Sunday, May 24, 2009

Kicking Tin

Last February, a commuter flight from Newark to Buffalo entered a spin and crashed while configuring to land, killing all 49 aboard and one on the ground. It was the worst airline crash in seven years.

How is it that a perfectly good airplane just stops flying? On hearing that it occurred at night where icing had been reported, I assumed the crew had been caught out by airframe icing, which, particularly in smaller airplanes, can lead to losing control while configuring to land. As it turns out, that is only part of the story.

Two weeks ago, the New York Times reported on the National Transportation Safety Board's (NTSB) cockpit voice recorder (CVR) transcripts and digital flight data recorder (DFR) reconstruction of the flight's final two minutes.

As non-specialist reporting goes, I must give credit where it is due: the reporter went about as far as he could, stuck pretty closely to the facts, and made no significant errors; it is a very credible piece of work. Since I am a specialist, though, I am going to dissect the mishap a little more closely than did Mr. Wald. I am going to speculate a bit, but I will make it clear when I am doing so.

First, a precis of the mishap itself. [Some of this goes beyond anything in the NTSB documents. It is not speculation, though, as the information comes from DFR based flight path simulation video.
  • The Bombardier Dash 8 was slowing to final approach speed, 118 knots, at glide slope intercept altitude, about 1700 feet above the ground.

  • A few seconds before extending the extending the flaps to 15 degrees, the airspeed decayed to Vmin (1.3 times the stall speed for the configuration).

  • At this point the stick shaker activated, and the throttles advanced to nearly full power. The stick shaker warns of approaching the stall angle of attack; i.e., the angle between the wing and the relative wind that produces the maximum lift, and beyond which lift quickly decreases.

  • The pilot flying (PF, the Captain in this case) reacted to the stick shaker by pulling on the yoke.

  • The airspeed then decayed at an increasing rate until the wings stalled, and the stall was allowed to progress into a spin.

  • It took approximately 12 seconds for the aircraft's ground speed to go from 150 mph to zero.

So, how did the crew convert a perfectly good airplane into a smoking hole? The reasons read like a laundry list of Things Best Not Done: lack of professionalism, complacency & inattention, inadequate flying skill, incomplete training, and a longstanding transport category aircraft instrumentation omission. Oh yeah, and airframe icing. Everything but the first is my speculation.

The story's headline highlights a specific infraction: the pilots, particularly the Captain, far from rigorously observed the sterile cockpit rule. It prohibits non-essential conversation any time the aircraft is moving below 10,000 feet, or when in a climb / descent within 1000' of the new altitude. Almost all the Captains I fly with hew closely to that requirement; it is striking how often mishap CVRs show the opposite. Further, the first officer (FO) was suffering a head cold, and should probably have removed herself from the schedule.

The report mentions possible fatigue due to commuting -- the Capt slept in the crew lounge, and the FO commuted all night. However, flight ops can be astonishingly abusive of sleep cycles. Fatigue comes with the territory*, and cockpit discipline is the primary means to combat its effects. Besides, no amount of fatigue short of deep REM sleep explains the mishap sequence, and really tired people tend not to chatter.

Complacency and its evil twin, inattention, are an unfortunate consequence of modern glass cockpit aircraft with flight management systems (FMSs). Since FMSs know a great deal about the aircraft, its position in space, phase of flight and destination, the airplane takes care of many things itself. Among other things, FMS equipped aircraft will set target airspeeds based upon configuration, maintain those speeds with a combination of pitch and thrust while the automatic systems are engaged, and will even resume automatic control of the throttles to prevent speed excursions outside the allowed range.**

FMSs are, in general, great contributors to flight safety. When the conditions are not permissive (lots of traffic, bad weather, fatigue), an FMS provides two advantages. First, through programming approach details, it allows shifting some workload to the far less busy cruise phase. Second, in the terminal environment, it allows the crew to tell the airplane what to do without getting involved in the nitty gritty of actually making it happen. A woodpecker does not always have the best view of a forest.

Unfortunately, FMSs, while very good, are not perfect.

Sometimes they just do weird, umm, things.*** Glass babies, those who have spent essentially all their time on FMS, as opposed to clueless round dial airplanes, have a tendency to become passive observers. Well, the good ones, that is. The rest become merely passive.

What FMSs don't know, they cannot learn. There is no teaching an FMS about a wing with ice (actually, there can be; more later). It knows what the speed target for a configuration is, but it can be caught trying to do the impossible: fly a configuration predicated airspeed for a wing it no longer has.

That computers are imperfect, to the point of perversity, should come as no surprise to anyone more sentient than an iPhone. To comprehend that perversity, though, requires sufficient skill to know when the machine has boarded the handcart to hell. It is at this point, perhaps in the interest of avoiding speculation, where an otherwise well written newspaper article comes up a little short.

