Posts about: "Audio Analysis" [Posts: 51 Pages: 3]

Visual evidence the rat was deployed, audio evidence the rat was deployed, evidence the APU inlet door was open, evidence the gear retraction was interrupted, evidence there was very little engine noise after departure, very obvious evidence that the aeroplane didn’t have enough thrust to stop it descending into the ground. All of this strongly suggests that both engines were to all intents and purposes, and for want of a better word, ‘failed’ now that could be intentional, accidental or because of some sort of technical malfunction or external factor. Nothing conclusive and no answers as to how or why, but not quite as wildly stabbing in the dark as your post would infer.

We agree that there was a lack of thrust. Possibly caused by a dual engine failure. But the sharpest frames in the video do NOT show the RAT and this is counter evidence to the RAT theory. If there were substantial technical failures who knows what sounds could be generated. I find the evidence weak at best. And we immediately get into a chicken-egg problem: did some power issue of unknown nature cause an engine failure or did a dual engine failure occur, resulting in a power loss? Both are extremely unlikely and need to be backed by quality evidence. The video is not it, in my opinion. I don't know the APU intake mechanism and whether it could open after the impact.

I realise the discussion has largely moved on, but for completeness I have analysed, filtered and compared the audio from several 787 videos\x97both take-off and landing (the latter with and without RAT deployment; take-off footage with the RAT out is rare).

Using the same style of frequency plots employed in the RAT analysis\x97and drawing on speed of different engine components data supplied by other contributors\x97it is straightforward to identify the engines\x92 high-power acoustic signatures objectively, despite variations among clips and the presence of Doppler shift.

What is clear is that the engine noise in the recording bears no resemblance to a typical 787 take-off profile. Whether the engines were merely at very low thrust/idle or fully windmilling I cannot say with certainty, but they were certainly nowhere near take-off power.

Make of that what you will.

The investigators will certainly soon know, but how sure are you that they'll share that in public?

Nope, you may not have read the whole thread - the full set of conditions which can trigger the RAT have been posted multiple times before your reply

I'd phrase it differently: the frequency (and harmonics) of the noise are consistent with it being a RAT. If you then assume that it was a RAT and infer distance and speed curves from the Doppler variation, you get plausible values for an aircraft. But the two things are somewhat linked so you can't really treat the second thing as confirmation of the first. (FWIW, I'm personally quite satisfied that it was indeed a RAT on the audio.)

I've been sitting on my hands for days now...but please stop linking RAT deployment evidence with some blurry cluster of pixelation, which proves nothing.
This Pprune-forensic audio analysis (and subsequent posts) by Kraftstoffvondesibel, although completely ignored by many it seems, is the only clear proof there was a deployed RAT sound recorded in the young boy's video clip. You may disprove the analysis as much as you like, but referring to blurry video regarding possible RAT deployment or not, is indeed " thousands of people are wasting countless hours "

I'm rather under the impression that many recent discussions are going backwards or h amsterwheeling.
But thank you for your valuable audio analysis

If you’ve read the whole thread, you’ll see that most of us who are acknowledging rat evidence are also citing this compelling audio evidence too.

if you had read the whole thread, you would have seen ths 2 independent audio analysis of the HQ video soundtrack, comparing with 2 other segments. It has been established that with 99% certainly that the RAT was deployed. Not to mention the ear witness who has heard 500\xd7 RATs close to Boeing field

The audio analysis in this case isn't anything obscure or vague. The same software is used in courts all over the world every day. You can be pretty sure the same software in some iteration is used investigating CVRs all the time. It shows something emitting the exact same sonic signature as a B787 RAT went over the camera at the same speed as the Boeing 787 seen going over the camera. It isn't a broken fanblade, or a motorcycle passing by, that would give a completely different result. It literally means hundreds of thousands of datapoints fall into the exact same pattern. If it were anything else, that would be easy to observe. It is a much more accurate tool than watching heavily compressed video. It is the same software used in hollywood to surgically remove airplane noises from historical dramas, btw. At least until AI came along and took on that job about 2 years ago. . Just because you aren't familiar with it doesn't mean it isn't accurate.
Just putting this out there, since many might be unfamiliar with this, you included.

