Monday, July 08, 2013

Let the Speculation Begin

Updates:

For particularly outstanding speculation, go here. He is a general aviation pilot and aerospace engineer. He used ATC tracking data to provide more factual bases upon which to speculate. All the news outlets should stop reporting and refer to him. (Gratifyingly, his conclusions are the same as mine.)

I had the seat positions reversed. PF was in the left seat, getting initial operating experience as a Captain. The instructor pilot was in the right seat; I have made changes where appropriate.

The NYT today has a classic case of journalistic buffoonery: taking a simple fact and following it to never-never land:
Investigators in the cockpit of the wreckage found the auto-throttle switches set to the “armed” position, meaning that the auto-throttle could have been engaged, depending on various other settings, she said. The disclosure is far from conclusive, but raises the clear possibility that there was a mechanical failure or that the crew misunderstood the automated system it was using.
In other words: we have no idea what we are talking about. If the journalist had spent any time at all -- say ten minutes with Wikipedia, and looking around for a subject matter expert -- said journalist would have learned that there are many autothrust system (ATS) modes. One of them could have been selected before being cleared for the visual approach, then forgotten.

It appears the PF might have selected "Level Change" as the vertical guidance mode. This would have clamped the throttles in idle until capturing the target altitude, at which point the Flight Management System (FMS) will transition to speed-on-thrust. The flight management system would then use pitch to control speed (too fast, raise nose; too slow, lower it). It is a mode I almost never use, although many guys do. IMHO, its only purpose is to cause the FMS to ignore any altitude restrictions en route to the assigned altitude because of reasons. It would have been a completely inappropriate mode to use on a visual approach, though, for several reasons. First, there is no altitude to capture. Second, by no later than 500' above ground level, stabilized approach criteria require the airplane be on speed, which requires a speed-on-thrust mode. Additionally, speed-on-pitch only makes sense if you are on speed already; otherwise, if steep in fast selecting speed-on-thrust will put the throttles in idle until capturing the target speed, at which point the FMS will manage thrust to maintain speed. It is possible that PF had selected Level Change while getting vectors on downwind and then forgot about it when cleared for the visual approach. Rolling out on final steep and fast would have camouflaged the incompatible mode until short final, when the airspeed finally went below target. It is still baffling how the crew could have let it get even 5 knots slow, never mind 20, or why the pilot monitoring did not note the inappropriate ATS mode.

An A330 Captain friend of mine sent me this:

Here's an email from one of the United 747 pilots that witnessed the 777 crashed in SFO. They were holding short of runway 28L at the time of the crash.
On July 6, 2013 at approximately 1827Z I was the 747-400 relief F/O on flt 885, ID326/06 SFO-KIX. I was a witness to the Asiana Flt 214 accident. We had taxied to hold short of runway 28L at SFO on taxiway F, and were waiting to rectify a HAZMAT [hazardous materials] cargo issue as well as our final weights before we could run our before takeoff checklist and depart. As we waited on taxiway F heading East, just prior to the perpendicular holding area, all three pilots took notice of the Asiana 777 on short final. I noticed the aircraft looked low on glidepath and had a very high deck angle compared to what seemed “normal”. I then noticed at the apparent descent rate and closure to the runway environment the aircraft looked as though it was going to impact the approach lights mounted on piers in the SF Bay. The aircraft made a fairly drastic looking pull up in the last few feet and it appeared and sounded as if they had applied maximum thrust. However the descent path they were on continued and the thrust applied didn't appear to come soon enough to prevent impact. The tail cone and empennage of the 777 impacted the bulkhead seawall and departed the airplane and the main landing gear sheared off instantly. This created a long debris field along the arrival end of 28L, mostly along the right side of 28L. We saw the fuselage, largely intact, slide down the runway and out of view of our cockpit. We heard much confusion and quick instructions from SFO Tower and a few moments later heard an aircraft go around over the runway 28 complex. We realized within a few moments that we were apparently unharmed so I got on the PA and instructed everyone to remain seated and that we were safe.

We all acknowledged if we had been located between Runways 28R and 28L on taxiway F we would have likely suffered damage to the right side aft section of our aircraft from the 777.

