Merged Discussion of femr's video data analysis

[femr2]

placement of their second datapoint was noncritical

See post above.

femr2's high-order polynomial approximations ignore the boundary conditions, so the acceleration they show at the left endpoint of their training data (which femr2 refers to as T₀) is quite sensitive to the positioning of that endpoint.

A particularly pointless observation in this context (which is not about T₀), but regardless, the Savitzky-Golay profile does not have that sensitivity.

At least one of femr2's criticisms of NIST's methodology is based upon his mistaken belief that NIST's nonlinear model is as sensitive to sample positioning as femr2's high-order polynomials.

Incorrect. There has been no suggestion of relative sensitivity levels. Suggestion that the NIST T₀ is ~1s early, sure. Note also that my later suggested T₀ (which relates to the NW corner of course) is based upon inspection of numerous profiles including the Savitzky-Golay acceleration profile...

T₀ = 11.8785s for my dataset is selected via inspection of:

• Raw displacement data
• Poly(10)_displacement model
• Poly(50)_displacement model
• Savitzky-Golay smoothed displacement data
• Wide-band symmetric difference smoothed displacement data
  
• Poly(10)_velocity derived model
• Poly(50)_velocity derived model
• Savitzky-Golay smoothed velocity data
• Wide-band symmetric difference smoothed velocity data
  
• Poly(10)_acceleration derived model
• Poly(50)_acceleration derived model
• Savitzky-Golay smoothed acceleration data
• Wide-band symmetric difference smoothed acceleration data

That's my best guess as to why he's been going on about NIST's sample positioning for the last several pages.

You didn't read back a few posts then. Your best guess is incorrect.

 

[oody]

Did you not read my post above ?


Fine, though I suggest ensuring that you read the thread so I don't have to repeat myself...

• The positional trace was not started until T=0.86s
• Rooftop structures obscured the roofline at T=0s
• Rooftop structures obscured the roofline until T=~1.6s
• Until the rooftop structures had passed, automated tracing of the roofile would be effectively impossible.
• Automated tracing processes are extremely unlikely to use inconsistent inter-sample frame counts, suggesting the trace may have been performed manually.
• Data from T=0.86s onwards suggests that early samples used the rooftop structures, not the roofline.

Times are all relative to the NIST T₀ of course, and this list may not be an exhaustive list of the features the datapoint separation data highlights.

Very good. Then my perception of your graph is accurate. The graph represents that regular samples were taken about 1/4 of a second apart, starting at some larger fraction of a second (0.86) into the process. And your dispute is with the starting point. That's fine. I think your graph obfuscates that point, rather than making it clearly. And I think your condescending manner and tendency to post in quantity rather than quality do not serve your cause well.

 

[W.D.Clinger]

femr2's high-order polynomial approximations ignore the boundary conditions, so the acceleration they show at the left endpoint of their training data (which femr2 refers to as T₀) is quite sensitive to the positioning of that endpoint.

T₀ = 11.8785s for my dataset is selected via inspection of:

• Raw displacement data
• Poly(10)_displacement model
• Poly(50)_displacement model
• Savitzky-Golay smoothed displacement data
• Wide-band symmetric difference smoothed displacement data
  
• Poly(10)_velocity derived model
• Poly(50)_velocity derived model
• Savitzky-Golay smoothed velocity data
• Wide-band symmetric difference smoothed velocity data
  
• Poly(10)_acceleration derived model
• Poly(50)_acceleration derived model
• Savitzky-Golay smoothed acceleration data
• Wide-band symmetric difference smoothed acceleration data

So your selection of the left endpoint for the training set used to compute your Poly(10) and Poly(50) polynomials was determined in part by those polynomials, which you could not have computed until you had selected that left endpoint.

 

[femr2]

your dispute is with the starting point

There is a number of features in the list you quoted from my post. The starting point is just one of them.

I think your graph obfuscates that point, rather than making it clearly.

The other points, specifically irregular time between samples, require graphing as presented.

