Miami Pedestrian Bridge, Part XIII 81

[JAE]

A continuation of our discussion of this failure. Best to read the other threads first to avoid rehashing things already discussed.

Part I
thread815-436595

Part II
thread815-436699

Part III
thread815-436802

Part IV
thread815-436924

Part V
thread815-437029

Part VI
thread815-438451

Part VII
thread815-438966

Part VIII
thread815-440072

Part IX
thread815-451175

Part X
thread815-454618

Part XI
thread815-454998

Part XII
thread815-455746


Check out Eng-Tips Forum's Policies here:
faq731-376

 

[jrs_87]

https://youtu.be/hBjntrebxj8?t=227

NTSB video appears to have misdated a crack photo. Note bolt or bolt shaped object on deck in this still and other(s).

 

[charliealphabravo]

That looks like the broken end of a black zip tie used to tie the banners.

 

[jrs_87]

Re: Video. The cracks do not appear to match March 10 progression. Seems this is March 14 or March 15 photo. Point is same zip tie is shown in still with another date. Not really important video anyway, so I digress.

 

[Vance Wiley]

Balance between demand and capacity of nodes and members as indicated on drawings

I realize what has gone down has already hit the ground and been defined by the NTSB. But the incongruity demonstrated in the contract drawings really bothers me. This set of drawings must be far below the standard presented by this engineering group.
One should expect the strongest connection at the point of the highest load and the strongest member at the point of highest load.
In a building, simple pad footings should be larger under the columns with the greater loads.
Without doing any calcs, I do not see that logical balance of demand and capacity in critical places in this structure.

[pre]Connection of nodes to deck:
Count and area of #7 hoops crossing shear plane.
Demand from NTSB attachment 73, Figure 63, Graph (eyeball interpretation).
EDIT - I now see a problem with interior nodes. First there is no Pc force - but
there is tension from the weight of the deck - like maybe 150 kips net.
There is PT clamping but that should be largely offset by the tension the PT was
intended to resist.

1. Node 1/2 5 hoops = 6 in^2. Demand 2600 K
2. Node 3/4 9 hoops = 10.8 in^2 Demand 1100 K Cap=10.8X60ksiX.8=583 Kips or X0.6=350kips as cast
3. Node 5/6 6 hoops = 7.2 in^2 Demand 500 K Cap=7.2X60X.9=388 Kips or X0.6= 233 Kips
4. Node 7/8 6 hoops = 7.2 in^2 Demand 500 K 388 Kips or 233 Kips
5. Node 9/10 6 hoops = 7.2 in^2 Demand 850 K 388 Kips or 233 Kips
6. Node 11/12 4 hoops = 4.8 in^2 Demand 2000 K 3.6 in^2 effective.


Notes: Node 1/2 Lots of contact area at this joint. Pc of about 940 Kips DL unfactored.
Larger diaphragm. Bars from member 1 contribute.
Node 11/12 One hoop outside shear zone. 3 = effective 3.6 in^2. Pc of 940 Kips DL unfactored.
3 - #11 bars from member 12 may contribute.[/pre]
Takeaway - 2000 kips demand at 11/12 has 3.6 in^2 dedicated reinforcing across the shear plane while 500 kips demand at interior nodes have twice that amount of dedicated reinforcing working. Interior nodes at deck are not adequate in shear friction using coeff of friction of 1.0 or 0.6.

Members:

Member 1 is 21” X 36” with 14 - #11 = 21.8 in^2 reinf and two hinges formed. The axial load is about 60 kips. Moment is not determined.
Member 12 is 21” X 34.5” with 9 - #7s and 3 - #11s =10.08 in^2 reinf and about 50 kips axial load. This is 10% less load, more than 50% less reinforcing, and no hinges provided. This member experienced pre collapse cracking of the north face because the larger #11 bars were all in the south face.
Member 11 is 21” X 24” with 8 - #7 bars = 4.8 in^2. 1% minimum area of reinforcing is 5.04 in^2. Reference Section B-B sheet B-40. Can anyone confirm more reinforcing in member 11? Load in 11 is 2000K slide /cos 31.8 degrees = 2350 kips. Member 11 cracked severely before collapse.
For comparison, members 6 and 8 are 21” X 24” and have 10 - #7 bars = 6.0 in^2. That meets the 1% minimum reinforcing required. The axial load is estimated (by me) to be less than 1000 kips.