Approximately two minutes later I was looking out the left side cockpit windows and noticed movement on the right side of Runway 28L. Two survivors were stumbling but moving abeam the Runway “28L” marking on the North side of the runway. I saw one survivor stand up, walk a few feet, then appear to squat down. The other appeared to be a woman and was walking, then fell off to her side and remained on the ground until rescue personnel arrived. The Captain was on the radio and I told him to tell tower what I had seen, but I ended up taking the microphone instead of relaying through him. I told SFO tower that there appeared to be survivors on the right side of the runway and they needed to send assistance immediately. It seemed to take a very long time for vehicles and assistance to arrive for these victims. The survivors I saw were approximately 1000-1500' away from the fuselage and had apparently been ejected from the fuselage.

[Original post begins]
Regrettably, I have another opportunity to provide some expert aircraft mishap commentary and speculation. News coverage of Asiana 214 has been reasonably good, considering that the reporters are non-specialists. What I have read has been somewhat garbled, but the gist of the thing is pretty much there.

First, a quick summary of the Asiana 214 mishap sequence:
  • Weather was good, and the winds were light.
  • The glide slope component of the Instrument Landing System was out of service.
  • The First Officer (FO) left seat pilot was the pilot flying (PF), and was undergoing Captain Initial Operating Experience (IOE) Given the number of hours in the 777, he almost certainly had fewer than 10 landings in the airplane. He would have had many more in a very realistic flight simulator.
  • The right seat pilot Captain (CA) was the pilot monitoring (PM), an instructor pilot, and the pilot in commandand possibly a line check airman.
  • The flight's intermediate approach was over San Francisco heading south-southwest.
  • Approach control cleared the flight for a visual to runway 28R (28 refers to the runway heading rounded to the nearest 10 degrees; R means the righthand runway).
  • The aircraft flew a left hand 180 degree turn to line up on final for Rwy 28L
  • On short final, airspeed was well below (perhaps as much as 20 knots) the target approach speed of approximately 145 knots.
  • The aircrew made a very late attempt to go around. The aircraft pitched up, but there was insufficient airspeed and power to convert the descent into a climb.
  • The aircraft impact point was approximately 2,000 feet short of the glide slope intercept, and 2,500 − 4000 feet short of the landing zone. Based on photos of the debris field, and some happenstance amateur video, it appears the tail of the aircraft struck the 13' high sea wall.

Screen shot of the KSFO airport diagram, with some crude additions. The flight path was from the lower left of the screen. The impact point was probably the right side of the "8" in 28L. The rest of the line shows the approximate path of the aircraft until it came to rest. The bracket above the runway represents the span of the runway between the glide slope intercept point and the end of the touchdown zone. The white bar across the runway just before the beginning of the bracket indicates the runway has a displaced landing threshold. The portion of the runway preceding the displace threshold may be used for taxi and takeoff, but not for landing. The purpose of the displaced threshold is to provide error margin in case of a short landing.


That the ILS glide slope, which provides precise guidance in poor weather, was out of service has received prominent mention, far beyond its actual importance. The normal approach glidepath angle is 3.0º. Given the precision of that number, it should come as no surprise that it is fairly critical. If the approach is more than a half degree too steep, then arresting the sink rate at the right height above the runway becomes increasingly difficult. More than a half degree too shallow and, particularly for large aircraft, there is the risk of the landing gear impacting approach lights. If the approach is very drug in, aimpoint errors increase exponentially[1].

Attaining and maintaining this glidepath isn't easy; due to human visual limitations, at night it is very difficult. Consequently all major airports, and even most small airfields, have Precision Approach Path Indicators. When not obscured by weather, they provide excellent glidepath guidance from about 10 miles out all the way to touchdown.

The PAPI is the row of four lights just to the left of the runway. Each light changes from red to white at a specific angle. When on the proper glidepath, the inner two are red and the outer two white. In this picture, since the 3rd light is just starting to turn read, the pilot is very slightly, about 1/6th of a degree, below glidepath.

So, the big question here is how, despite extremely benign conditions and airport facilities completely adequate for those conditions, the pilots managed to convert a fully functional airliner into junk.

Some initial reporting indicated the aircraft had been flying a steep final. However, the video clip, which captures the last 10 seconds before impact, seems to contradict that: while the distance and oblique angle makes it difficult to tell for sure, to my eye the glidepath was very shallow. Additionally, the aircraft is visibly slow. Because there is a lot of airplane behind the main landing gear, large aircraft have a landing deck angle of about 5º in order to avoid striking the tail. SWAGging, I'd put the deck angle at more than 7º.

This isn't as contradictory as it sounds, though. One of the most challenging problems in flying large aircraft is energy and flight path management — which is really two ways of saying the same thing. My guess is that traffic was relatively light, and approach control cleared Asiana for the visual when the aircraft had just passed the field, leaving it up to the aircrew to decide when to start their left turn to final. Since the pilot flying was in the right seat, he would not have been able to see the runway. Instead, he would have had to rely on other means to ensure sufficient distance from the airfield to get the airplane slowed, configured, and on the proper glidepath.

