Ethiopian airliner down in Africa
The software in the two ADRIUs is different, it has to allow for the fact that the two vanes are on opposite sides of the cockpit, so I believe they probably use different correction tables, or the algorithm to apply to them is subtly different, possibly something as simple as using a wrong sign.
Join Date: Nov 2018
Location: Vancouver
Posts: 68
Likes: 0
Received 0 Likes
on
0 Posts
...
2) Looking at a picture of the AoA module, I am presuming that the A to D convertor is in the body, and it is of the ‘rotary’ kind, i.e. the A to D converts the rotated position of the vane shaft into a digital signal. Most of the A to Ds I worked on were typically 12 bit parallel output, I suspect that more modern ones would be a serial output. Either way the signal would be very noisy, responding to every vibration that the vane felt....
2) Looking at a picture of the AoA module, I am presuming that the A to D convertor is in the body, and it is of the ‘rotary’ kind, i.e. the A to D converts the rotated position of the vane shaft into a digital signal. Most of the A to Ds I worked on were typically 12 bit parallel output, I suspect that more modern ones would be a serial output. Either way the signal would be very noisy, responding to every vibration that the vane felt....
it looks as though, at least for the NG, the AOA sensor provides an analog output to both the ADIRU and the SMYD box.
I couldn't find a diagram of the ADIRU Analog interfaces, but here is the SMYD Analog Interfaces diagram.
The three AOA sensor wires are labeled "SIN", "COS", "COM", whose meanings are obvious.
[img]https://3.bp.%62%6C%6F%67%73%70%6F%74.com/-qCwEtMS3-yU/W-xF4QyEZLI/AAAAAAAAFCo/mpLqB83Yau4ph1pPKUjgqSKBFmPoYOWugCLcBGAs/s640/Screen%2BShot%2B2018-11-14%2Bat%2B7.56.25%2BAM.png
3) The offset is not due to a ‘stuck bit’. The offset in AoA as the aircraft first started to move is about 12 deg, increases to about 17 deg and then settles on about 22 deg. A stuck bit would be 22.5, or 11.25, or 5.625; to get 17 deg would take two ‘stuck’ bits. Besides which serial buses of themselves (e.g. ARINC 429) do not have ‘stuck’ bits, they have multi bit corruption, usually recognised as a ‘bad message’ and rejected.
4) The AoA digital signal goes into the applicable ADIRU for further processing. Part of this processing *should have been* to reject any invalid input!! Another part will be to smooth the data to get rid of the noise, but this is the same software in both ADIRUs, so would not give an offset.
5) ‘Light bulb’ moment -
The software in the two ADRIUs is different, it has to allow for the fact that the two vanes are on opposite sides of the cockpit, so I believe they probably use different correction tables, or the algorithm to apply to them is subtly different, possibly something as simple as using a wrong sign.
I know which bit of code I would be diving into now!
4) The AoA digital signal goes into the applicable ADIRU for further processing. Part of this processing *should have been* to reject any invalid input!! Another part will be to smooth the data to get rid of the noise, but this is the same software in both ADIRUs, so would not give an offset.
5) ‘Light bulb’ moment -
The software in the two ADRIUs is different, it has to allow for the fact that the two vanes are on opposite sides of the cockpit, so I believe they probably use different correction tables, or the algorithm to apply to them is subtly different, possibly something as simple as using a wrong sign.
I know which bit of code I would be diving into now!
The AoA disagree is pretty close to 22 degrees from half way down the runway onwards. I suspect that at low speeds, say during taxi, the AoA vane does not respond to the airflow accurately.
I still quite like the One Bad Bit explanation although I agree that it could be a software issue. As you clearly will know it is quite common for software (in general, not in necessarily in aviation) to set a memory Bit to store a status. Perhaps the wrong bit got changed?
Do these computers have hardware memory protection such that one "process" cannot interfere with another? I suspect that they will not have such protection but I am not sure. I read somewhere that they were 80286 based . The 286 does not have integrated hardware memory protection although I suspect that an addition Memory Management Unit might be able to provide it.
Join Date: Jul 2013
Location: Norway
Age: 57
Posts: 140
Likes: 0
Received 0 Likes
on
0 Posts
I have no knowledge of this position system. But could it be so simple as to swap the cos and sin wires and then the output would be in the other direction? So no need for different software in the ADIRU?
SIN and COS tend to be a resolver type of angle “encoding”. A very robust passive type of design. You need an excitation AC signal and you get two AC signals back, one with the SIN of your angle and the other with the COS of your angle signal. Typical those signals are differential and isolated from GND, because they come out of a transformer style winding. That makes them robust against noise on the harness. A possible scenario, where you might get wrong or offset values if there is a short between harness lines either the output or the excitation line which is not detected. That might tell us, why exchanging the sensor did not improve the problem.
For Boeing it should be easy to make some tests on the ground to determine what harness failure can create a 22 degree error in their “normal” flight orientation.
For Boeing it should be easy to make some tests on the ground to determine what harness failure can create a 22 degree error in their “normal” flight orientation.
Join Date: Jun 2009
Location: Dorset
Posts: 31
Likes: 0
Received 0 Likes
on
0 Posts
A correction table would also explain the different offsets. I would expect the table to be a 'look-up' table using air speed as the selector. As the airspeed went up it would reach a point where the next value in the table has to be used, so there would be a step change in the correction. So, I would expect the first value in the correction table to be about 6 deg (to give a 12 deg offset), the next value to be about 8 deg and the final value to be about 11 deg.
