts in flanges and/or webs on the shop drawings, after design review is complete? Should there be more specific sequences on the shop drawings for cuts and welds in heavy plates? Should fabricators be required to hire detailers that prepare shop drawings directly rather than the steel subcontractor or the fabricator hiring the detailer?  Should the fabricator be allowed to communicate directly with the other parties?

TJPA says it is fully cooperating with the MTC panel to determine the causes of the fractures. Though TJPA is not providing a reopening date, Turchon says, “It is our highest priority to reopen the transit center as soon as safely possible and that requires our full attention and effort to provide the peer review panel with all of the information they are requesting as they do their important work of determining the cause of the failure."

“It is our duty and responsibility to be fully transparent with the public as to the cause and proposed repairs,” he adds. “In that effort, we have made thousands of pages of inspections, records, drawings, etc., available on our website and we have provided information via our board meetings and the press to keep the public informed.”

However, interested parties, including some on the building team, say sifting through the thousands of pages posted by TJPA is like looking for a few needles in many haystacks.

Preliminary Findings

During a Dec. 13 public meeting of the TJPA’s board of directors, Robert S. Vecchio, CEO of metallurgist LPI, presented findings to date about the brittle fractures, which he called a preliminary root cause assessment. The findings are based on analysis and tests done on material samples cut from the flanges in the fractured regions. LPI’s finite element stress analysis is ongoing.

The investigation to date “suggests the probable cause of the girder fractures is the formation of cracks in the girder weld-access-hole radii prior to service,” said Vecchio. LPI’s client is TJPA through TT and Turner Construction Co., the facility’s construction manager, according to Vecchio.

Shallow surface microcracks developed during “thermal cutting of the weld access holes in the highly hardened and brittle martensitic surface layer,” said Vecchio. The WAHs were not ground and polished, which would have eliminated the microcracks, which are a normal consequence of thermal cutting, he added.

Larger so-called pop-in cracks subsequently formed in two of the four flanges, potentially during the butt-welding of flange plates.

– Robert S. Vecchio, CEO of metallurgist LPI

Larger so-called pop-in cracks subsequently formed in two of the four flanges, potentially during the butt-welding of flange plates, said Vecchio.

Finally, “rapid low-energy fracture of the flanges occurred as the girder was subjected to service loading, on top of the normal stresses due to fabrication,” said Vecchio. “Pre-existing defects formed at high temperatures during the thermal cutting of holes and butt welds,” he added.

Vecchio stated that microhardness testing indicated the presence of a “very brittle zone” that “enabled” the fine cracks to form. However, there was nothing deficient in the welds or in the steel plate.

 “Toughness levels at the surface of the material were good,” he said, but “the toughness in the centerline of the plate is low and cracks formed,” though the “plate did meet the requirements.”

Proximate Cause, Not Root Cause

In a public comment after Vecchio’s presentation, Rafael Sabelli, a principal with structural engineer Walter P Moore (WPM), said that calling the LPI investigation a “root cause assessment” was misleading.  “This was just the first step and the proximate cause, the mechanism of how the crack formed and propagated,” said Sabelli, representing WPM, which is a consultant to Herrick.

“A root cause assessment has to include an investigation of loading, fatigue, design and specification issues," he added.

The same “fabrication methods have been used on previous projects and will continue to be used on future projects,” said Sabelli. The hanger connection design detail has not been used on previous projects and “will probably not be repeated on future projects,” he maintains.

Another expert, who declines to be named, says, “Vecchio’s presentation seemed to assume the starting point was a given condition and there were not other ways to design the hanger connection. There are probably a dozen alternatives that could have been employed.”

The fractured flange plate, near the midspan of the tapered built-up plate girder, is 4 in. thick and 36 in. wide. Each bottom flange is a total of about 80 ft long.

Specification Met

The standard tests for fracture toughness are called Charpy V-Notch tests (CVNs), which test the material at certain temperatures. The requirement is for the steel mill, in this case ArcelorMittal, to test down to one quarter of the plate’s thickness. Tests at the mill were performed on samples 1 in. into the material. Certified mill test reports indicate the material met the specification, says a source close to the investigation.

Herrick confirms. The 4 in. plate came from the Coatesville plant of ArcelorMittal. It was certified to meet A572 13a Grade 50 type 2 and was CVN tested at +70F.

But LPI’s CVNs, at various locations and at different temperatures, are all over the place, says the source. And 2 in. down, the material was “superbrittle.” The material met the specification so perhaps the spec should be changed, suggests the source. “If you want higher fracture toughness in the middle, you need to ask for it,” he says.

Hazleton argues that the identical holes in the bottom flanges (shown above, with fracture) are not weld access holes because they don't meet the required size and shape of WAHs, were not needed for welding and are in the flanges, not the web. Image: TJPA/LPI

During his presentation, Vecchio showed magnified pictures of microscopic cracks in a feature he described as a weld access hole. “He explained that these cracks became larger during the welding process and became the initiation points for the brittle fractures that occurred sometime after the girders were put into service,” says Robert Hazleton, Herrick’s president. 

This image shows the mark for a hole in the bottom flange of a fabricated First Street girder. Image: TJPA/LPI

“It is important for the public to know microscopic cracks are normal and expected in thermal cutting operations of heavy plate,” Hazleton explains.