My hypothesis is that Asiana turned in too soon, leading to a steep final and consequent difficulty in reducing airspeed. This is a classic setup for getting behind the plane, particularly for someone with little experience in type. Subsequently, the high sink rate and task saturation caused the PF to descend well below glidepath before correcting. However, the correction was insufficient to put the aircraft on the correct glidepath before touchdown. Critically, the PAPI would have been all red, which means it was providing no additional guidance. Complicating matters, final was over calm water, which practically eliminated the pilots' ability to visually estimate altitude[2].

So far, so obvious. The puzzler is how the airspeed got so slow. Somehow, the auto throttle system (ATS) was no longer controlling speed. In my aircraft (MD11), the ATS may be manually overridden at any time for any duration, but will remain active. However, in the 777, near as I could tell from talking to a 777 FO this morning, overriding the ATS for greater than a certain time results in the ATS disconnecting. In my speculative scenario, the PF would have held the throttles in idle during the steep descent, and failed to note the ATS commanding increase thrust after the aircraft went through the desired glidepath. Subsequently, the ATS disconnected. the PF selected an inappropriate ATS mode, which the PM failed to catch, and because the pilots were fixated outside on saving the approach, they neglected to monitor airspeed. The 777 does have airspeed limit protection, but the aircraft hadn't reached that threshold before they suddenly swapped ends — the sudden increase in angle of attack triggered the stall warning — just before impact. The stall warning would have triggered airspeed protection, but by then it was too late.

Recapping the speculation: a short turn to final created a steep approach which task saturated the PF leading to late recognition that they had gone well below glidepath. Failure to regain the glidepath, aggravated by the loss of depth perception, caused an inability to detect that even though the cockpit might make to the runway, the tail wouldn't. When they finally recognized the problem, there was insufficient energy available to change the flight vector.

Presuming my speculation is reasonably close to reality, the mishap report will have one cause: aircrew failure to adhere to established stabilized approach criteria.

At my company, and Asiana must be similar, in good weather the aircraft must be on glidepath, fully configured, checklists complete, on airspeed and engines spooled up no later than 500' above ground level. (There is even a synthetic female voice that makes the 500' callout.) If any of those conditions aren't met, or if any significant deviation occurs below 500' AGL, then the first person to detect the deviation must direct a go-around.

Since 500 AGL is nearly two miles, the better part of a minute, from the runway, the aircrew had failed to execute a go-around well before it comes into view in the video, eons before the crash.

When I first heard about this crash, I immediately thought I caught a whiff of the same stench of incompetence wafting through AF447. On thinking this one through, though, it is clear that given the right (and therefore, thankfully rare) combination of circumstances, I too could have become intimately familiar with how narrow the dividing line sometimes is between a successful flight and a disaster. [added:] This mishap is starting to look like having the same fundamental problem as AF447. When asked to fly the airplane, the pilots couldn't. It isn't easy to maneuver an airliner onto final while configuring to land, but doing so should be an unquestioned part of a pilot's skill set. (You would think so, anyway; however, on a recent flight into Anchorage we had the same setup as this mishap, with the added complication that the PAPI was also inop. The Captain required significant coaching to fly a proper final.) That isn't to say even good pilots won't goon it up, and get so channelized on the task that they fail to make the decision to go around. (I saw this happen once in the flight simulator, with a 35 knot overshooting crosswind thrown into the mix.) What is most unfathomable about this mishap is how two other pilots could sit through the whole thing without saying a word. Over dinner a couple nights ago, the consensus among the pilots at the table was that there is a widespread lack of basic flying skills caused by too much reliance on automated flight systems. Sadly, that sounds about right.


[1] To illustrate this, assume a 90 degree glidepath; i.e., straight down. A small change in the flight vector causes a small change in the impact point. Now assume a glidepath of not quite zero degrees. A small change in flight vector will create a huge change in impact point.

[2] In a previous life, my base lost an F-111 when the pilot, while on an overwater range, failed to detect a slow descent over calm water, despite looking outside. On another clear but hazy and windless day, I noticed a boat in the sky. A guy I know was conducting an simulated attack on one of our carriers in the Med; he watched an F-14 attempting a stern conversion on him fly into the water. Had Asiana been flying over land instead of water, I am certain their peripheral vision would have clued them in long before they got to the runway.