Maybe you should describe your analysis then and present the data? I'm a physicist so chances are I might actually not be so unfamiliar. Please include an analysis of raw data quality, spectral resolution and binning as well. And don't state it is the 'exact same' as this is statistically impossible.

As one of the two authors who have offered an acoustic analysis of the AI 171 'RAT' video here I feel obliged to make some observations.

First of all Kraftstoffvondesibel and I did our work completely without knowledge of each other. We reached the same conclusions more or less, most likely using different software (I used four different methodologies) and in subsequent discussions between us we've found our specific results (frequency etc) match very closely - including the doppler shift that we've also both commented on.

While this doesn't absolutely prove the RAT scenario (I was, at least initially, somewhat more circumspect on this matter as anyone reading my first post would find) it does go some distance towards reconciling our respective methodologies and outcomes. Given the confluence and discussion I am now more satisfied that the RAT was deployed than I was before - although for the reasons espoused earlier I totally recognise the [scientific] challenges to this view.

In this regard I too would not 'mind' if our view regarding RAT deployment were proven incorrect. What Kraftstoffvondesibel and I have done is simply use a little science and apply a quantitative analysis to the available data that many had qualitatively argued over. One might hope that, amongst other things, it could have dispelled the RAT/no RAT question to some degree and reduce thread noise, but unfortunately it seems some haven't read the thread through, or perhaps lack analytical reasoning.

This site isn't the place for a full academic paper+peer review and for me the real subject of this thread is the tragedy of AI 171 and what might determinable from what we know in the hope it will be useful to those closer to the coal face. If you are a physicist, scientist or engineer and have the capability then the same data we used is also available to you to do your own analysis. Should you do so we'd welcome hearing the outcome - whether it concurs with our results or not - as this is the scientific way. If you want to include the full nitty-gritty detail of how you've gone about then for a very small percentage of us that might be interesting, but it'd be peripheral to the main issue - and bear in mind that RAT deployment itself is merely an indicator of other potentially more serious issues. It is not likely to be the cause per se !

To conclude, I understand it's a long thread but as far as the RAT question is concerned, I'd encourage you to read the relevant posts at least and if you remain unconvinced, and you have the skills, then why not conduct your own analysis and let us know the results? Otherwise perhaps as a group we could move on from this, remembering that this is a terrible event which at the very least deserves informed useful discussion rather than wild and/or repetitive speculation.

FP.

I don't believe I owe you anything, I believe this is done adequately previously and has already taken up enough time on this thread. I am of the opinion that we have shown the RAT being deployed satisfactory enough to be of use for speculation in this thread. I find repeated comments about the bad video being the only evidence a bit disrespectful, though. Even from a mere physicist. It is based on a spectrogram over time. The source file shows audio up to about 16 kHz, it is unknown whether this limitation is in the file format (ie. 32kHz sampling rate) or microphone. Doesn't matter much. The frequencies above 16kHz is not important in this context as it is not where the sound energy is anyway. The audio will have been lossy data compressed, but it does not affect these prominent properties of the audio. It does make me hesitant to draw conclusions from the parts of the spectrum with more broadband noise and several intersecting sounds. Noise floor suggests 16 bit sampling depth. Spectral resolution? N/A All samples are included. The spectrogram covers the entire frequency range recorded, It shows comparatively the same overtones of the fundamental expected from the technical specifications of the 2 bladed RAT running at it's intended RPM, the doppler characteristics fits completely with a reasonable range of passing speeds and distance to the passing source plotted out. Compareatively, All the harmonics are identical both in pitch and seperation to a recording of a known B787 landing with RAT deployed, while the Doppler fall shows a longer time frame in the landing video taken from a further distance. As expected. The overtones easily discernable in this recording falls in the 220-2700Hz range. Below that, there is other noise centered around 150Hz, which gradually fades towards the end of the recording. This, as far as I can find in available information, fits with an idling or even windmilling B787 engine, but this is not conclusive. This falls in a range of the spectrum where there are other noise sources and the signal/noise is low and of a broader band characteristic, these masking frequencies is where the lossy data compression might play tricks, so I do not weigh that heavily. Recordings of landing B787 without the RAT, shows none off the same characteristics, and completely lack the tonal components and exact overtones shown with the RAT deployed. More importantly, compared to videos of B787s taking off with normal take off thrust, the latter shows distinct tonal elements, but with very different overtones,, both in separation and composition, again possible to relate to known quantities of the rotational speed and elements of the engine at high power. The AI recording shows none of this.

