The problem here is: in the transition from the -NG to the -MAX, Boeing added this protection and didn’t tell anyone.
Imagine someone added Adaptive Cruise Control to your existing car without telling you - the first time the car braked on its own, you’d freak out. The car is suddenly behaving in a way it should be able to.
If you know the system exists, you can recognise what’s happening and deal with it. If you don’t know it exists, the behaviour is going to be absolutely baffling, and no in-flight diagnostic procedure has a step for “did the manufacturer add an important safety device and not tell me?”
https://en.m.wikipedia.org/wiki/Scandinavian_Airlines_Flight...
https://www.havkom.se/assets/reports/English/C1993_57e_Gottr...
In fact, as a general rule of thumb, I believe any safety feature that would automatically change the state of the aircraft should be well educated to the pilots.
Totally out of my depth here, but how is it possible that they "didn't tell anyone" about a feature so crucial, how could pilots not know how the plane's computer will behave in such a fundamental, non edge-case scenario?
The FAA signed off. I guess the lawyers will argue if it was legal or not.
Here's some other examples of what happens when the stabilizer is trimmed without the pilot knowing. Sadly Capt. VanderBurgh died a couple years ago. Had he not it would have been fascinating to see his take on the Lion Air wreck.
https://www.youtube.com/watch?v=WfNBmZy1Yuc
And here's a quick overview of how the stabilizer functions in a 737:
Do the stabilizer trim tabs really have more authority than the elevator? That seems odd.
Do you really understand how your trim works? Many do not, and why it matters. Alex Fisher - GAPAN
It's quite an interesting one if you are into that kind of thing.
I couldn't even hazard a guess. Boeing told the Brazilian authorities about it. See the table on Page 18 under MCAS.
http://www.anac.gov.br/assuntos/setor-regulado/profissionais...
For comparison here's the FAA equivalent:
http://fsims.faa.gov/wdocs/fsb/b-737_rev%2014.pdf
Note that MCAS is distinctly absent from the difference tables (also) on page 18.
The FAA signed off on the NG to MAX training as "B" level differences. This means no sim time necessary[1] (although apparently American Airlines doesn't have a MAX sim in the first place). The Brazilians were made aware of MCAS and thought that also qualified as a "B" level difference. The question then is: did the FAA think MCAS constituted a C or D level difference?
1: See the table on page 14 for a list of the FAA definition of different categories of differences
[0] https://www.seattletimes.com/business/boeing-aerospace/faa-e...
I recently got a hell of a shock in my fathers BMW: I was driving with cruise control engaged, depressed the clutch, and when I released it, found the car re-accelerating to match it's previous speed. Every other car I've driven has disengaged cruise when the clutch is engaged, so this took be quite by surprise and could have put me in danger.
What gave me a scare was lowering the cruise speed on the steering wheel by clicking the button a few times, which in fact caused the car to actively break quite hard instead of naturally lowering its speed.
EDIT Now that I think about it, my BMW does not disengage the cruise when you shift up and ... I'm not sure if it does when you shift down.
My BMWs (E46, E39) all have two clutch switch outputs: one for the starter and one for cruise control.
If the clutch switch was inoperative and the cruise control was engaged, the cruise would not know the engine is disengaged from the transmission and would rev the engine up to redline.
I guess a good thing to note about this is that aviation is so close to optimum safety level that it's hard to find unambiguous ("pareto") improvements; you can just find a different tradeoff. The gradient is pretty close to zero.
What if a malicious pilot had been part of the threat model? instead of terrorist-proof cockpit doors, attacker-proof cockpit doors: split the cockpit in half (strong plexiglass), each has their own combination code for their own door. As soon as one pilot starts disobeying ATC, they can choose which pilot gets control of the airplane...
True, the pilots can no longer circle j* during their boring flights...
To the extent the ideas from the peanut gallery are terrible, it is because intellectual property does not allow the peanut gallery to inspect the full airplane designs, schematics etc... I feel confident the community at large (or the relatives and acquaintances of the deceased and their lawyers and or engineers they appoint) could provide better criticisms. But if my aunt had balls she would have been my uncle: intellectual property kills. What is the statistical value of a human life? What is the statistical value of intellectual property?
>No one can foresee all possible threats, or reliably predict which ones will come to pass.
I think that is an open question. But let's pretend you are right for the sake of argument. Improvement does not necessitate foreseeing all problems, excluding failure modes as they are discovered is still improvement, even if they aren't absolute panaceas.
