Atlas Air 3591 NTSB Public Docket Opened
Champagne anyone...?
The brake was out as they were descending so I assume that was why he was guarding it. The brakes were manually retracted as G/A was "initiated". Watch the Youtube vid in the post above - about 3:35 onwards.
Ray, interesting charts, #177.
Concentrating on the 'certainly split' region, then this is coincident with change of stab trim.
As you note, which part of the system is driving the trim; which direction - presumably nose down.
Pilot trim switch input, but it is not clear if the AP was disengaged.
Or the AP, auto trim, unlikely re the stick deflection; but if so the direction should be consistent with the selected mode, which was …
Or AP overridden by stick deflection; is this CWS mode, and if so what would the trim be expected to do. Perhaps minimise the stick force as a 'follow up' trim function; but then why does the trim not stop / reverse when the sticks are reversed. The stick 'reversal' is coincident with the split region, where trim reversal might be expected, but with a split, which stick has priority.
Much of the above depends on the stick force felt by the pilot, which is affected by the elevator and trim positions and their relative position in feeding back the feel force.
We don't know what the pilots felt nor how that might have influenced their perception.
Concentrating on the 'certainly split' region, then this is coincident with change of stab trim.
As you note, which part of the system is driving the trim; which direction - presumably nose down.
Pilot trim switch input, but it is not clear if the AP was disengaged.
Or the AP, auto trim, unlikely re the stick deflection; but if so the direction should be consistent with the selected mode, which was …
Or AP overridden by stick deflection; is this CWS mode, and if so what would the trim be expected to do. Perhaps minimise the stick force as a 'follow up' trim function; but then why does the trim not stop / reverse when the sticks are reversed. The stick 'reversal' is coincident with the split region, where trim reversal might be expected, but with a split, which stick has priority.
Much of the above depends on the stick force felt by the pilot, which is affected by the elevator and trim positions and their relative position in feeding back the feel force.
We don't know what the pilots felt nor how that might have influenced their perception.
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
... then eased back for short time before applying more push to -2.5° Elevator
12:38:42 further pushed to -7° Elevator. That's 12/13s after G/A
12:38:46 FO shouts "my speed", that's 8s after the first push
12:38:48 FO shouts "we're stalling", 10s after first push
It's simple, and just needs looking at the report again.
He (or both) was hit by Somatogravic Illusion. Consistent with the thrust increase.
The apparent pitch up was calculated with ~20° at 12:38:42 and 80° at 12:38:46 in the report.
He was not the first commercial pilot to react with pushing down. He did as far as producing 0...-1g
I bet he was partially blind and deaf now, well, overwhelmed. That was especially a problem for him! From then on don't take his calls for real. Did he really believe in a stall?
If you believe you're up to 80°, you can guess your speed is gone very quick
If you feel g-force of 0 or -1, you can think that you're falling
Regarding "speed", it's not 100% ruled out the speed tape on EFIS was affected from another breakdown. NTSB however found it very unlikely. Or he confused overspeed range with something different.
The hearing mentioned some hundereds pilots known to have been affected by somatographic illusion, MOst of them in Gereral Aviation,
The hearing discussed how that effect is dealt with in trainings
Last edited by Ray_Y; 26th Jul 2020 at 17:47.
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
Concentrating on the 'certainly split' region, then this is coincident with change of stab trim.
As you note, which part of the system is driving the trim; which direction - presumably nose down.
Pilot trim switch input, but it is not clear if the AP was disengaged.
Or the AP, auto trim, unlikely re the stick deflection; but if so the direction should be consistent with the selected mode, which was …
Or AP overridden by stick deflection; is this CWS mode, and if so what would the trim be expected to do. Perhaps minimise the stick force as a 'follow up' trim function; but then why does the trim not stop / reverse when the sticks are reversed. The stick 'reversal' is coincident with the split region, where trim reversal might be expected, but with a split, which stick has priority.
We don't know what the pilots felt nor how that might have influenced their perception.
As you note, which part of the system is driving the trim; which direction - presumably nose down.
Pilot trim switch input, but it is not clear if the AP was disengaged.
Or the AP, auto trim, unlikely re the stick deflection; but if so the direction should be consistent with the selected mode, which was …
Or AP overridden by stick deflection; is this CWS mode, and if so what would the trim be expected to do. Perhaps minimise the stick force as a 'follow up' trim function; but then why does the trim not stop / reverse when the sticks are reversed. The stick 'reversal' is coincident with the split region, where trim reversal might be expected, but with a split, which stick has priority.