 

[femr2]

So your selection of the left endpoint for the training set used to compute your Poly(10) and Poly(50) polynomials was determined in part by those polynomials, which you could not have computed until you had selected that left endpoint.

In part. Poly(10) and Poly(50) have been produced locally with all manner of ranges, not just the final one posted here. I have *Poly(10)* data which starts from my time 0 (>11s earlier) if you're that interested.

 

[oody]

There is a list of features in the list you quoted from my post. The starting point is just one of them.

The other points, specifically irregular time between samples, require graphing as presented.

Thank you for bearing with me. I really was having a problem understanding what you were conveying with that particular graph and why. I think perhaps alternate names for the axes might have helped, but I confess new ones don't some to mind quickly. I'd like to think on it and perhaps suggest alternative names for the axes at a later date.

 

[W.D.Clinger]

So your selection of the left endpoint for the training set used to compute your Poly(10) and Poly(50) polynomials was determined in part by those polynomials, which you could not have computed until you had selected that left endpoint.

In part. Poly(10) and Poly(50) have been produced locally with all manner of ranges, not just the final one posted here. I have *Poly(10)* data which starts from my time 0 (>11s earlier) if you're that interested.

Understood. The obvious question (as you may have anticipated but cleverly didn't answer): What criteria did you use?

Here's one possible answer: You selected the value for T₀ that maximized the discrepancy between the acceleration derived from your polynomials at T₀ and the acceleration derived from your Savitzky-Golay or some other smoothing of your raw data.

Since you have done the computations yourself, you must understand why I wrote "maximized the discrepancy" in that sentence.

 

[W.D.Clinger]

Thank you for bearing with me. I really was having a problem understanding what you were conveying with that particular graph and why. I think perhaps alternate names for the axes might have helped, but I confess new ones don't some to mind quickly. I'd like to think on it and perhaps suggest alternative names for the axes at a later date.

"i" and "x_i-x_i-1" would have been better.

 

[femr2]

What criteria did you use?

It's a judgement call following spending a lot of time viewing the data in various ways with varying processing. Has to be really. I'm sure there are a heap of other things I could add to the list, but hey ho. It's probably too long already. Perhaps inclusion of the Poly_acceleration entries in the list is suggesting an inflated importance to you, but the've been part of the process. More focus is upon velocity than acceleration for T₀ selection, and it does help to be looking at the profiles in greater detail at times...

 

[carlitos]

Does the use of passive voice help you deal with the fact that there is no purpose to all of this?

 

[tsig]

AE, anyone who believes that accelerations close to or exceeding g are normal doesn't know anything about the mechanics behind any collapse, including intentional demolitions.

I posted earlier to the first analysis of Frank Greening days after the release of the NIST function. He is a debunker and was probably the best poster on this forum.

He could see the problems with near g acceleration and the greater than g hump from his very first post in 2008.


In a different universe it is considered normal. He understood more in the first couple of days than you have understood in 3 years. Where in your universe do buildings collapse with this acceleration range?

Anywhere where g=9.8 m/s²

 

[tsig]

What is generally accepted for WTC1 is fake data and fake significant tilt caused by a fake south wall failure.

That is where you seem to stand with WTC1` and further proof of why real data is far superior to fake data.

Details discussed in the WTC1 features list thread for a long time.


So while you watch femr generating data, remember that you have ignored all his WTC1 data to date while forming your history and analysis on fake data.

If you were to switch to real data everything you thought you knew about WTC1 collapse initiation would go "poof".

This is why real data is so important.

And we get that real data by micro-analyzing videos right?

 

[carlitos]

Compressed, converted VHS copies of 30 fps videos actually. Using those allows micro-pixel analysis, time accuracy to the ten-thousandths of a second and error bands to like 15 digits.

 

[pgimeno]

Have any qualified experts told you near g accelerations as witnessed are expected or normal?

Yes.

Like the big blunder on the early WTC 1 movement, this shows how serious you are as a researcher.

Leaving your mistake resulting from assuming a wrong answer to the above question apart, who said I am a researcher, in the first place?