Takeaway for members is that significant incongruities exist on the contract drawings regarding capacity and demand between members.
Member 11 likely did not meet minimum reinforcing requirements (pending information to the contrary).

It seems the designer's staff had difficulties with the complexity of this staged ABC construction. No telling what might have developed in stage 3.

 

[Hiway.1]

Vance Wiley
I'd be interested in your take on a report uploaded to the NTSB docket here on October 22:

Link

I'm not a structural engineer but it appears to provide the first clear explanation how Figg improperly analyzed the shear load on the 11/12 node and grossly under-estimated the shear load on the interface at the cold joint.
It's a report by Modjeski-Masters that details alleged errors in Figg's calculations in properly extracting loads from the finite element model. Apparently they are under hire by Louis Berger from what I can see and I may be wrong.
They claim in the report to repeat Figg's error in interpreting the finite element model and show how they should have used the model to extract loads 2-3 orders of magnitude above what they actually used.

 

[Vance Wiley]

Rather than keeping silent and having everyone think I am just dumb, I am going to ask this and remove any doubt.
What does one use for the "lower bound" of structure weight ( Pc ) across a shear plane?
Would that be 1.0 X DL? The NTSB hit the design team with that criticism.

 

[Vance Wiley]

This FIU fiasco is a good illustration of why not being a structural engineer is a good career decision. And that does not surprise me when I see the common sense evident in your recent post.
I will read the report in your link. I am probably not the person to do a critical review of the use of a
FEA program. I am an old guy out to pasture.
I was learning Fortran in 1962 and by 1980 I was translating Wilson's ETABS into interpretive Basic for use in a CP/M computer with 64K ram and two 600K floppies - 8".
I will get back to you.
Perhaps the FEA gurus will jump in?
Thanks,

 

[Vance Wiley]

I may wish I had waited a bit but here goes.
The Modjeski - Masters Report is a review requested by attorneys for Berger, it appears. The date is 10/19/2019 so I doubt it influenced the release by NTSB this week.
The first thing evident is the posturing in favor of Bergen. I have not studied the contracts between FIGG and Bergen and I cannot venture an opinion on the accuracy of the information presented. Nor do I have experience in peer review of highway structures, so I best be silent there also. The comparison presented between FDOT and MnDOT seems to show FDOT has more interest in the overall layout of traffic and "fit" if you will, while MN seems to focus a bit more on the structure and details.
The explanations of LUSAS software use, mesh sizes, standoff distances, and decreasing mesh size or changing from linear to quadratic definitions seem plausible to this untrained eye.
What stands out to me is the amount of correspondence with LUSAS tech support. It seems the staff engineer designing this structure was self training and should have had more supervision.
M & M apparently duplicated the findings and identified the method of interpretation by the design engineer and from their experience can define the errors made. What is particularly disturbing is the statement that after learning how to get good answers from the software, the design was created from the earlier results. One problem I have seen with too much output (did I read 3400 sheets?) is that it all looks alike and nothing catches one's attention.
This project needed the oversight of an experienced engineer who could step back and see the entire picture while seeing the small things. The old saying "Take care of the dimes, the dollars will take care of themselves" seems appropriate.
And no doubt the history of M & M with NTSB will give weight to their report.
I apologize if this sounds like a poor book review. My thought is it will benefit Berger's position.
Thank you,
PS - I reserve the right to edit anything that sounds 'reel stoopid' -

 

[Michael L]

When you perform a slice in LUSAS, you specify which part of your model the software considers when it is integrating elements that intersect that slice (defined as “Extent” in the dialogue box, and you choose a pre-determined “Group”). The report suggests that one side of the slice was not “active”, which may suggest the “Group” chosen was the diagonal strut brick elements only, and the deck/roof elements were not considered. Given this truss bracing member should have similar axial stress (if not bending moment) along its length, it seems logical (with the benefit of hindsight) to slice at least 1 or 2 elements back from the face, at least to be sure you’re not affected by any averaging between adjacent elements, shape functions or other complexities of finite elements.
Can’t help to think whether a brick element model is more complex than is really needed for a truss diagonal carrying fairly uniform DL and pedestrian loading.