We don't know what the pilots felt nor how that might have influenced their perception.
After G/A, A/P was in G/A P mode and G/A R mode (pitch and roll)
The final seconds it changed from G/A P mode to ALT and ALT HOLD mode, that was somewhere at 3000ft (guess what was dialed in into MCP)
The FDR data does not mention CWS. I don't know the 767 and its behaviour with manual overriding the G/A mode
TRIM changed from ~5 to ~4 Units, that is towards nose down. I believe it followed colums with some delay. Again, I hav no knowledge of 767.
"Stick force": add the forces once the split between CPT and F/O took place. I don't know if it's just a momentary breakout force or if the split produces permanent forces.
Ok, i will try again in more simple terms....
Why, why, does the speed brake does NOT "stow" when thrust levers advance? Because both conditions does not make sense!
Regarding the seconds and somatographic illusion this person should never be in a flight deck
Why, why, does the speed brake does NOT "stow" when thrust levers advance? Because both conditions does not make sense!
Regarding the seconds and somatographic illusion this person should never be in a flight deck
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
Regarding speed brake I repeat
It was procedure, and in fact they were retracted some time after G/A. This made the scenario very likely, but NTSB can't really know.
Touching the extended speed brakes lever is a good practice not to forget it. Happened too often in the past. Simple as that,
Touching the extended speed brakes lever is a good practice not to forget it. Happened too often in the past. Simple as that,
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
de minimus non curat lex
The data from this tragic event needs to part of the UPRT simulator program and certainty part of any command training and recurrence training yearly.
Prospective employers minds will clearly be concentrated by this tragic event, and improve the Due Diligence process before hiring new pilots.
I would suggest that the process starts from day 1 of flying, even if this is for PPL issue, prior to any thoughts of professional flying.
Last edited by parkfell; 27th Jul 2020 at 10:34. Reason: modified & extended final paragraph
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
I think investigation is completed. And then there's this remark at "findings" ... nothing of the following were factors ... training and certification of crew (from memory)
And no safety reccom to address the issue of somat. illus.by training, simulation, right?
Rather a device/automation that shall prevent such CFIT in future, as it seems to work in military aircraft.
And no safety reccom to address the issue of somat. illus.by training, simulation, right?
Rather a device/automation that shall prevent such CFIT in future, as it seems to work in military aircraft.
Ray, et al,
There is no need to invoke an illusion to account for the pilot's nose-down action.
Consider the inadvertent GA mode, not recognised; aircraft starts to pitch up and thrust increases.
The pilot's objective is to continue the descent; the pilot 'naturaly' pushes forward on the stick (opposing the GA pitchup stick force), forgetting that the AP is still engaged; i.e. overpowers the AP without disengaging.
An unknown factor is the mechanism of AP / auto trim movement in this condition. Either:
- Overpowering a normal pitch mode, the trim could react to oppose the pilot stick input, but not necessarily for GA mode.
- If the 767 (this version) enters CWS mode with overpower, then trim follow-up action is more likely; the trim movement and direction aids the pilots stick input.
Thus the erroneous nose down stick input could have been a subconscious, automatic, or startle response to the unseen GA mode, which was then aided by nose down trim.
Does the 767 have a CWS mode, is this activated by overpowering the AP ?
What would the trim do when overpowering the AP, particularly in CWS mode if fitted and activated ?
'Rather a device/automation that shall prevent such CFIT in future, as it seems to work in military aircraft.'
Already tested as a proposal for EGPWS auto pull up for obstacle / unauthorised location (after 9/11); rejected due to civil certification requirements. Also that AP 'muscle' is usually much less than the human input - pilots can override the AP - back to square 1; what happens if the AP is owerpowered.
There is no need to invoke an illusion to account for the pilot's nose-down action.
Consider the inadvertent GA mode, not recognised; aircraft starts to pitch up and thrust increases.
The pilot's objective is to continue the descent; the pilot 'naturaly' pushes forward on the stick (opposing the GA pitchup stick force), forgetting that the AP is still engaged; i.e. overpowers the AP without disengaging.
An unknown factor is the mechanism of AP / auto trim movement in this condition. Either:
- Overpowering a normal pitch mode, the trim could react to oppose the pilot stick input, but not necessarily for GA mode.
- If the 767 (this version) enters CWS mode with overpower, then trim follow-up action is more likely; the trim movement and direction aids the pilots stick input.
Thus the erroneous nose down stick input could have been a subconscious, automatic, or startle response to the unseen GA mode, which was then aided by nose down trim.
