Crash Investigation Reveals Gaps in Airline Safety System

Wed, May 20, 2009 — David Evans

Articles, Featured

The more officials from Colgan Air tried to distance themselves from, and blame, the dead pilots for the crash, the worse it appeared. Under occasionally sharp questioning by National Transportation Safety Board (NTSB) investigators, it was apparent there was a vast gulf between Colgan’s stated commitment to safety and the awful reality. Carrier officials, in fervently mouthing their practices were the “industry standard,” were revealing grievous weaknesses not only at Colgan Air, but throughout the industry.

The occasion was three days of fact finding hearings, 12-14 May, into the crash earlier this year of Colgan Air flight 3407, a Bombardier Dash 8-Q400. Colgan was operating the airplane as part of Continental Express. On a 12 February flight from Newark to Buffalo, the airplane was slowing and turning onto the heading for the runway, in icing conditions, when it experienced a stall and gyrated wildly through the sky. Neither Captain Marvin Renslow, who was the handling pilot, nor First Officer Rebecca Shaw, the pilot not flying (PNF), responded correctly to the stall situation. The airplane plummeted, in a severe nose-down attitude, about 2,000 feet to impact. All 49 persons aboard were killed, plus an individual on the ground. (See Air Safety Journal, ‘Buffalo Icing Crash Raises Issues of Airplane Design & Crew Training.’) It was the worst airline disaster in the U.S. since the Comair regional jet crash on takeoff August 2006 at Lexington, Ky., which killed 49.

The Dash 8-Q400. Colgan Air was flying for Continental Express, hence the Continental livery.

The Dash 8-Q400. Colgan Air was flying for Continental Express, hence the Continental livery.

Harry Mitchel, Colgan’s VP of Flight Operations, said that in the 30 seconds before the stall warning the crew was not monitoring their primary instruments, per their training, and they were not aware that airspeed was bleeding off (some 50 knots of speed was lost in less than 30 seconds).

They were also flying in icing conditions with the autopilot activated. In the past few years, the NTSB has repeatedly warned of the hazard attendant to flying in icing conditions with the autopilot on. Specifically, the autopilot will add pitch to counter the negative effects on lift and drag of ice buildup until it has no more trim available, at which time the autopilot will snap off and hand the problem to the surprised crew. Perhaps more importantly, with the autopilot turned off, changes in the handling characteristics of the airplane would be felt by the pilot in time to prevent the stall.

That the airplane was in icing conditions was clear from the cockpit voice recorder (CVR) transcript of the pilots’ remarks:

FO Shaw: Is that ice on our windshield?

Capt. Renslow: Got it on my side. You don’t have yours?

FO Shaw: Oh yeah, oh, it’s lots of ice.

Capt. Renslow: Oh yeah, that’s the most I’ve seen – most ice I’ve seen on the

leading edges in a long time. In a while, anyway, I should say.

The anti-icing systems were activated, but ice was still accumulating. NTSB Member Deborah Hersman asked, how was this possible?

The crew was looking at the leading edges, in the “dwell period” between boot inflations, said Bombardier’s Harlan Simpkins. There are two other possibilities. First, ice is not always completely removed with each boot cycle and forms a site for further accretion. Second, distributed roughness can sometimes form in certain icing conditions. Areas of distributed roughness do not touch and are not effectively removed when the boot cycles.

Intercycle ice accretion on a de-icing boot, at the sixth cycle, with one-minute cycle intervals.

Intercycle ice accretion on a de-icing boot, at the sixth cycle, with one-minute cycle intervals.

Roughness of only 0.08 inch thickness resulted in an increase in stall speed that resulted in a 1996 accident involving a Cessna 560 in Augsburg, Germany.

Roughness of only 0.08 inch thickness resulted in an increase in stall speed that resulted in a 1996 accident involving a Cessna 560 in Augsburg, Germany.

This response about dwell period does not quite suffice. The anti-ice system has two speeds – slow and fast – and the crew was using the fast speed. The various Bombardier witnesses, in extolling the effectiveness of the accident airplane’s anti-ice systems, were rather blasé about the ice buildup in the period between boot activation.