>It's not possible to split cockpits down the middle because some critical controls such as circuit breakers exist only one one side or the other, but are reachable by both pilots. Adding duplicate circuit breakers on both sides isn't practical for space reasons, and would introduce additional failure points.
Space reasons? I'm sure the passengers appreciated their extra millimeter of legroom while crashing into the alps... why not each their own breaker and the circuit only breaks when their breakers agree on breaking? i.e. logical AND. Yes any additional complexity would make the problem space more complex, and should be dealt with using formal verification etc...
>And allowing ATC to block out a pilot from flight controls would itself be a failure point. What if ATC is compromised?
Obviously, the plane only listens to ATC pilot selection when one pilot disagrees (visualize a big red disagree button if you will). As long as none of the pilots has the disagree button engaged the plane ignores ATC. ATC can not initiate selection of pilot. Yes there situations that still admit malicious pilots like: collusion between 2 pilots, collusion between a pilot and an ATC controller. Use things like sortition, and reward ratting out proposals of collusion...
Edit: spelling
1. Prevent any passenger on the flight from crashing the plane
2. Prevent anyone from crashing the plane into a location of their choice
Locking cockpit doors prevent the first one. Locking cockpit doors plus 2-in-cockpit rules (as mandated by the FAA, and not implemented by GermanWings) do a good job of preventing the second.
yeah, in the Airbus A300 just kick the rudders a few times fully left to right (see American Airlines Flight 587) [1]
> 2-in-cockpit rules (as mandated by the FAA, and not implemented by GermanWings)
Just to clarify, GermanWings was not under FAA jurisdiction, so the 2-in-cockpit rule did not apply.
Also, EASA implemented the 2-in-cockpit rule in 2015 after the GermanWings accident [2], but withdrew it in 2016 and replaced it with a more flexible rule [3] after consultation with stakeholders, as "the rule to have 2 persons in the cockpit at all times introduces additional safety and security risks." [4]
As I said somewhere above, there's rarely scope for a pareto improvement in aviation.
[1] https://en.wikipedia.org/wiki/American_Airlines_Flight_587
[2] https://www.easa.europa.eu/newsroom-and-events/news/easa-rec...
[3] https://www.easa.europa.eu/newsroom-and-events/news/minimum-...
[4] https://www.eurocockpit.be/news/end-2-persons-cockpit-rule-s...
Kicking the rudder fully left to right a few times would most likely put any modern airliner outside of its structural limits (and would probably injure lots of passengers even if it didn't). I believe the A300 had very sensitive rudder controls, which made it possible to create extreme loads on the rudder via a sequence of relatively small back-and-forth movements.
The English language is a bit poor on existential and universal quantifiers, which renders the above sentence a bit ambiguous: it could be interpreted as 1) "For every flight and prevention method there will always exist some opportunities for a pilot to crash the plane" (with which I tend to agree) or 2) "For every flight and prevention method, the pilot will at all times be able to crash the plane" (with which I don't agree)
>(How do you prevent an intentional cartwheel on takeoff or landing? There are regions of the flight envelope where 2-3 seconds of pessimal input are unrecoverable. ATC override won't help you there)
I tend to agree.
>So given that, we aren't trying to prevent the pilot/copilot from intentionally crashing the plane
This doesn't follow. We can still limit the attack scenarios to say situations like landing and taking off. It doesn't need to be binary, a reduction in threats is still desriable. For example if Lubitz had been even more antisocial in his destruction by suicide, he could easily have crashed the plane into the surface storage for nuclear waste next to a nuclear power plant. The reactor itself does not house that much nuclear material, while the typical surface storage next to it tends to contain a lot of nuclear waste. He might still kill the passengers on landing and take off, but as crude as it sounds, it's still better than losing the very same passengers and having a nuclear contamination on our hands!
The 2-in-cockpit rule sounds rather weak in my opinion, as it becomes a matter of who is stronger or who strikes first, the element of surprise etc...
From a systems engineering standpoint, everything you have described would add complexity and would likely be a net negative for safety.
I believe FAA of US specifically alerted German government's air travel agency (?) about the need to have another crew member (like flight attendant) to be in the cockpit if one of the two had to leave cockpit to use restroom etc, exactly to avoid this kind of scenario.
But the Germany govt agency apparently thought it was unlikely to happen and ignored the recommendation.
I could be wrong, but pretty sure about reading that bit while reading about the incident...
Article suggests that pilot didn't know about new system nor how to disable it. In case of faulty sensor (Boeing can't stupid enough not to have quorum of sensors, can it?) crew should be able to disable possibly deadly autopilot.