Does the 767 have a CWS mode, is this activated by overpowering the AP ?
What would the trim do when overpowering the AP, particularly in CWS mode if fitted and activated ?
'Rather a device/automation that shall prevent such CFIT in future, as it seems to work in military aircraft.'
Already tested as a proposal for EGPWS auto pull up for obstacle / unauthorised location (after 9/11); rejected due to civil certification requirements. Also that AP 'muscle' is usually much less than the human input - pilots can override the AP - back to square 1; what happens if the AP is owerpowered.
Join Date: Jan 2006
Location: US
Posts: 2,205
Likes: 0
Received 0 Likes
on
0 Posts
Why don't they? No idea. We'd have to be back to the later 1970's or early 1980's to ask why it was designed and certified that way. Perhaps it was "one step too far" for the parties involved in transitioning to the modern glass cockpits? Now we accept glass cockpits but there were significant trainings issues training pilots coming off of 'round dial gauges' to 'glass' /FMC cockpits.
Join Date: Jan 2006
Location: US
Posts: 2,205
Likes: 0
Received 0 Likes
on
0 Posts
Deploying squadron overseas had the last departure 'lose his radios' and returned to base. Low IMC using a/c radar in trail departure procedures. So the other aircraft continued. On landing they were told that the last guy had crashed. Investigation was vertigo and pushed over...staying in burner to rejoin quicker? Distracted by radar? Somatogravic illusion.
Vertigo is a real threat if fighters, even in VMC. I think the F-15 record is 12 G's pulling out of a dive after getting vertigo while dog fighting in VMC. Late 1980's out of Langley AFB. A funny moment in his video briefing on what happened - F-4 flashes across the screen. Based on his perception on his attitude the F-4 would be climbing vertically - "if I'd seen that I would have realized I had vertigo. F-4's don't go straight up like that." That always had the viewers laugh...except the F-4 guys?
"
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
4 submissions of investigation parties published in the NTSB dockets today
dms.ntsb.gov Public docket
dms.ntsb.gov Public docket
Last edited by Ray_Y; 27th Jul 2020 at 18:57.
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
Quoted from Party Submissions
There you have some answers! One of the last 4 files
INTERNATIONAL BROTHERHOOD OF TEAMSTERS Submission:
(...)
1. Autopilot Does Not Disconnect
Discussions with Boeing’s Air Safety Investigation team revealed that the autopilot system is designed such that control column input force that overrides a single channel autopilot servo on the 767-300 will not result in autopilot disconnect, but rather an illumination of the autopilot caution light. A review of the Boeing 767 manual reveals no notes or cautions to this effect. However, in the simulator training environment the built-in protections of simulator control loading will disconnect the autopilot and release the control columns when excessive force is applied during a maneuver or malfunction of the simulator control loading.
The IBT investigators believe this difference has the potential for a ‘negative learning’ transfer from the simulator training environment to actual line flying. While it did not appear to have a direct impact on the accident, the Boeing 767 manuals should be corrected to highlight this system architecture difference for all operators.
2. EFI Switch and Symbol Generator Malfunction
Investigators have been unable to determine the exact failure scenario that required the rapid transfer of controls back and forth between the First Officer and the Captain. Based on a review of the CVR, it appears there may have been a right symbol generator failure of the First Officer’s EFIS59 screens. Selecting the EFI switch to ALTN60 mode would have sourced the center symbol generators. Since this EFI switching event occurred so close to the final upset event, it is possible that the First Officer may have mistrusted his EFIS screen readings and relied instead on sensory input as to aircraft attitude. This EFI/symbol generator event should not have affected the round-dial (analog) airspeed indicators on either pilot’s instrument panel or the backup ISFD in the center left panel for airspeed/attitude crosschecking purposes.
(...)
EDIT: And read their recommendations.
INTERNATIONAL BROTHERHOOD OF TEAMSTERS Submission:
(...)
1. Autopilot Does Not Disconnect
Discussions with Boeing’s Air Safety Investigation team revealed that the autopilot system is designed such that control column input force that overrides a single channel autopilot servo on the 767-300 will not result in autopilot disconnect, but rather an illumination of the autopilot caution light. A review of the Boeing 767 manual reveals no notes or cautions to this effect. However, in the simulator training environment the built-in protections of simulator control loading will disconnect the autopilot and release the control columns when excessive force is applied during a maneuver or malfunction of the simulator control loading.
The IBT investigators believe this difference has the potential for a ‘negative learning’ transfer from the simulator training environment to actual line flying. While it did not appear to have a direct impact on the accident, the Boeing 767 manuals should be corrected to highlight this system architecture difference for all operators.
