Posted on 03/18/2019 1:14:55 PM PDT by SeekAndFind
I do not trust AI in aviation. Because one cannot know how the AI is making its decisions, one cannot predict where the AI’s rules will cause their own mischief.
EVERYTHING is speculation.. until PROVEN!
Even the ancient variants had comparators for the auto pilot systems. I agree a hard trim input system should be based on two AOA guages.
JLAGRAYFOX
“Why did the First Officer (Co-Pilot) on the Ethiopian Airlines crashed B737MAX aircraft have only 200 active flight operating flying hours on the B737MAX aircraft.”BatGuano
When I flew my first trip on the B-737, my experience was about 30 hours in a 737 simulator and 25 hours in the airplane with a line check pilot. 200 hours doesn’t seem like much but it could take over two and a half months to accumulate. More important would be the background and type of flight education and experience prior to being hired by Ethiopian. My experience, Military trained, fighter and multi-engine “heavy” experience over a 6 year period of time, was more than this Co-pilot may have had. Avoid third world airlines.
The first officer didn't have 200 hours in the 737 Max. From what I've read, he had 200 total flight hours in any kind of aircraft. Chesley Sullenberger was shocked that that a pilot that inexperienced would be flying a jetliner.
CAPTAIN SULLY WEIGHS IN ON BOEING 737 MAX CRASH
MARCH 16, 2019
https://liveandletsfly.boardingarea.com/2019/03/16/sullenberger-737-crash/
However, I am also astounded that this system and the AFM were approved as is.
As I understand from earlier reading, There were two AoA gauges, but the one used depended on which pilot was flying at a given time. That means no redundancy. First, the AoA gauges need to have self diagnosis capabilities. A faulty sensor should not be controlling the actions of anything. A sick pilot is able to at least diagnose that he is sick and needs to see a doctor and not fly a plane. The gauges have difficulty knowing when they are malfunctioning. They should at least know when to not have confidence in their measurements and abstain from voting.
I watched an episode of Air Disasters on Sunday about QANTAS flight 72 in 2008 where there was a software error in the code for the device sending data to the fly by wire systems. It was reporting altitude as angle of attack. It was showing large quick changes in angle of attack, and commanded the plane into a negative G dive. The captain was able to get control again, but it tried to dive again. Thep pilots finally got control and diverted to a closer airport to make an emergency landing. The captain was a former US Navy pilot who used his experience as a fighter pilot to do a high speed landing in order to be able to recover in case the plane started going nose down close to the ground.
Malfunctioning AoA sensors causing dangerous commands to be sent by fly by wire systems is not an unprecedented issue.
What kind of “intelligence” cannot articulate the relationships it detects and teach them to humans in English?The ability to do that is the marker for when AI has broken free of its “artificial” context and simply become intelligence.
What kind of “intelligence” cannot articulate the relationships it detects and teach them to humans in English?The ability to do that is the marker for when AI has broken free of its “artificial” context and simply become intelligence.
In fact, intelligence can learn from articulated teaching in English, learn more (including the limits and errors of the aforementioned articulated learning from English) from experience, and teach in English. And also learn from the responses of those it teaches.
No human, after all, could earn a PhD without all of those abilities . . .
My favorite anecdote about AI:
The Army was using AI to detect camouflaged tanks. They used satellite imagery from Germany to train the neural network. The AI got to 95% effective.
In Desert Storm, they fed similar imagery from Iraq, and the AI completely failed.
The AI was counting leaves in the German pictures, which was highly correlated to a hidden tank being corrected. In Iraq, it found to few leaves to count.
I stand by my assertion that AI necessarily means not really understanding how the machine makes decisions - otherwise it wouldn’t be AI. I do not want to fly on an aircraft governed by AI - and I am in the business of approving software for use on airplanes.
The first problem was that the AI did not say in English that it was counting leaves.And the more general problem is obvious: the system learned from a far-from-globally representative data set. In the case of an airplane, I would think that a good approximation of a globally representative data set could be constructed. An awful lot of flights are in the air as you read this, and they generate a lot of data.