1. YES - for automation 2. YES - for sensible and safe handling of sensor failure 3. YES - for allow human to override auto-pilot
If instruments cannot measure key environmental indicators such as velocity, temperature etc - no amount of automation will save the plane.
Instrument meteorological conditions (IMC) / Instrument Flight Rating (IFR) flights are when the plane is flying through darkness, or through conditions that do not allow for a judgement of the visual elements and therefore pilots can easily make incorrect judgement calls on the position of the plane, leading to a crash.
The pitot tube is a primitive equipment to measure wind velocity and easily can be jammed by ice, insects etc. I think it was the Pitot tube malfunction in this plane that caused the incident.
I believe the issue is that this hidden system (MCAS) relies on AoA data which can, per the above, not be validated by the pilots or the computers. Thus the fear is that the plane will go full nose down for no obvious reason. Granted the emergency AD indicates some secondary indicators that your AoA vanes have gone wonky.
Per the AA email:
> The MCAS function becomes active when the airplane Angle of Attack exceeds a threshold based on airspeed and altitude. Stabilizer incremental commands are limited to 2.5 degrees and are provided at a rate of 0.27 degrees per second. The magnitude of the stabilizer input is lower at high Mach number and greater at low Mach numbers. The function is reset once angle of attack falls below the Angle of Attack threshold or if manual stabilizer commands are provided by the flight crew. If the original elevated AOA condition persists, the MCAS function commands another incremental stabilizer nose down command according to current aircraft Mach number at actuation.
IOW hey the plane might try to kill you and while you're busy trying not to die at 5,000 ft please disable the electronic aids and grab the trim wheels by hand. Noting, of course, that it take the computer ~30 seconds to move the stabilizer from one end of its travel to the other. It'll take a person longer if you're cranking it by hand. This is, of course, all after the pilots have realized what the problem actually is. All of this at five thousand feet where you might not have 30 seconds to respond. I'd suggest that if this scenario is at all close to what transpired those pilots didn't have a chance.
Just to clarify, with two AoA sensors, you can know that there is a problem (if they disagree), but you don't know which one is erroneous.
What I find surprising about this crash is that even if there's an indication of unreliable readings, the automation proceeds to actively do stuff - I thought Boeing philosophy was to hand everything to the pilots in such a case.
> I'd suggest that if this scenario is at all close to what transpired those pilots didn't have a chance.
Yeah, absolutely devastating. In the time they had, how were they supposed to diagnose that error condition (automatic down trim), given that a) it sneakily recurs every now and then, and b) it was not prepared/trained for?
The AoA disagree alert is an optional feature on the 737[1]. My understanding is that the AoA display is optional as well[2] but does not break down the info per vane. I don't know if the gauge and alert are bundled together or available separately. So maybe you can know, maybe not.
1: https://ad.easa.europa.eu/blob/2018-23-51_Emergency.pdf/EAD_...
2: https://cimg2.ibsrv.net/gimg/pprune.org-vbulletin/432x481/e7...
If you watch a cockpit video of an airliner taking off you will usually hear the co-pilot announce "80 knots" and the pilot reply "cross-checked". What they are doing is checking that their air-sensor data agrees (within a reasonable margin) for the most critical information at that stage of flight (since takeoff speed is very important with the modern wing shape on an airliner).
Similarly they have A and B autopilot systems which are driven independently by two AHRS units (except in special cases like during auto-land where both systems are operational).
Which is all to say that I think they likely have two separate AoA sensors. Although, perhaps being an optional element the failure of one doesn't automatically trigger a AHRS disagree message.
Correct, the 737 NG has two separate alpha vanes[1] and I believe the MAX does as well. However the "alpha vanes disagree" alert is a paid option per the emergency AD. Likewise the AoA indicator is a paid option. There is redundancy, but the plane may be configured such that the pilot cannot determine if there is a failure.
Failure of one or both alpha vanes on an NG isn't a good thing, but failure of an alpha vane on a MAX could cause MCAS to essentially try to kill you and without that AoA disagree alert you may not know why because you've never been informed about this system, and at low altitude you likely wouldn't have time to figure out what's going on.
Edit: if that all sounds fucking insane, it is. That's why American and Southwest pilot unions are livid[2].
1: https://www.faa.gov/air_traffic/separation_standards/ase/201...
2: https://www.seattletimes.com/business/boeing-aerospace/u-s-p...
In the 737 Max this probably got exacerbated to the point that it was possible to fly the plane into a stall by sharply increasing power in a high AoA situation (typically in a go-around). This was probably different enough to the 737NG that they felt it necessary to add the MCAS system to prevent having to do, what they considered, excessive differences training in that phase of flight.