 

[Settingsun]

The Hyatt collapse is how I respond to criticism that my new design is heavier than some dodgy contraption that's 'stood the test of time'. The Hyatt had something like 10% of the required live load capacity [(failure load - dead load)/ design live load]. Still stood for about a year. If it had even 30% it may still be standing today as a ticking time bomb. The test of time is tough in some ways and mild in others.

The lack of action before the FIU bridge collapse seems like a case of everyone not wanting to be the whimp by saying they need to stop. A well-known human trait we should actively look out for in ourselves. Link below.

https://www.dowellwebtools.com/tools/lp/Bo/psyched/16/Smoke-Filled-Room

 

[Pete_K]

The explanation in the M&M report about result accuracy vs. mesh size (and convergence due to mesh refinement) is something I would expect anyone doing competent FEA work to know. If Figg's stress analyst (who repeatedly in his interview stated that he is a stress analyst and not a designer) didn't already know this, but his work was being used to determine demand, that is a major problem. That is basic FEA knowledge.

 

[The Mad Spaniard]

I wish I had access to the output of all the analyses performed to look at the forces in the elements for all the relevant loads (DL, PT, and Construction LL) independently. This is because in the reports It is not clear to me if they are refering to Service loads or factored load combinations (Strength I) when they mention forces and compare them. It is confusing.

Regarding the M&M report, I was shocked to see that the results were taken from a LUSAS mesh of 1ftx1ft. This is to represent connection that has an about 2ftx2ft incoming member. The 1x1 elements are too large for what you are trying to eveluate. I would have been more confortable with a 0.25x0.25 and another 0.4x0.4 to compare results (and two different types of elements plus some basic hand calcs). A 1x1 is more that enough for the deck and canopy but not for the connection.


Again, another example of FE output fooling the analyst that may have a small budget to work with. But the pics are impressive (and pretty), so they must be right.

My rule is that "hand calcs"(in this case truss analysis) usualy puts you within 20% of the correct number. Trust but verify.

 

[PA PE]

I'm looking to retire myself in a few years. In college I was interested in structural engineering and took all the courses, including prestressed concrete. After graduation, I found work doing water/wastewater work and thankfully have been doing that for 38 years! It's still fun to make "suggestions" to the structural engineers on the projects.

When I started in consulting there were no personal computers and all calculations were done by hand. I got good at it and never adopted the computer software. We have had some complex projects that used computer models, and of course this work goes to the young, tech savvy, engineers who don't have the experience to recognize questionable results. With experience, and some creativity, it is possible to come up with a way to do a hand calculation that can get an approximate answer in order to provide a reality check on the computer output. When I was a young engineer, an older engineer gave me the best advice ever: "It's better to be approximately correct than exactly wrong." That's been my motto ever since and I am trying to impress that on the younger generation now.

I'm looking at this disaster in more of a qualitative way, not being able to follow the technical discussions of load factors and such. However I am getting the impression that these calculations, even if done correctly, are working with rather small factors of safety. Is that correct?

Aren't the various published design guides and standards to be taken as a minimum? In my opinion, a good conservative design has to take into account uncertainties - settlement of a footing, contractor's mistakes, accidents, vandalism, the future, etc., and then add a little more, "just in case". It's really no consolation if you can find someone to "blame" for a disaster - It's still a disaster. Would a structure like this survive a truck crash and fuel fire underneath? It's no accident that the structures built 100 years ago are still standing.

Enough of the soapbox, I better stop now.

 

[JStephen]

Just to point out, the structures built 100 years ago are still standing because you're only looking at the ones that remain!
Long ago, I got involved with some finite element work and ran into issues trying to extract overall loads from it somewhat like Figg did. But not having used it for 15 or 20 years, I would have figured that problem was all "fixed" now, but apparently not. And in my case, the whole point of using the FEA was that the "approximately correct" analysis was shown to be incorrect to some extent.

 

[PA PE]

JStephen: Ha, good one. You got me. Therein lied the danger of being up on a soapbox for too long.