The latest techniques let us separate such things as reverbration from the source, when superimposing the reverberation/ambience and background noise of the AI crash urban environment on the clean, dead open field recording of the known B787 w/rat, they do indeed sound exactly the same to this very skilled and experienced listener. Although this is not courtesy of the computer analysis. It is just another angle of confirmation.

All in all, i think this source audio is excellent. The source is an iphone, their mems based microphones, although noisy shows great spectral balance and is comparable to basic measurement microphones of professional application. There is plenty of information to analyse from in this sample.

And again, I can't see it in the video either, and until I put on some really expensive headphones and fired up the software I was of a different opinion. I bowed to the science.

Edit: I took an extra look, I am prepared to say the fall off at slightly above 16kHz is from the original recording, this is probably a limitation in the microphone, as it is not a hard cut-off before a 16kHz Nyquist frequency as it would be with a 32kHz sampling rate, there is dither noise from 16-20kHz fitting with the source being 16 bit.

Amen to that!

Now, if you and Kraftstoffvondesibel wouldn't mind conducting similar audio analysis of a moped whizzing past, that might help to dispel one of the most popular competing theories.

I appreciate the time taken to produce information that might help us advance our understanding. I am not qualified to mark your homework, but I can recognise that you have spent time on it, shown your workings, and self-reported the limitations. That sets you apart (from some) and earns my respect.

(I just hope you haven't been proven wrong in the time it has taken me to post this...)

While I get your light intent there is a reasonable question there, and one which was posited many posts ago.

There are various reasons why the signatures we obtained are not that likely to be from a motorbike, but to answer briefly - and pictures being 1000 words 'n all that - here's a quick comparative [visual] analysis. At the top is a motorbike passing by away from the observer, the other is AI 171 doing the same thing:




I expect the difference will be clear. That said one could level a number of criticisms at the method I've used to show this, the sources obtained etc (eg. this is motorbike, not a moped as I couldn't find one in the time available), and if you really wanted to fake it one could, but I doubt that's occurred here.

Would say more but I'm short of time and think this is sufficient clear for now.

FP.

Alright, just because that isn't very difficult. The first segment is a Moped/motorcycle whizzing past. Not from the same city, but it is in India, and I listened to a number of them to find a representative one, and the closest I could find in lenght/timing that isn't completely out of bounds with the streets in the area the AI video was filmed. it is taken uncompressed from a professional sound library and is a very high quality recording. The sound source is in direct line of sight, much like the AI recording when the aircraft is presumably just coming into the clear from above the roof. A moped/motorcycle in the AI video, would probably not be direct line of sight down on the street, and would have an even more ambient character, but shorter ambience. The second segment is the Air India video.
One thing we haven't discussed as a characteristic is how the AI recording moves from indirect sound with a shorter ambience, to direct sound, to a very much more ambient distant character in the same amount of time as the large aircraft moving overhead.
The moped/motorcycle in the sample, keeps about the same small street ambience throughout.
Notice how the harmonics are in different places with different spacing and with different relative intensity, the emphasis on the internal combustion engine is lower in frequency than the open Rat, but with added exhaust noise.
Thank you.