> A former Boeing executive, speaking on condition of anonymity because discussion of accident investigations is supposed to be closely held, said that Boeing engineers didn’t introduce the change to the flight-control system arbitrarily.
> He said it was done primarily because the much bigger engines on the MAX changed the aerodynamics of the jet and shifted the conditions under which a stall could happen. That required further stall protection be implemented to certify the jet as safe.
Wow, the information in that Seattle Times article is really damning! Differences training from the 737 NG to the MAX consisted of a one hour iPad session (plus crosswind training because the permissible roll is reduced due to vertical wing tips). Livid indeed.
The series of events that caused the accident are a long story, but power was knocked out to the pilot's controls (where the one stick shaker was installed), but not the copilot's controls (which didn't have a stick shaker due to the selected options). TBH, it's doubtful that the pilots could have recovered in that specific situation, but the chances of success dropped to basically zero when they didn't have a device capable of communicating what was happening to them in time.
Of course, this is obviously different than having no warning system for the type of failure whatsoever (as appears to be the case on the MAX), but it was still a little surprising for me.
>Instrument meteorological conditions (IMC) / Instrument Flight Rating (IFR) flights are when the plane is flying through darkness
In the US flight in darkness is not flight in IMC. Neither does darkness impose instrument flight rules. Recall that IMC is governed by ceiling, proximity to visible moisture, and visibility: Fail one of those criterion and you're in IMC, governed by IFR.
A pilot lacking an instrument rating may fly in pitch black, no moon, (high) overcast over an ocean and still be VFR compliant. Whether it's wise or not is a different issue...
But, that's theoretical. I'm hearing pilots get massively underpaid for the responsibility they have, unless they gain an x amount of flight hours and years of experience so they can do big intercontinental flights (= more flight hours, = more income). It's up to the industry to fix this problem, also to fix their own shortage of pilots. Pay pilots a decent wage when they're starting out on their career. Give them the opportunities to gain the flight hours.
IMHO the long term solution to improving safety is to make the pilots increasingly redundant. With the current generation of technology this is not yet feasible since they are depending on a wide range of technologies dating back decades. Also a lot of the planes flying are decades old designs. This makes them hard to automate and human-computer interaction has not evolved to the point where a computer can take over these tasks and deal with all the critical human interactions that are involved with operating a plane. In principle the problem is solvable, however. It's mostly just a matter of enough sensors and computation redundancy and improving communication technology to get humans out of the loop.
Mostly pilots these days take executive decisions that boil down to religiously following checklists for basically every scenario imaginable and programming the auto pilot to act accordingly. The auto pilot is activated right after take off and typically disabled on final moments before touch down; or in some cases after landing. Or as in this case, in an emergency.
That's not to marginalize their role. Flying an airliner is a two man operation and they tend to be extremely busy dealing with flying complex procedures, routing around weather, ad hoc queues from controllers, cross checking each other, etc. Most of that stuff requires pilots to have good situational awareness. Most of that awareness is created through reading their instruments, communicating on their radio, and looking outside (when weather permits).
All of that could be automated but it would require a complete rethink of how this business works. For example modern planes are basically equipped with multiple redundant computers and fly by wire control (i.e. a computer controls everything). Yet, critical information is passed to these computers via a non digital communications channel involving people trying to exchange crucial bits of information over a badly congested, low quality radio channel with limited range. This is a ridiculously convoluted, error prone, and hopelessly inefficient way of communicating. The only reason it exists is because agreeing and standardizing on something sensible is going to take decades and has taken decades already.
Most of the chatter on the radio is people cross verifying completely routine information. Worse, people on both sides tend to be very overloaded with information and yet lack the mechanisms to share information other than verbally. Controllers can be juggling communication with dozens of planes and pilots are bogged down in a barrage of instructions, complex procedures, and checklists. A lot of the training focuses on teaching both to stay on top of this (this is very hard). A lot of accidents result from their failure to do so. Emergencies are stressful and stress makes all this even harder.
So, a completely computer based system would do away with most of that to reduce task overload for pilots and controllers and ultimately reduce both of them to the role of remote managers that intervene by exception and very rarely. Ultimately such a system would run itself. Military drones are slightly ahead of the curve here.
This is really arguable.
People always go on about how voice radio communication on a shared channel is from the Stone Age, but what’s better, and why, and can you prove it?
I feel like people drastically underestimate the degree to which the current radio system succeeds in providing situational awareness. We are wired to processs and respond to input from human voices extremely quickly and intuitively with lots of nuance, and our visual and tactile systems are pretty well saturated with information while flying.