For all airplanes, there is a certain cadence to configuration for approach. While I am not a Dash 8 pilot, I am virtually certain that the pilots were in step with that cadence. The aircraft should have been roughly 15 knots above Vmin as the flaps went to fifteen degrees, and Vmin should have decreased by around 15 knots as the flaps extended.

For those of you at home in the pixel audience keeping score, a final approach speed of 118 should have a Vmin of around 113, and Vstall with landing flaps of roughly 87 knots. Vstall at the 15 degree flap setting would have been higher, but not much.

To put it in more notional terms, the crew was operating the airplane appropriately, but completely failed to take on board that it wasn't flying correctly: Vmin was far too high for the altitude, weight and configuration. The FMS was calculating Vmin on angle of attack, and target speeds on configuration. Those didn't add, and the pilots were too lackadaisical and ill-informed to break the code.

Being surprised is a wonderful thing. During Christmas and birthdays, that is. In the air, surprise is never, ever, good. Despite the wrapping being long off the gift -- the crew had almost 20 seconds to suss that the "foot" (the red vertical bar on the airspeed display indicating every bit of what red intuitively means) was where it most certainly did not belong -- neither pilot gave the tiniest sign of acknowledgment. Where mental warning flags and alarm bells should have been waving and clanging furiously, there was mere uncomprehending passivity.

So when the stick shaker fired up, it was a Jack-in-the-box moment to a crew who wasn't even aware the crank had been wound. Here is where inadequate flying skill comes into play. Yanking on the yoke when it is vibrating like a bad rental car at highway speed is just wrong. Like turning right at Indy, it is one of those elemental things one would think lies well into the realm of instinct.

This is where training needs to take a hit. My airplane is fully automated. And, oh by the way, it is the only one I have flown that doesn't require memorizing, or even knowing -- because the information is not available -- pitch and power settings for various airspeeds and configurations. Perhaps the Dash 8 is similar. Adding injury to insult, the flight manual essentially requires flying with the flight director (FMS computed steer-to bars) on. This does incredible violence to the control-performance concept of instrument flying (To abbreviate: setting specific attitude and power in order to obtain desired altitude, airspeed and heading. Repeat quickly and continuously). There is one sure result here. Performance cross check will atrophy because it almost never matters. Flight departments need to encourage turning off all the magic when conditions are permissive. I know of at least one that, until a couple days ago, did not.

All these paragraphs, and not one word about something that featured prominently, even amongst the impertinent chatter: icing. It is what caused Vmin to be so much higher than it should have been.

There are two things to be said here.

First, so what. There is no way the crew could to mount a response to that about which they were oblivious.

The second points an accusatory finger at aircraft designers. Angle of attack (AOA), the angular difference between the wing and the airflow, is essentially an instantaneous measure of the amount of lift the wing is producing with respect to how much lift it can produce. For a given combination of g-load, altitude, airspeed, weight and configuration, there is precisely one angle of attack. Not only is AOA the most fundamental measure of performance, it is also mechanically the simplest.

In fighter type aircraft, AOA is (g-limits notwithstanding) everything. In the realm relevant to the case at hand, the airspeed corresponding to landing weight is the crosschecked with AOA. If they match, wonderful. If not, something is up. Oh, and fly AOA on final.

Oddly, transport category aircraft ignore AOA so thoroughly that it isn't displayed anywhere. Had the the FMS been designed to fly AOA and compare the resulting airspeed against target airspeed, it would not have been trying to do the impossible. (Since the pilots had already proven themselves blind to the obvious, I doubt the addition of AOA would have caused them to change the outcome.) I can think of at least several crashes that would have been prevented with more emphasis on, and clear display of, AOA.

Flying can get very ugly very quickly. Once airborne, there is no pulling off to the side of the road. Cars do not fall through the pavement, or suddenly decide to head for the nearest ditch, if they get slow. This crew was undisciplined, oblivious and incompetent. I admit up front I don't know precisely what distinguishes profession from occupation, or occupation from marking off time. Despite that definitional inability, though, I insist that taking people into the air qualifies as a profession, and this pair comprehensively failed to meet even minimal requirements: they were passengers like the other 47 on board, distinguished only by getting paid for the trip.

++++

Miscellaneous notes about the article.

Officials of Colgan Air, the company that operated the Continental flight, offered startling testimony that pointed fingers at their own pilots.

John Erwin Barrett, the airline’s director of flight standards, said neither pilot in the twin-engine turboprop was paying attention to the flight instruments. A month after the crash, he said, a Colgan crew in the same model plane, a Bombardier Dash 8 Q400 on approach to Burlington, Vt., received the same warning of a stall, and recovered smoothly, landing without incident.


Officials at Colgan Air should be aware of when their own foot is in the crosshairs. Getting to the point of stall warning on a transport category aircraft means something has already gone badly wrong, no matter how successful the recovery.