I also enclose a doppler distance

I hesitate to chip in in these accident threads. Keep them clean. However, as as a few comments above brushes my audio expertise, I will comment.

A very simple audio analysis give me this:
The 3 segments horisontally, are of different videos of B787s passing overhead/landing. The vertical drop you see is the doppler effect.
In other words, these are spectrograms over time which makes these distinctions easier than a simple static spectrogram.
1. B787 landing with RAT extended.
2.Air india crash
3. B787 landing without RAT


It's a 5 minute laptop job, and it would look much prettier and clearer if I spent some time with it, (Gain to color match, and spectrally match to compensate for microphone placement and type),
but it is 85% conclusive even when done as simple as this IMO.
(I do have legal forensic audio experience)
The RAT was out judging from the audio evidence. You can see the the equally spaced overtones of the propelller match when passing overhead resulting in the Doppler effect, the difference in length of the doppler is caused by distance and the slightly varying frequencies shown in the starting point is caused by a difference in speed. But the harmonic content match.
In the 3rd segment you see none of these overtones at all.

I hesitate to chip in in these accident threads. Keep them clean. However, as as a few comments above brushes my audio expertise, I will comment.

A very simple audio analysis give me this:
The 3 segments horisontally, are of different videos of B787s passing overhead/landing. The vertical drop you see is the doppler effect.
In other words, these are spectrograms over time which makes these distinctions easier than a simple static spectrogram.
1. B787 landing with RAT extended.
2.Air india crash
3. B787 landing without RAT


It's a 5 minute laptop job, and it would look much prettier and clearer if I spent some time with it, (Gain to color match, and spectrally match to compensate for microphone placement and type),
but it is 85% conclusive even when done as simple as this IMO.
(I do have legal forensic audio experience)
The RAT was out judging from the audio evidence. You can see the the equally spaced overtones of the propelller match when passing overhead resulting in the Doppler effect, the difference in length of the doppler is caused by distance and the slightly varying frequencies shown in the starting point is caused by a difference in speed. But the harmonic content match.
In the 3rd segment you see none of these overtones at all.

There has been much discussion here about RAT deployment. Various claims either way have been made, based on individual perspectives of available video and audio.

I am very mindful of just how awful a tragedy this is, and have significant misgivings about disproportionate interest in others misfortune where it carries no purpose, but also recognise that for some people knowing and learning what happened ASAP could be very important, particularly given the present circumstances.

Thus while I sincerely hope that early detail from investigators will give some clarity, in an effort to reduce needless speculation regarding RAT deployment I have taken:

(1) an audio sample from the video of AI171 passing by in which people claim to hear a RAT
(2) an audio sample from a 787 video with RAT deployed on test by Boeing
(3) an audio sample from a JAL 787 video with RAT deployed

And passed these through a FFT in order to gain a more quantitative view of the noise spectra from each event.

A spectrogram of the results is presented below. I hesitate to make any conclusions per se, but observe that there are similarities as well as divergences between them. In all three samples there is a relatively consistent signal roughly centered in the range 113-146Hz that could be what gives the characteristic 'buzz' sound of (apparently) a RAT in operation.

JAL ~141Hz
Test ~146Hz
AI171 ~113-134Hz (prob doppler variation here)


Spectral comparison AI171, B787 with RAT, JAL 787 with RAT



Spectral comparison #2 AI171, B787 with RAT, JAL 787 with RAT


These frequencies seem consistent(ish) with what I got from this video [[url]https://www.youtube.com/watch?v=O1r3CuRwjPc] in which a 787 RAT is being tested - albeit in this case the blades are hydraulically powered and not driven as a turbine. This test showed a fundamental frequency of 135Hz with relevant harmonics above (the second harmonic at 270Hz is higher SPL, no weighting):




It's important to note that the initial recordings are necessarily different; these are not controlled conditions, the recording equipment is probably quite diverse and almost certainly not ideal, and the environmental conditions will also be different. Moreover all of these audio samples have come from video files referenced here, one has no way of determining the provenance or veracity of these sources and, crucially, I have no prior experience of analysing/extracting RAT acoustic fingerprints (nor have I sampled 'control' audio of a 787 passing by /without/ RAT!).