2. EFI Switch and Symbol Generator Malfunction
Investigators have been unable to determine the exact failure scenario that required the rapid transfer of controls back and forth between the First Officer and the Captain. Based on a review of the CVR, it appears there may have been a right symbol generator failure of the First Officer’s EFIS59 screens. Selecting the EFI switch to ALTN60 mode would have sourced the center symbol generators. Since this EFI switching event occurred so close to the final upset event, it is possible that the First Officer may have mistrusted his EFIS screen readings and relied instead on sensory input as to aircraft attitude. This EFI/symbol generator event should not have affected the round-dial (analog) airspeed indicators on either pilot’s instrument panel or the backup ISFD in the center left panel for airspeed/attitude crosschecking purposes.
(...)
EDIT: And read their recommendations.
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
Quoted from Party Submissions
ATLAS AIR Submission:
1.2.2.1 Atlas Air Stall Recovery Guidance
...
1.3.2 Personnel Info > F/O
Employment History
...
3.3 Safety Recommendations
1. To the FAA: Require air carriers to implement ground and simulator training to recognize spatial disorientation/somatogravic illusion induced by sudden unexpected acceleration.
2. To Boeing: Make the activation of the go-around mode more conspicuous to the crew, including an aural alert.
3. To the FAA: Require air carriers to implement training programs related to recognizing inadvertent activation of go-around mode.
4. To the FAA: Implement Congress’s 2010 mandate to establish an electronic database for pilot records.
5. Reiterate, to the FAA: Develop a process for verifying, validating, auditing, and amending pilot training records at 14 Code of Federal Regulations Part 121, 135, and 91K operators to guarantee the accuracy and completeness of the records. (A10-20).
And many more to read
1.2.2.1 Atlas Air Stall Recovery Guidance
...
1.3.2 Personnel Info > F/O
Employment History
...
3.3 Safety Recommendations
1. To the FAA: Require air carriers to implement ground and simulator training to recognize spatial disorientation/somatogravic illusion induced by sudden unexpected acceleration.
2. To Boeing: Make the activation of the go-around mode more conspicuous to the crew, including an aural alert.
3. To the FAA: Require air carriers to implement training programs related to recognizing inadvertent activation of go-around mode.
4. To the FAA: Implement Congress’s 2010 mandate to establish an electronic database for pilot records.
5. Reiterate, to the FAA: Develop a process for verifying, validating, auditing, and amending pilot training records at 14 Code of Federal Regulations Part 121, 135, and 91K operators to guarantee the accuracy and completeness of the records. (A10-20).
And many more to read
Testpanel:
Airbus FBW speed brakes DO automatically retract : if the thrust levers are set above MCT. (TOGA gate is above MCT gate, obviously).
From A320 FCOM: (my bold)
Speedbrake extension is inhibited, if:
In my opinion Airbus significantly improved aircraft design in many areas, including removing the 'external' TOGA switches to avoid accidental operation and re-siting the speed brake and flap levers away from the thrust levers.
Why, why, does the speed brake does NOT "stow" when thrust levers advance?
AFAIR Airbus' speed brakes don't retract either. Instead it gives you and ECAM alert.
Why don't they? No idea. We'd have to be back to the later 1970's or early 1980's to ask why it was designed and certified that way. Perhaps it was "one step too far" for the parties involved in transitioning to the modern glass cockpits? Now we accept glass cockpits but there were significant trainings issues training pilots coming off of 'round dial gauges' to 'glass' /FMC cockpits.
Why don't they? No idea. We'd have to be back to the later 1970's or early 1980's to ask why it was designed and certified that way. Perhaps it was "one step too far" for the parties involved in transitioning to the modern glass cockpits? Now we accept glass cockpits but there were significant trainings issues training pilots coming off of 'round dial gauges' to 'glass' /FMC cockpits.
From A320 FCOM: (my bold)
Speedbrake extension is inhibited, if:
- SEC1 and SEC3 both have faults, or
- An elevator (L or R) has a fault, or
- Angle-of-attack protection is active, or
- Flaps are in configuration 3 or FULL, or
- Thrust levers above MCT position, or
- Alpha Floor activation.
In my opinion Airbus significantly improved aircraft design in many areas, including removing the 'external' TOGA switches to avoid accidental operation and re-siting the speed brake and flap levers away from the thrust levers.
Last edited by Uplinker; 28th Jul 2020 at 09:52.