No sane human being could tell you that type of motion is expected or normal in a collapsing building, even one being demoed intentionally.

You mean especially in one being demoed intentionally. Physics don't prevent over-G accelerations if a building is allowed to fall by itself. If anything, that proves that it was a collapse. See my previous video; it shows how a façade can fall at free fall. Observe how the façade pivots over one of its ends, which ideally would cause the opposite end to reach over free fall acceleration (although in this case it is possibly not happening, as the pivoting stops near the end of the movement). That is just an example illustrating how over-G acceleration is possible in a collapse.

Perhaps if someone can produce one example recorded or filmed somewhere on earth of a building undergoing this range of acceleration you'd have a better case.

I have. Here it is again. Funny that it's you who complains about how many times you need to show something.

 

[femr2]

Compressed, converted VHS copies of 30 fps videos actually.

This Dan Rather footage converted from VHS ? Source please. (I imagine you are confusing with the Cam#3 footage NIST used)

30i fps, 60fps or 60i fps, or more accurately 29.97i fps or 59.94 fps but not 30 fps, actually.

Using those allows micro-pixel analysis

Centi-pixel is about the limit.

time accuracy to the ten-thousandths of a second

For frame timestamps which are specified in μs, of course.

Try dividing 1s by 59.94

and error bands to like 15 digits.

Nonsense.

 

[carlitos]

This Dan Rather footage converted from VHS ? Source please. (I imagine you are confusing with the Cam#3 footage NIST used)

Oh, if I am wrong, I sincerely apologize. Was this a Beta? I forget the provenance of that

I stand by the assertion that your claims of accuracy (to the ten-thousandth of a second, for example) are absurd on their face, and the number of significant digits you employ are ridiculous. That you can't understand why is your problem. It's self-evident in this thread.

 

[Major_Tom]

Pgimeno, tht was a very insightful answer. It is serious? That is your justification for accepting the WTC7 early acceleration profile as normal and expected?

When you say you are not a researcher, you certainly are not kidding.

And for anyone else? Is there another regular poster at JREF that may need a bit more analysis before reaching the same conclusions as pgimeno or do the regular posters agree with his logic?

 

[oody]

And for anyone else? Is there another regular poster at JREF that may need a bit more analysis before reaching the same conclusions as pgimeno or do the regular posters agree with his logic?

Quite possibly, people are simply choosing not to wander into your old-west saloon during a bar fight. I'm not sure you should draw too many conclusions from that.

 

[pgimeno]

Pgimeno, tht was a very insightful answer. It is serious? That is your justification for accepting the WTC7 early acceleration profile as normal and expected?

This was your claim:

No sane human being could tell you that type of motion is expected or normal in a collapsing building, even one being demoed intentionally.

And your request:

Perhaps if someone can produce one example recorded or filmed somewhere on earth of a building undergoing this range of acceleration you'd have a better case.

Which I fulfilled. Or do you believe that that wall didn't experiment this range of acceleration?

If you want to modify or retract it, you're welcome to do so. I rebutted it by offering an example of a free falling façade.

The WTC7 case is to me clearer since I saw the stabilized video you previously posted. It seems to me that the building's south side failed first, causing an horizontal movement in the north side as the building bent towards the south, and a curved one in the south side with a greater vertical component. The façade then arrested the horizontal component of the movement somehow (the why is still a bit of a mystery to me), and the north wall had to catch up with the downwards movement as the rest pulled from it, which would probably cause the peak in the downwards acceleration visible in the north side.

 

[Major_Tom]

Pgimeno, I understand that you sincerely believe this.

Is there anyone else besides Pgimeno that sees this acceleration profile for a large building to be normal and expected?

Any other evidence of it besides that wonderful video by Pgimeno?

This is about as far as you've come as a group by 2011? And is this the same type of logic that was used to determine WTC7 fell naturally?

If so, guys, I think I've located the main problems with our discussions and debate:

1) Use of fake data
2) Lack of the capacity to critically analyze observations


This seems the heart of our whole imisunderstanding