Additionally it's been quite a long time since I did any work with [turbine] noise so given these and other variables I'm not prepared to make any declarations per se, but perhaps more knowledgeable people could. That said, my feeling from what I see is that RAT deployment is not dis proven, and that the apparent fundamental frequency difference between the samples may be explainable by - amongst other things - difference in a/c airspeed, bearing in mind that AI171 was on TO, the others were landing.

Ultimately what I've done here is extremely rudimentary and while it would be possible to go into much more depth I'd hope that more definitive answer would be forthcoming by then, however if anyone wants to discuss specific methodology etc off-line please PM, no wish to add to noise on this thread.

FP.

It was hard to let it go, so I spent a bit more time with the audio, using filtering and matching to see if I could be even more sure.

It's a pity uploading audio to this site isn't as easy as uploading photos, but I can say it took very little filtering and matching to make the Air India audio become nearly indistinguishable from audio taken of B787 with known RAT extended during landing.

I can't see it in the photos either, but in these circumstances the audio is a lot more trustworthy, and from my audio point of view RAT deployment is 100% confirmed.

In the off chance that the audio I borrowed from a confirmed RAT event was somehow faked, I plotted the technical data I could find of the B787 RAT (4000rpm, 2 blades) combined with a height estimate and asked the O3 model with deep research to estimate doppler shift and speed.
The result matches the above documented 270-200Hz (in one of the harmonics) Doppler shift observed in 1.7 seconds.

Further to the audio analysis of a deployed RAT; given an audio print, known noise source [RAT] frequency, location of a receiver and relative location + direction of moving object, and using the principles of doppler effect etc, it should be possible to calculate [ground]speed of a passing aircraft.

Thus, having been informed that the source frequency of a RAT is 145Hz (thank you that person - please ID yourself if you wish), and with regard to the [assumed] RAT acoustic print available for AI 171 we are close to being able to determine an approximate airspeed during descent (given the video from which we get this acoustic data appears to start during the descent).

There are several variables in play here that we don't know exactly (the positions of observer and aircraft, height agl of observer and aircraft etc), and some I will ignore for the moment (effect of air temperature/pressure for example) but if one makes an educated assessment from the visual cues in the video I calculate an initial relative airspeed of circa 120kn. Necessarily this is an extremely rough number, in fact the range (making various positional assumptions etc) appears to be between ~100kn to ~150kn, but if it turned out be either side of this I wouldn't be completely surprised.

As with the initial audio analysis to determine whether a RAT was deployed or not I offer this for those to whom this sort of data may be important. Not sure if it's of any real use or not, and I wouldn't hang your hat on it, but here it is in case it assists. Should it matter I expect some sort of corroboration could be obtained by making an assessment of vertical speed from the variously available videos and extrapolating from there.

My previous caveats over veracity of source data etc remain.

FP.

Correct. That was the original purpose of the calculation. In addition to the sound itself having the measurable harmonic signature from other rat videos.
What this plot also does however is tell you the speed if you know the height or height if you know the speed.

The iphone used to film this were pictured somewhere, knowing the iphone model, and thus the characteristics of the camera, and the dimensions of the airplane it wouldn't be impossible to calculate height from the video imo.

Just throwing it out there if anyone sees the use and feels the call.

My personal amateur speculation still centers around the cut off switches.
I have spilled coffee and sweet tea over complex electro/mechanical switches/panels before(large format audio consoles with 8000 buttons) and seen unexpected things happen.

I am sure the switches are spectacularly well built, but they are in close proximity and thus prone to the same external factors.
Does anyone know if these two cut-off switches in such close proximity has the exact same installation, or they differentiated in some way that makes a freak failure mode in one not neccesarily affect the other the same way?