Captain Renslow had commanded the autopilot to level off the airplane at 2,300 feet after descending from a higher altitude. He had the airplane in a turn at that altitude to line up with the runway. The airplane is more vulnerable to stall during a turn. As the aircraft slows and is configured for landing (extending the landing gear, deploying flaps, etc.), the lower wing or inside wing in the turn will stall before the outside or higher wing. When the lower wing stalls, that wing will drop, the greater drag on that side causes the aircraft to yaw toward it, and the nose drops into a spin spiral. Compounding the asymmetric condition is the fact that the approaching loss of control is masked by the autopilot soaking up the effects of the ice-load by insidiously auto-trimming the aircraft.

Twenty seconds after deploying the landing gear, and coincident with the flaps extending beyond five degrees, the stick shaker activated, automatically disengaging the autopilot. Immediately after the autopilot disconnect, the pilot advanced the power levers to 75%. The yoke was pulled back. It was at this point that the aircraft made its initial roll to approximately 50 degrees. In addition, the Captain and FO were applying force to the rudder pedals, out of phase with each other, for about 18 seconds.

Pulling the nose up, rther than unloading the aircraft, was the worst possible response to the stall.

Pulling the nose up, rther than unloading the aircraft, was the worst possible response to the stall.

The stick shaker warns that airspeed has degraded to a point close to stall. Immediate action, defined as within three seconds, is needed to avoid a stall. Specifically, the nose must be lowered immediately and vigorously (not just relax the back pressure on the yoke, as the autopilot trimmed the aircraft nose up). Angle of attack must be decreased and airspeed must be gained. The technical term is to “unload” the aircraft. It should be mentioned that the speed tape on the instrument panel will feature a rising “barber pole” indicating the rising stall speed of the airplane. However, it is evident that neither Capt. Renslow nor FO Shaw was monitoring the speed tape/barber pole.

Both pilots were evidently caught by surprise. “Jesus Christ,” muttered Renslow. FO Shaw screamed, and the CVR tape suddenly ends.

With the aircraft then pointed to the ground, the correct control response would nevertheless have been to push the yoke forward, as it was vital to unload the aircraft and lower the angle of attack.

With the aircraft then pointed to the ground, the correct control response would nevertheless have been to push the yoke forward, as it was vital to unload the aircraft and lower the angle of attack.

It is apparent that Renslow pulled back sharply on the control yoke, pitching the nose skyward. This was the opposite of what he should have done. When the airplane was then pointed at the ground, he still did not unload the airplane; rather, he maintained a death-grip on the control yoke, persisting in pulling it back.

Power was added, but only to 75%  not full power.

FO Shaw added to the problem by retracting the flaps. She did this without coordination with Capt. Renslow. Often, after recovery from stall, if the flaps are retracted, the airplane promptly enters another stall. It is best to leave the flaps alone or extend them further if close to the ground.

If the control yoke had been moved forward promptly rather than back into his chest, and if the throttles had been firewalled, there is a fair probability that control of the airplane would probably have been restored, even at the flight idle power setting and certainly at 75% torque.

Time from stick shaker to crash was on the order of just 26 seconds. The crew, tired and caught by surprise, never correctly diagnosed the problem.

The tragic end to Colgan Air flight 3407.

The tragic end to Colgan Air flight 3407.

Colgan officials basically said the two pilots acted in a way that was contrary to the airline’s training. Paul Pryor, Colgan’s chief simulator instructor, said, “Both pilots received stall recognition training in the simulator.” And, he added, “They would have been tested” on their recovery technique.

All well and good, but Colgan’s haste to attribute this accident to pilot incompetence is too facile, too convenient.

A review of some of the facts revealed at the hearing indicates that the two pilots were products of the system, and they were put in a position of fatigue and confusion that made recovery from the icing upset problematic.

First of all, they lacked experience. The two pilots each had less than 200 hours in the Dash 8-Q400. While they had accumulated more hours in other kinds of aircraft, they had far less experience than airline pilots hired by the major carriers. In fact, regional airlines are often regarded as training grounds for pilots to accumulate flying hours necessary for meeting the minimums required by the majors.

Second, both were fatigued. Captain Renslow commuted from Florida before starting his duty period. FO Shaw commuted from Seattle. Based on text messages she sent during this commute – indicating she was awake during this time – it is estimated that she was awake for about 36 hours before her duty period with Colgan even began. It should be mentioned that Charles Lindbergh later recounted that he was hallucinating from fatigue during his solo trans-Atlantic flight in the Spirit of St. Louis, and he had been awake before that pioneering flight about the same number of hours as Shaw.