WAHs occur at areas of concentrated stress, therefore, the code mandates grinding after thermal cutting in these specific areas on heavy sections. Then, the code mandates nondestructive testing after grinding, he continues.

The code also mandates WAH’s be a certain size and shape to minimize stress concentrations. “These holes do not meet the size and shape of a weld access hole,” says Hazleton, which is why they were not ground or subjected to nondestructive testing.

Not Weld Access Holes

Hazleton gives two more reasons they are not WAHs: They were located in the flange instead of the web and they were not required to access a weld.

WAH shape, size and finishing requirements are all dictated by the American Institute of Steel Construction (AISC) “Specification for Structural Steel Buildings (AISC 360-10),” Chapter J, Section 1.6: “The access hole shall have a length from the toe of the weld preparation not less than 1 ½ times the thickness of the material in which the hole is made, nor less than 1 ½ in.”

After completing the joint on these girders, 2 in. x 4 in. holes remained in the bottom flanges, one of each side of the web. “If we followed the text above, the holes in the flange would be 2 in. x 12 in.,” says Hazleton.

The commentary on J1.6 offers three examples of weld access holes including that required for built-up shapes. Both code and commentary only discuss weld access holes cut in the web of structural members, not the flange.

“Code and fabrication professionals, as well as the inspector in this specific case, would not recognize these holes as weld access holes and as such would not recognize the need for grinding,” says Hazleton.

Weld experts and the AISC concur that the holes in question are not WAHs, as presented in the AISC specification.

With thick material, there is a need for clearer rules about what should be ground after thermal cutting. “There is ambiguity now,” says the source close to the investigation.

Hazleton also points out that there are no holes in the flange adjacent to the hanger shown on the design documents.

Not a Smart Thing to Do

“Flanges do all the work,” says another expert observer, who declines to be identified. “Cutting material out of flanges is not a smart thing to do,” whether for the hanger slot or for weld terminations, he adds. “And if you had to cut a hole, there are better ways to do it,” he says.

On shop drawing P537BB, dated Sept. 24, 2014, TT noted the need for a weld access hole in the web, by adding “MISSING WAH. (TT),” with an arrow pointing to the web, just above where the hanger slots through the flange. P537BB was approved as noted.

Subsequently, RFI 978, was issued on Dec. 2, 2014, by Skanska to Webcor/Obayashi. It suggested the WAH location was in error and that holes were required in the flange for this application.

The note on the RFI, presumably offered by steel detailer Candraft, working for Skanska, said: “It is not clear what is meant by this approval comment. A WAH is not required in the 4” vertical plate but it seems WAHs are required where the [complete joint penetration] weld for the 4” flange plates terminates at the 4” vertical plate. Please refer to detail 9/S1-5026 and supply the dimensions for the WAHs if required.”

Webcor’s RFI response stated that dimensioning of WAH at this location was a means and methods issue and shall be executed per American Welding Society’s AWS D1.1. TT added: “We disagree with the reason for the request: insufficient information. We categorize this RFI as clarification on means and methods and called out on contract documents.”

Red Flags

This raises a number of red flags, says Hazleton. WAH dimensions have a direct impact on the level of stress concentration at that location. Those dimensions are dictated by the code when designing a WAH and most certainly are not a means and methods issue, he adds. Hazleton suggests the cuts are weld termination holes, as suggested by the RFI.

 “This is a unique detail that requires a hole that is not covered by code,” he says. “These holes needed dimensions and finishing requirements to be considered by the design team and detailed in the contract documents.”

All parties involved, including TJPA, declined to comment in any detail beyond what was in the RFI response, until after the investigation is complete.

The timeline and fabrication sequence are important to understand why the First Street girders did not fracture, say sources. Herrick fabricated the two First Street girders, including the full-penetration welds, before it thermally cut the notches in the flanges. Hazleton explains that in March 2015, the inspector, armed with the RFI about the WAHs, directed Herrick to cut the notches into the already-fabricated First Street girders. Herrick complied, without grinding and polishing.


Soon thereafter, Herrick received a revised shop drawing (see pop-up image, at left), dated April 4, 2015, showing “cuts” in the bottom flanges. For Fremont girders, the fabricator proceeded to cut the notches before it welded the bottom flanges.

Same Connection Design

The First Street girders had the same connection design, the same material and no grinding of the holes, says Engelhardt. The only difference was that at First Street, the holes were thermally cut after the full penetration welding was done.

When Fremont girders fractured, which some say may have been in July, there was a sudden redistribution of forces that might have damaged the end connections of the girders to their columns. The MTC panel may do calculations to determine if anything is overstressed as a consequence of the energy released.

The MTC panel also is scrutinizing other structural elements with heavy plates to determine if conditions there might be favorable for brittle fracture, says Engelhardt.

Though there is much review work to be done, there is a sufficient understanding of the cause of the fractures to proceed with the repairs, he adds. Earlier this month, TT submitted a bolt-only repair scheme for review by the MTC, which the panel generally supports. The fix would bypass the fracture regions, where the test specimens were removed, by sandwiching the flanges, on both sides of the web, between 14-ft-long steel splints.

The TJPA plans to provide an update on the repair scheme and the investigation at its next board meeting, scheduled for Jan. 10.