What’s your superior suggestion? Saying “computers” or “automation” is going to have to address the fact that those things have some issues as well.
Much of the pilot training for an instrument rating is about staying on top of the radios, dealing with the enormous task load, and planning ahead as to what is going to be required next. As soon as you loose the plot there, you are in trouble: you lose situational awareness, controllers get pissed, things get very stressful, mistakes get made, etc. Many private pilots chicken out from ever getting their instrument rating since it is so enormously challenging and intimidating and since maintaining it is such a big commitment as well.
Most of this is for the perfectly valid reason that this extremely limited and unreliable medium of shouting out letters as words over congested and distorted radio channels is literally the only way to communicate with controllers.
The fix is obvious: send messages over authenticated digital channels with some integrity/sanity checks using some well defined protocol. Start requiring this for all commercial aviation and ban planes from critical air space that are not able to communicate that way. Important messages should require acknowledgement from pilots. There should be zero doubt for pilots about the current status of controller provided altimeter settings, weather report, most recent instructions, etc. Likewise controllers should be able to pick up planes the second their radios and transponders are turned on. This should not be optional. There should be no need for transponder codes. There should be zero doubt about which plane is where for the controller,the entire history of that plane, it's current journey, it's pilot,all previous communications, etc. None of that should need to be communicated or verified by voice, ever.
Voice channels should be reserved for emergencies. Controllers should be able to talk to any plane without delay. These channels should have stuff like caller id and other fancy stuff any decent mobile phone has been capable off for decades. There should be no need to ever yell out call signs and hope that the other side got it jotted down correctly. Etc.
This is not particularly hard to design technically and hasn't been for a long time. Sure you'd want a sane design and lots of built in checks and balances and back up systems, etc.
But agreeing that it needs to be designed, and then agreeing on a particular design is what the problem is here. If you doubt there is a problem, consider that essentially every major plane crash gets blamed on "pilot error" and translate that as "they accidentally crashed the plane because they got confused about what to do under the enormous task-load". When shit goes wrong, having a lot of confusion about who said what, who did what, what was the pilot, plane doing, etc. seems standard. There's no need for that, at all. E.g. Most of the stuff on a black box should be stored remotely long before the plane hits the ground.
There’s no question that staying on top of a complex flight has a heavy mental load. But I’m not convinced radio is the problem.
The information has to be conveyed somehow. Having humans at every step is a clear sanity check.
Either you cut humans out of the job of decision making and have computers make decisions, which introduces all sorts of new issues, or you have to give humans information and context.
And as long as humans are in charge, humans talking to other humans is underrated as a means of getting a lot of information across quickly and effectively.
Basically, aviation has a long history of increasing safety by taking tedious complicated tasks away from humans (e.g. dedicated navigators, radio operators, and engineers used to be a thing). This is just the next obvious thing to tackle. I'd also argue pilot stress and task load is the biggest killer in the aviation industry and the single biggest root cause of every crash ever.
Also, I'd argue this is an extra tool. Of course talking to controllers and other pilots in the area should always be an option. Especially, at uncontrolled fields. Wouldn't it be nice if you can skip the formalities of telling over and over again who you are, where you are, etc. because the other side already knows this?
From an instrument proficiency point of view, I'm much more worried about efficiently understanding the information on the approach plate, maintaining situational awareness with respect to the approach path, adhering to minimums, and so on. Missed something that a controller said? "Say again." There's a reason it's aviate, navigate, and communicate--in that order.
You bring up a fine point: aviate, navigate, and communicate - in that order. Very sensible when everything is manual and you are task saturated to drop tasks in that priority; which is what this is about. If you automate communication and remove most of the task load related to that and navigation, it frees you up to let the auto pilot do the aviation bit.
1 - http://fortune.com/2014/03/03/why-airlines-are-running-out-o... 2 - https://daytonabeach.erau.edu/admissions/estimated-costs/
I think this is outdated from the post-2008 crisis time, when Airlines cut back on flights. Currently there is a massive shortage of pilots, some make 150k ~ 300k in the Middle East of China.
I was going to quote parts of it, but it's all quite insightful that, making 150k-300k is not quite how it goes. Take a read.
I remember here in HN, there were cries why auto pilot didn't take over at that time and save lives. "Pilots should never be allowed to stall the plane."
Well, Boeing's new software did exactly that. Correct the situation when the pilot wasn't. And that crashed the plane.
Interesting indeed. Lesson: There's a lot to look beyond before taking a decision, even when the obvious decision is just in front of everyone.