And after all, simulators can generate all the data you need - and if the AI system gets bad skinny from the simulator, so do all the human pilots who train on them.
But for a system to be smart enough to be competent to fly the plane, it would need a lot of anomaly data - from broken sensors in the real world, and (simulated, presumably) data on behavior of aircraft with all manner of possible degradations/failures of the controls.
Back in the piston-engined airliner days (but post-WWII) there was a story in the Reader’s Digest about a flight which lost its tail in a collision, and dived toward the ground. I was astonished to read that the pilot was able to crash-land the plane by the expedient of firewalling the throttles to lift the nose.
Seems like the least instinctive thing you could do under the circumstances, but then - he had no other pitch control, and it (nearly) worked. Well enough that there were a lot of survivors. If an AI system had had that couth built in, and had firewalled the throttles quicker, the plane might have pulled out AGL instead of slightly below it.
Airbus would have cleaned Boeing's clock in sales if the 737 Max had a different type rating.
Boeing dragged its feet on a 737 replacement. When Airbus upgraded the 32X series with the NEO, Boeing knew it needed to play catchup.
Boeing really needs a clean sheet re-design of the 737. The airframe debuted in 1968....and it was based on the older 707.
Not to mention the Pratt GTF (on the Neo) is quieter, more fuel efficient, and has much lower nitrous emissions.
Yes, the GE/Safran CRM has had teething issues due to being “rushed’ as well!
AOA is a powerful tool, but in my experience it is less reliable than normal pitot static instruments. Even in carrier based planes where AOA is the primary landing speed instrument, we always checked it against airspeed before commencing the approach.
By incorporating it in the MCAS system, Boeing is making AOA a primary flight instrument. I would think you could add a comparator between the two guages, with an associated light, and corresponding procedure.
Boeing dragged its feet on a 737 replacement. When Airbus upgraded the 32X series with the NEO, Boeing knew it needed to play catchup.
Boeing really needs a clean sheet re-design of the 737. The airframe debuted in 1968....and it was based on the older 707.
Boeing didn't have enough time to develop a whole new type. I'm pretty sure this will be the last major revision of the 737. Boeing needed a plane to compete against the A320 NEO series. I don't really find anything wrong with doing the Max. The 737 Max is lighter at equivalent passenger densities and has the advantage on shorter flights of about 500 miles or less. The Max is more efficient in climb. It's only on the longer flights where most of the time is spent cruising that the A320 has better seat mile costs.
Boeing is probably waiting for GE and CFM to come out with a reliable geared turbofan engine before committing to a totally new type based on technologies of the 787 and 797. It makes no sense to try to build a whole new type before the engines are mature. Any new type will benefit from the new technologies that come along in the next 10 to 15 years.
The problem with the 737 Max MCAS is being tied to only one sensor. If that sensor is faulty, the MCAS can command the nose of the plane to point nose down. Any 737 pilot who is not proficient enough with how to handle runaway trim can get into trouble really quickly. The MCAS needs to be easier to disengage. Other than MCAS, the 737 Max the same as the 737 NG.
By incorporating it in the MCAS system, Boeing is making AOA a primary flight instrument. I would think you could add a comparator between the two guages, with an associated light, and corresponding procedure.
Probably more than just compare the two gauges. I don't think I'd want to be flying ETOPS 180 in a 737 that can't figure out which AoA is faulty. Two gauges just means you can tell that one of them isn't working correctly. How are the avionics going to figure out which one to use?
What you FAIL to realize is that the 737 Max does not fly like a 737 NG on climb out.
Boeing knew this and used MCAS to cover the difference. The FAA approved it.
But what you FAIL to understand is that, at first, Boeing did not disclose the details of this system to pilots. That is what led to the letter from the pilots union.
When things are going wrong in the cockpit and the stick shaker is going while the plane is nosing down, 737 NG pilots will do what they do in a 737 NG simulator.
There are no 737 Max simulators....yet. They are being built because the pilots demand it and Boeing realizes it screwed up big time.
There is a reason why there is a CRIMINAL investigation into this rushed plane. Both Boeing and the FAA are complicit in the deaths of over 300 people.
I didn’t fail to realize that the Boeing deliberately did not disclose the MCAS to pilots. I thought that was rather bizarre when I read about that after the Lion Air crash.
The non-Lion Air pilot who was sitting in the cockpit jumpseat on the same plane that crashed the next day, recognized that it was a runaway trim situation and correctly shut off the trim. There were 3 incidents previous to the Lion Air crash on the same aircraft that crashed. In all 3 cases the pilots successfully dealt with the condition. The 4th incident on the same plane, there was no one in the jumpseat to show the correct way to handle the runaway trim. But this begs the question. Why was a plane with a faulty AoA gauge allowed to fly 3 more times without being properly repaired? No airline in the US would be allowed to to that.
I think pilots flying for US based airlines that fly the 737 Max would correctly handle the runaway trim situations. Still it appears that the MCAS system as certified is orders of magnitude more likely to generate runaway trim situations. How a fly by wire system that was supposed to work so far behind the scenes that pilots would not be told about it, and flight manuals didn’t mention it, could be activated by non-redundant sensors boggles my mind. Quite frankly I was really shocked when I first saw the new tail for the 737 Max. It looks to be smaller than the 737 NG. That seemed to be quite a departure from the 737 NG. If I were a pilot, I think I would want to learn about what those differences were for.
Actually Southwest does have a 737 Max simulator. Guess what. That simulator has no MCAS failure modes.
The FAA need to get it’s stuff together. They risk losing reciprocity with other aviation regulators around the world.
I don’t see how this plane was rushed. The program was announced in 2011. It was studied for some time before that. The first flight was in January 2016. The first revenue flight was in May 2017. The main problem has been the limited supply of engines being produced by the CFM consortium. There are lots of 737 Max aircraft being stored without engines. Fortunately there haven’t been reports of engine reliability problems in use unlike the Pratt & Whitney Pure Power geared turbofans that have had huge teething problems with the A320 NEO. And don’t forget the Rolls Royce Trent 1000 engines on the 787. They have had huge problems with corrosion. Apparently there are issues with how they tested their engines. It is RR equipped 787’s that fly over Asia where lots of sulfur dioxide is emitted mainly form Chinese electrical generation plants and steel mills bur into the atmosphere that the corrosion issues are showing up. Guess what! The EU prohibits RR from testing their engines in Europe using high sulfur fuel. They likely would have discovered the issues in the development phase rather than after in operation.
“Why was a plane with a faulty AoA gauge allowed to fly 3 more times without being properly repaired?”
Red flag and very disconcerting.
There is debate if the AoA guage is faulty or just intermittently giving erroneous readings.
The Ethiopean Airlines CVR showed that Captain and FO had different AoA readings, That was one reason why the Captain handed control over to the FO after he thought he got the plane climbing again.
I think they will find that because they did not turn off MCAS, the Captain pulling on the yoke and changed trim only temporarily reset MCAS. After 10 seconds, MCAS turns back on and wants to trim the plane to nose down. This happened as the Captain handed controls over to the FO and MCAS re-activated and put them into a nose down situation again.
My understanding is that the trim forces can be so much that manual yoke and trim cannot overcome it.
Don’t forget, they were not far from the ground, nose down, stick shaking and warning buzzer trying to get the plane to climb. They may have forgotten about the Auto Trim Off switches altogether, resulting in their deaths.
Turning Auto Trim Off is not a 737 NG procedure....and that is a problem.
The worst problem is having MCAS use only one AoA sensor input. Two is better, but then the computer has to choose which one to believe and which one to ignore. Three is always the best because the computer can look for 2 that agree and ignore the oddball.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.