Many recent crashes have raised questions about training and judgment, as opposed to mechanical failure — for example, the crash of another twin-engine turboprop, in Kirksville, Mo., in 2004, in which the crew violated operating rules and joked and yawned as they descended into the trees. The pilot of American Airlines Flight 587, which crashed in Queens in November 2001, used the rudder in a way that made the plane swing back and forth until the tail broke off.


The Kirksville crew was woefully inattentive during taxi to the runway. They were most certainly not joking and yawning as they left the ground.

As for the American pilot, he was an ex-Eagle driver, accustomed to using the rudder to control roll; unbeknownst to him (see Surprises, above) he was flying along the wing vortex of a preceding B747-400. Airbus neglected to mention in the flight limitations section any restrictions on rudder use. Why? Because no one foresaw that one-off circumstance.


* E.g. I will depart Newark for Europe in five hours at 11 pm body clock time.

** The FMS in my aircraft does all these things. Since the Dash 8 is a newer aircraft, I suspect, but do not know, that it has the same capability.

*** E.g. Enroute Here from There, the FMS adjusted the power a little high, then picked at an electronic hangnail while the airspeed increased by roughly 15 knots to Vmo (max Mach). Then, about the time the Captain and I were doing the Spock arched eyebrow thing, yanked the throttles to idle. If I had done that on the DC-9, which didn't know automatic from potatoes, I would have gotten slugged. Verbally, of course.

5 Comments:

Blogger Harry Eagar said...

Some 30 years ago, the cockpit transcript was published of a passenger jet that collided with a small plane on approach to the San Diego field. Cockpit indiscipline again.

A deadheading pilot hitching a ride was chattering about his pension outlook to the working crew. His last words were, 'O shit.'

Not sure of the human engineering of that one. The story accompanying the transcript did not say, but my guess is that the pension-concerned pilot outranked the guy driving the plane, who may have been shy about telling him to shut up.

Human engineering is the hardest kind.

The Buffalo pilot was a friend of one of my prospective sons-in-law, who just got promoted to captain by Colgan. He considered him a good pilot, but they fly different machines, so I don't know how specific that judgment was.

What say you about the pilot's failure to pass flight training tests? Relevant?

May 24, 2009 8:14 PM  
Blogger Bret said...

"Unfortunately, FMSs, while very good, are not perfect. "

Unfortunately, apparently not all pilots are perfect either.

Why don't they add dealing with icing to the software?

May 25, 2009 5:03 AM  
Blogger erp said...

As a passenger, I'd like to think the people in the cockpit took their job seriously and had the skills to cope with emergency situations. I'd hate to think that the off hand attitude that has infected every other profession has caught up with pilots too.

May 25, 2009 6:56 AM  
Blogger Hey Skipper said...

Harry:

The San Diego midair really isn't a good example.

While I am less than entirely sure about this, I think that midair preceded the sterile cockpit rule.

More importantly, though, the mid-air problem has been essentially solved by purely technical means: TCAS (Terminal Collision Avoidance System). Well, that and better defining airspace control areas around airports.

The Buffalo pilot was a friend of one of my prospective sons-in-law, who just got promoted to captain by Colgan. He considered him a good pilot ...I hate to be harsh, but he wasn't. He was undisciplined in the cockpit; noted the presence of icing, but failed to think through the implications, and was gob-smacinkgly unaware of basic aircraft performance.

His check ride bust count points out the difference between becoming a civilian pilot vs. a military pilot. For a civilian, you are as good as your checkbook is deep. Not so for military pilot training. That many busts would have earned a position in another career field.

That, in and of itself, doesn't necessarily make him a bad pilot -- maybe he was just very unlucky on checkrides. And it doesn't necessarily mean a specific military pilot is better than a specific civilian pilot.

No major airline would likely hire a guy with that many bust, if the airline knew about it. Apparently, he was less than forthcoming with the information.

++++

Bret:

Unfortunately, apparently not all pilots are perfect either.You have hit the crux of the problem. Is the system overall safer with pilots heavily relying on FMSs, or with less reliance on the machine, and more upon fallible humans, just to keep them in the loop?

In my airplane, there is concern we have overdone the automation thing. Yet it would be silly to ignore the increase risk of pilots bending the airplane because they failed to control, say, power as well as the machine.

Clearly, I think permissive conditions require turning the magic off, because the ability to closely monitor the airplane when the going gets rough is related to perishable flying skills.

++++

erp:

As a passenger, I'd like to think the people in the cockpit took their job seriously and had the skills to cope with emergency situations.Most do, especially at the top of the heap. This crew was far less experienced, and making much less money, than any pilot at a major airline

There is a lot of autonomy that comes along with being a Captain. Some people just can't, or won't, do what they should, when they should.

As for me, I just don't want to run the risk of sounding bad on a post-mishap CVR analysis.

May 25, 2009 1:20 PM  
Anonymous Anonymous said...

This comment has been removed by a blog administrator.

June 15, 2009 3:43 AM  

Post a Comment

<< Home