Join Date: Dec 2014
Location: USA
Posts: 41
Likes: 0
Received 0 Likes
on
0 Posts
Join Date: Mar 2006
Location: Vance, Belgium
Age: 61
Posts: 262
Likes: 0
Received 3 Likes
on
2 Posts
VicMel, what you write does not match the FDR data on AOA values that were shown in the interim report of the Lion Air accident.
It doesn't fit how SIN/COS resolvers work either.
There is a good analysis of what type of resolver, mechanical or electronic failure could match such an almost constant 22° offset on the site of Peter Lemme.
It is worth a read. It's here:
https://www.satcom.guru/2018/12/angl...ure-modes.html
I made a quick check of what would produce a constant 34% positive parasite signal on the sinus entry combined with a 16% negative parasite signal on the cosinus entry.
It gives the series (in degrees)
Real Angle - Reported Angle - Delta
-2 19.97512 21.97512
-1 21.00953 22.00953
0 22.03623 22.03623
1 23.05537 22.05537
2 24.06711 22.06711
3 25.0716 22.0716
4 26.06899 22.06899
5 27.05944 22.05944
6 28.04309 22.04309
7 29.02008 22.02008
8 29.99056 21.99056
I have no idea on what could cause such parasite signals.
Could be inductive from a 400Hz current (like the excitation current), or could be a DC parasite signal in the entry of the analog to digital converter.
Don't misunderstand me ; I do not state that this is what caused the AOA failure.
Just looking at failures that give similar symptoms.
It doesn't fit how SIN/COS resolvers work either.
There is a good analysis of what type of resolver, mechanical or electronic failure could match such an almost constant 22° offset on the site of Peter Lemme.
It is worth a read. It's here:
https://www.satcom.guru/2018/12/angl...ure-modes.html
I made a quick check of what would produce a constant 34% positive parasite signal on the sinus entry combined with a 16% negative parasite signal on the cosinus entry.
It gives the series (in degrees)
Real Angle - Reported Angle - Delta
-2 19.97512 21.97512
-1 21.00953 22.00953
0 22.03623 22.03623
1 23.05537 22.05537
2 24.06711 22.06711
3 25.0716 22.0716
4 26.06899 22.06899
5 27.05944 22.05944
6 28.04309 22.04309
7 29.02008 22.02008
8 29.99056 21.99056
I have no idea on what could cause such parasite signals.
Could be inductive from a 400Hz current (like the excitation current), or could be a DC parasite signal in the entry of the analog to digital converter.
Don't misunderstand me ; I do not state that this is what caused the AOA failure.
Just looking at failures that give similar symptoms.
Last edited by Luc Lion; 29th Mar 2019 at 15:57.
Join Date: May 2010
Location: Boston
Age: 73
Posts: 443
Likes: 0
Received 0 Likes
on
0 Posts
In which case the software in the ADRIUs *has* to be different, one side has to add the correction, the other has to subtract it. I can well believe from bitter experience that the software team did not know enough about the hardware subtlety to appreciate that the ADIRU software has to have a L and a R version. This would explain why testing and peer reviewing did not spot anything wrong, the software met its spec!
A correction table would also explain the different offsets. I would expect the table to be a 'look-up' table using air speed as the selector. As the airspeed went up it would reach a point where the next value in the table has to be used, so there would be a step change in the correction. So, I would expect the first value in the correction table to be about 6 deg (to give a 12 deg offset), the next value to be about 8 deg and the final value to be about 11 deg.
A correction table would also explain the different offsets. I would expect the table to be a 'look-up' table using air speed as the selector. As the airspeed went up it would reach a point where the next value in the table has to be used, so there would be a step change in the correction. So, I would expect the first value in the correction table to be about 6 deg (to give a 12 deg offset), the next value to be about 8 deg and the final value to be about 11 deg.
I agree that a 'stuck bit' does not seem to fit the data since there would be discontinuities as the data crossed the point where the stuck bit should change, especially true since there are 2 A to D converters, one each for SIN and COS. Easily ruled out if we had access to the raw data.
Mention has been made of the AOA analog signals being connected to 2 boxes, this opens a possibility that an electrical fault in one box could cause an offset for both.
Whilst all of you Tech ‘bit’ people provide valuable information and possible scenarios, could you please consider why ‘failures’ appear to be very rare and so far only relate to two aircraft / three vanes.
How something fails does not necessarily explain why (when) it failed.
Random, probabilistic, bit count, world clock ?
How something fails does not necessarily explain why (when) it failed.
Random, probabilistic, bit count, world clock ?
Join Date: Mar 2009
Location: Poway, CA
Age: 72
Posts: 33
Likes: 0
Received 0 Likes
on
0 Posts
I see the WSJ information as a little misleading. While it apparently states that "...the MCAS system activated prior to its dive..." it can not be assumed that it was the cause of an initial upset.
But both of these aircraft were a few months old...a harness problem should have manifested iteself on the first flight, no?
Last edited by Australopithecus; 29th Mar 2019 at 21:55. Reason: spelling
Join Date: Mar 2001
Location: North Cornwall
Age: 73
Posts: 428
Likes: 0
Received 0 Likes
on
0 Posts
An interesting article from Vox.com https://www.vox.com/business-and-fin...ndal-explained
Join Date: Feb 2017
Location: Missouri
Posts: 37
Likes: 0
Received 0 Likes
on
0 Posts
It IS quite interesting that two planes may have had similar problems after just a few months of operation. Certainly a mystery, and Boeing and several CAA's are REALLY wanting to know what the heck happened!
I'd love to know more about exactly how these sensors work, and how the signals get into digital form in the computers.
It is looking like they MIGHT just be bare resolvers in the sensor.
Jon
Join Date: Feb 2018
Location: Canberra
Posts: 1
Likes: 0
Received 0 Likes
on
0 Posts