 

[samwise753]

Rather than keeping silent and having everyone think I am just dumb, I am going to ask this and remove any doubt.
What does one use for the "lower bound" of structure weight ( Pc ) across a shear plane?
Would that be 1.0 X DL? The NTSB hit the design team with that criticism.

Load Resistance and Factor Design typically specifies a 0.9 load factor as a lower bound for Dead Loads. I have seen this both in ASCE 7 and AASHTO Bridge Design Specifications. The intent is to use 90% of the dead load in instances where you are counting on dead load to help you. For example, determining uplift loads on foundations and piles, or, in the case of this bridge, the Pc clamping force. The provision they used from the AASHTO code for interface shear doesn't explicitly state using 0.9 or 1.25; you have to go back to the load combinations chapter where it tells you to use the factor that gives the most conservative capacity estimate and/or the worst case load demand.

 

[samwise753]

Was a FEA even necessary here? Why couldn't a simple STAAD/RISA model be used instead? It seems a FEA over-complicated the issue.

 

[Vance Wiley]

You don't get off that easy. You could easily follow the factors.
My take on LRFD is it was intended to better predict failure conditions and if you know exactly when it will fail you can work closer to the edge. Somehow it was also supposed to provide uniformity in performance of the various components and thereby make structures more efficient (read cheaper).

So the loads are boosted and basically compared to 90 % of known failure of components.
In the case of this pedestrian bridge - which is a heavy structure - the actual weight is boosted by 25% using a Load Factor of 1.25 and the Live Load is boosted by 75 % by using a factor of 1.75. In this structure weighing over 10 kips/foot that becomes 10k X 1.25 = 12.5 kips/foot. Live Load of 90 psf and 29 feet wide is 2.6 k/ft and factored becomes 2.6k X 1.75 = 4.6 K/ft. Total is 12.5k + 4.6 k = 17.1 K, up from 12.6 k/ft. That gives us a SF = 17.1/12.60 = 1.36 against specified failure of components. When we consider possible variations in the anticipated performance of the materials and components by using a factor of 0.9 the design SF becomes 17.1/12.6/0.9 = 1.67 overall if the construction and all components perform as specified. In a structure half as heavy the SF would increase.
Using concrete in bending, the old days used allowable compression stress as 0.45 F'c which was a SF of 1/.45 = 2.22. So yes, LRFD works closer to the edge.
And factored loads do not indicate how the structure actually performs under service conditions. So deflections, creep, shrinkage, and elastic response still have to be determined using real loads and real properties of materials. So the net effect of LRFD is full employment for engineers.
I designed to real loads and allowable stresses and then did LRFD for the federales.
Another two cents - - because of the critical performance of the web members in this structure, I suggest using reinforcement levels between the specified range of 1% and 4 % which match the demand/capacity ratio. Example - if the member is half loaded ( demand/capacity = 0.5) use 2.5% reinf (1+.5 X (4-1) = 2.5). This should probably be used in building columns also. Losing one of those could also ruin your day.
And in a structure like this there should be lots of confinement reinforcing. Like in seismic design.
There is nothing in the codes that say the finished structure must be designed to just barely not fail - at least not yet. The designer can provide as much capacity as he deems necessary.

 

[Vance Wiley]

Thank you for the lesson on "lower bound".
Makes sense. Used the same idea when designing for seismic loads to account for vertical acceleration.

 

[Ingenuity]

In MCM's post-failure submittal to the NTSB ["MCM Party Submission; Findings, Conclusions, Recommendations, and Attachments"] document #628566, Section 5 "Engineering Analysis and Discussion", MCM present several analysis methods and calculations:

Basically, they compared SAP2000 and ABAQUS to some hand/2D calculations, and noting that they are within 3% of each other.

MCM make a compelling statement:

As discussed later, the value of 1,710 kips for interface shear at Node 11/12
arrived at by simple hand calculation greatly exceeds the value of 571 kips that FIGG extracted
from its LUSAS finite element model (see Section6.2.1). A hand analysis should have thus
revealed to FIGG that it had greatly underestimated the shear force at this critical connection.

I was generally impressed with MCM's tech submittal content to NTSB - I assume they engaged an engineering consultant to assist with the analysis section, because as they state: