Comment by myth2018 | Hacker News Reader

Comment by myth2018 | Hacker News Reader

Parent Full thread

Mass media in general is stressing too much the fact that 737 MAX is allegedly based on dated designs.

I think it's fair to say that MAX versions present some compromise solutions (like the now-infamous MCAS, which is there to compensate for the "unnatural" bigger engines). But I think that is not the main point. They would be good solutions it they worked as intended.

There are some other more fundamental and more daunting, afaik unanswered questions.

Like, why does such a critical system like MCAS take only a single AoA sensor as input, when there are two sensors available? Specially considering that the inputs from both are hardware-available to MCAS (the new software version is going to take data from both).

Boeing affirmed in its manuals that the elevators would be able to compensate for the trimmed vertical stabilizers. Now the preliminary report in the Ethiopian's crash shows that the pilots wheren't able to perform such compensation, even by pulling the control columns all the way back.

Those and some other issues are much more critical.

>Like, why does such a critical system like MCAS take only a single AoA sensor as input, when there are two sensors available? Specially considering that the inputs from both are hardware-available to MCAS (the new software version is going to take data from both).

Are you saying the answer in the article didn't give enough information?

"Airbus addressed this potential problem on some of its planes by installing three or more such sensors. Former Max engineers, including one who worked on the sensors, said adding a third sensor to the Max was a nonstarter. Previous 737s, they said, had used two and managers wanted to limit changes.

The angle of attack sensor, bottom, on a Boeing 737 Max 8.CreditRuth Fremson/The New York Times “They wanted to A, save money and B, to minimize the certification and flight-test costs,” said Mike Renzelmann, an engineer who worked on the Max’s flight controls. “Any changes are going to require recertification.” Mr. Renzelmann was not involved in discussions about the sensors."

No it doesn’t answer the question. If there are two sensors already installed, not requiring change, how come the MCAS only uses one, not both?

Instrumentation on an aircraft is usually designed with a pilot and copilot set. Everything down to the pitot tubes which feed the information is unique to that side of the aircraft. This harks back to the day when it was the only sensible solution when your guages are actually just directly reading airpressure differences.

Clearly these signals can be cross connected, because that's the solution Boeing are testing at the moment, but it's outside the normal design of aircraft systems.

Uhm.. I don't know.. I'm aware this is the case of airspeed indicators and many others, but, for instance, the autopilot is fed with readings of Pitot tubes from both sides and it disengages when the sensors disagree beyond sensible thresholds. Besides, such issue is brought clearly to the knowledge of pilots.

In the MCAS case, however, we are speaking about a computer which not only interferes in the flight controls, but also does that in a way impossible to override and it's too difficult for the pilots to spot the root cause.

To be clear, I think MCAS was a massive engineering failure for Boeing, I'm just pointing out why I think they didn't automatically say "hey, we've got two sensors, lets compare the readings before we act on it".

However, for the autopilot example, I don't think you are quite right. Typically there are multiple autopilots (2 or 3 is quite common), and each is driven by a different set of flight data. In some scenarios multiple autopilots are engaged at same time, for example during CATIII auto-landings but as far as I'm aware that is the exception rather than the rule.

It's true that the autopilot may disconnect if there is a warning like 'IAS Disagree', I suspect this is driven by a separate monitoring process though, rather than being an integral part of the system.

The article also states that 2 sensors is not enough as then you have two sensors that can disagree with no way to figure out the correct one.

Yes but at least you know the reading is faulty and avoid applying dangerous commands. Now I don't know if the stall it is meant to avoid is a greater risk than MCAS pushing the plane in the wrong direction.

Stalls can be extremely difficult to recover from as it means the wings have lost lift, and therefore the control surfaces (which you need to regain stability) have reduced or even complete loss of effect - a so-called 'deep stall'. As bad as MCAS is, it could theoretically (in practise, couldn't) be switched off in this scenario and the plane would be flyable. In the imagined scenario where the plane pitched up and began to stall, Beoing's logic is that without MCAS, the plane would be essentially doomed. Air France 447 crashed due to a (pilot-induced) deep stall; it was otherwise stable at cruising altitude.

The root cause is without doubt relying on a single sensor, and then downplaying the importance of the system so that nobody opted for the additional expense of the extra sensor. Boeing also have to answer for their lack of transparency; their flight control logic has always left the pilot fully in control of the plane, and can override any automatic system. This sets them apart from Airbus, which under almost all circumstances will defer to the computer.

In ways, the 737 MAX crashes are the antithesis of the 447 crash - the pilots thought they were in full command of the plane, whereas an automatic system designed to protect them malfunctioned, versus the pilots in the Air France plane believed the computer would protect them from exceeding the plane's capabilities, whereas the plane's computers could not get reliable data and passed full control to the pilots.

The analysis is not as simple as asking whether a stall is more or less dangerous than an MCAS failure. Firstly, MCAS does not prevent a stall; it is intended to make it harder to accidentally stall (and no sane pilot would deliberately stall an airliner in normal operations), in order to compensate for the design change that made it easier to do so. When considering alternatives such as whether to disable MCAS on a sensor discrepancy, one should ask both how likely each possible scenario in each alternative is, and how much risk it adds.

Where the risk analysis seems to have gone most wrong is that Boeing apparently grossly underestimated the difficulty of both figuring out what actions were needed to respond to the symptoms of MCAS failure, and to perform them. I don't know whether it was a significant factor in the former, but when the AofA sensor failed, it caused the stick shaker, as well as MCAS, to kick in.

The other mistake in analysis seems to be that when the power of MCAS was increased after initial flight testing, the additional risk it created was not properly taken into account. In particular, the ability of MCAS to drive the trim all the way forward appears to have been an unintended and overlooked side-effect of one design change.

It doesn't answer the question because it's still a speculation.

Full investigation is hopefully going to tell us what has really happened.

>Like, why does such a critical system like MCAS take only a single AoA sensor as input

The classic approach is to have three sensors, so in case one fails you can know which one. Having two only indicates something is wrong but is not useful on the fly.

Of course, even with triple-redundant systems failures can still occur.

Air France 447 [1] three independent air data systems, two of them failed due to environmental conditions

XL Airways 888T [2] three independent AOA sensors, two failed because the plane was washed without the right covers in place

US Airways 1549 [3] two independent engines, both disabled by bird strike at the same time (No fatalities)

Qantas Flight 72 [4] three independent inertial reference units, bug in voting system if a single sensor's output had multiple spikes 1.2 seconds apart (no fatalities)

An in the data centre, no amount of power-supply redundancy will save you if a technician pulls out the power cables on the wrong server :)

[1] https://en.wikipedia.org/wiki/Air_France_Flight_447#cite_ref... [2] https://en.wikipedia.org/wiki/XL_Airways_Germany_Flight_888T [3] https://en.wikipedia.org/wiki/US_Airways_Flight_1549 [4] https://en.wikipedia.org/wiki/Qantas_Flight_72

Three with a disagree algorithm is definitely what I’d expect out of such a critical system, but two with signal averaging would still be much better than just one.

Or alternatively with two, and disabling MCAS if they disagree, seems a better solution than having one and having no way to tell if it is working (keeping in mind both can still fail simultaneously). Not an ideal solution but better.

This is one of the features of the MCAS software update.

Unless I am misunderstanding what signal averaging is (quite possible) isn't it possible that in situations where the average of a signal is still going to crash the plane, a 50/50 guess is actually more likely to end up with a better chance?

If true, it's possible signal averaging isnt necessarily the best choice

Averaging would be disastrous in this case. You don't want to use faulty data to average inputs to a flight control computer.

> The classic approach is to have three sensors

Which can also fail:

In 2008, on a customer-acceptance flight of an Airbus A320, two of the angle-of-attack sensors froze and those two sensors then outvoted the third. When the pilots went to demonstrate the stall-prevention system, they were not aware of the malfunctioning sensors. The plane crashed, killing the seven people on board.

The same problem arose again on a 2014 Airbus A321 Lufthansa flight leaving Spain. Eight minutes after takeoff, two of the angle-of-attack sensors froze at the same pitch. This time, after a drop in altitude, the pilots were able to regain control and complete the flight. [1]

I don't think the fundamental problem with MCAS was the number of sensors, but that it was too difficult for the pilots to override MCAS when it faulted.

1. https://www.seattletimes.com/business/boeing-aerospace/a-lac...

Boeing has been using only two for quite a time. Having a failed and a working one would simply indicate that something is wrong, but this information is valuable anyway -- it could be used to prevent MCAS from engaging based on wrong data, exactly one of the features the new software update is bringing.

Besides, those AoA sensors are EXTREMELY reliable. So reliable that some have raised the hypothesis that the real problem is not in the sensors themselves, but in some piece of hardware or software between them and the flight computers.

It seems plausible to me since failures in those sensors are too rare in the other planes but, despite that, they allegedly failed in two 737 Max 8s and in a really short timespan.

They designed it precisely like this because it's a dated design. To use two sensors (or three) would have required new cockpit components and new pilot training, which is what they were trying to avoid by reusing the same Type certificate originally from 1968.

Reply on news.ycombinator.com