… the simulator does not replicate this hazard
Ray, #193, #194
Very informative links and a great contribution in identifying explanatory text; thank you. From one set of answers there are more questions.
"Overpowering aircraft auto-flight system does not disconnect the autopilot; this is incorrectly represented in the simulator."
Your question #177 about trim activation and direction of stab movement remains open.
From your chart, trim movement is correlated with stick displacement ('certainly split' region), possibly because auto-trim runs to reduce stick force as it would to reduce autopilot servo load (because the autopilot/auto-trim is still engaged)
If so the nose down pitch change from the erroneous forward stick would be aided by stab trim such that the recovery nose up pull is most disadvantaged by the need for more force to pull up *. Subsequently, stab trim tracks the nose up stick input, but too late.
The remaining question is why the pilot pushed forward at the time. The Boeing view suggest mis-assessed stall condition, my previous - the mindset of intended descent, or other views of illusion; we don't and cannot know for sure.
However, what we can learn is if the 767 AP is overpowered, the the auto-trim moving the stab to reduce stick force to help the pilot can be a disadvantage if the control direction has to be reversed.
Also that the simulator does not replicate this hazard.
* shadow of 737 MAX
Very informative links and a great contribution in identifying explanatory text; thank you. From one set of answers there are more questions.
"Overpowering aircraft auto-flight system does not disconnect the autopilot; this is incorrectly represented in the simulator."
Your question #177 about trim activation and direction of stab movement remains open.
From your chart, trim movement is correlated with stick displacement ('certainly split' region), possibly because auto-trim runs to reduce stick force as it would to reduce autopilot servo load (because the autopilot/auto-trim is still engaged)
If so the nose down pitch change from the erroneous forward stick would be aided by stab trim such that the recovery nose up pull is most disadvantaged by the need for more force to pull up *. Subsequently, stab trim tracks the nose up stick input, but too late.
The remaining question is why the pilot pushed forward at the time. The Boeing view suggest mis-assessed stall condition, my previous - the mindset of intended descent, or other views of illusion; we don't and cannot know for sure.
However, what we can learn is if the 767 AP is overpowered, the the auto-trim moving the stab to reduce stick force to help the pilot can be a disadvantage if the control direction has to be reversed.
Also that the simulator does not replicate this hazard.
* shadow of 737 MAX
Only half a speed-brake
Yes, I get it that once trimmed down it is harder to pull up, but that make-believe similarity above - - - like seriously bro' WTF? [/grumpy .... exits the stage via the trap-door
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
Originally Posted by alf5071h
...what we can learn is if the 767 AP is overpowered, the the auto-trim moving the stab to reduce stick force to help the pilot...
I asked myself if this is a sort of CWS function in the 76, a leftover from former Boeing aircraft. Anyway, if you deflect Column to pitch, as A/P would do, the trim will follow with a delay, as A/P would do, well, rather that's what a still active A/P does, right?
I see a 5 second delay in the FDR, when trim starts to follow control input. And yes, when trim followed nose-down then ...
Originally Posted by alf5071h
...the recovery nose up pull is most disadvantaged by the need for more force to pull up
Ethopian suffered from large Trim down, combined with increasing airspeed. The result was eventual complete loss of pilot elevator authority, impossible to counter trim.
This ATLAS flight had eventually added 1 Trim Unit nose down, also combined with increasing airspeed. Had it not been G/A with trim up before, they'd ended with like +3.8 units. BUT: at the end they were able to pull +4g. So loss of control authority wasn't a factor.
Join Date: Jun 2020
Location: Europe
Posts: 22
Likes: 0
Received 0 Likes
on
0 Posts
There's so much to read in that, I recommend to anybody interested. The other one is Atlas Air Submission. GE had nothing to suggest, and Boeing 's answer is underwhelming.
After reading few of these party's answers, I wonder if NTSB will also pickup these safety improvements:
Add the problem of Somatographic Illusion into training, simulation, enhance simulators fidelity to this (I think it's time!)
Install 767-400 Throttles into older types, those having TOGA switches moved away
Add G/A aural clues (I don't like that one really, more clutter)
Prevent G/A by more logic (With different throttles this is unnecessary IMHO)
After reading few of these party's answers, I wonder if NTSB will also pickup these safety improvements:
Add the problem of Somatographic Illusion into training, simulation, enhance simulators fidelity to this (I think it's time!)
Install 767-400 Throttles into older types, those having TOGA switches moved away
Add G/A aural clues (I don't like that one really, more clutter)
Prevent G/A by more logic (With different throttles this is unnecessary IMHO)