Third, Shaw had a cold, as evidenced by coughing recorded on the CVR. Coupled to fatigue, a cold is likely to render one even more “out of it.”

Fourth, both she and Captain Renslow engaged in extraneous conversation during the time when “sterile cockpit” rules were in effect. To be sure, engaging in conversation can be seen as a stratagem to stay awake, particularly at night, but the conversation also was a distraction. There are numerous occasions on the CVR where Shaw has been distracted from essential radio conversations and other preparatory-to-landing responsibilities. Renslow, in response to Shaw’s remark that “I’ve never seen icing conditions,” told a story about flying the Saab 340, where he saw ice build up and the airplane would “keep on truckin.’ ” This chit-chat occurred below 10,000 feet, where Colgan rules dictate that a sterile cockpit should prevail and only conversation related specifically to operation of the aircraft and the flight plan is allowed.

Fifth, neither pilot had actually experienced a stall in simulator training, which puts them smack in the mainstream of all other airline pilots. Simulators are not programmed with the flight characteristics beyond stick shaker. They had been exposed to approach to stall, and the resultant stick shaker, but they had not experienced an actual stall and the resultant stick pusher, which automatically moves the yoke forward to drop the nose. This motion can be overcome with about 70 lbs force by the pilot, but the stick pusher movement is described by pilots as dramatic and forceful.

Sixth, Captain Renslow had not disclosed to Colgan that he had failed at least two check rides before being hired. Colgan Vice President Mary Finnigan testified, “If we had known when Captain Renslow was in training that he had falsified his application and left off two failed check rides, he would have immediately been terminated.

Basically, Colgan Air officials heaped criticism on the two pilots, asserting in a statement provided reporters that they had “thorough initial and recurrent training on how to recognize an impending stall situation” and that Colgan’s flight crew schedule “provided rest periods for each of them that were far in excess of FAA requirements.”

Acting NTSB Chairman Mark Rosenker was not convinced. “I am concerned about the winking and nodding that I have seen in some of the policies in your company,” he told a line-up of glum Colgan officials at the hearing. “I don’t believe it is only in your company. I believe there are industry issues that we must examine here.”

First, the pairing of an inexperienced captain with an inexperienced first officer seems to be asking for problems. Neither pilot displayed adequate crew resource management (CRM) or airmanship skills. FO Shaw may have had 1,600 hours total air time, but she remarked during the fatal flight that she got more experience in icing conditions in one day flying the Dash 8-Q400. Since the accident, Colgan has instituted a policy that no pilot can be paired with another who has less than 75 hours in type; now, pilots must have a minimum of 100 hours. This policy seems more like an exercise in tokenism. The fact is that either figure – 75 or 100 hours – denotes a paucity of experience.

Secondly, Colgan insisted that pilots have a responsibility to show up for duty rested and ready for work. “Pilots are professional and must be rested; it’s the professional pilot’s responsibility,” declared Dean Bandavanis, Colgan’s director of flight operations.

At the same time, Colgan schedules pilots for 12, 14 and even 16 hours of duty. There is no evidence that anyone at the management level has actually gone on a three-day trip involving duty days of 12, 14 and 16 hours to experience the cumulative toll of fatigue.

Nor has anyone from management visited the crew operations room at, say, 2 or 3 o’clock in the morning, to see if crewmen are sleeping in the couches. It was suggested that Captain Renslow was doing this, as he had checked in around 3 a.m. FO Shaw remarked to a fellow crewmember that one couch in the Newark crew operations center ought to have her name on it, as she had slept there so often.

Colgan has a policy that crew members are not to sleep in the operations center, and in fact the lights are kept on all night to discourage the practice.

Three points are germane here:

• There is no evidence that management officials have ever visited the crew center at 2 or 3 o’clock in the morning to check compliance with their rule. If they had made such an unannounced visit, they would have observed pilots sleeping on the couches, including Captain Renslow.

• Of the 137 pilots assigned to the Newark base, 93 commuted to work. Among the commuting pilots, 49 travelled from domicile more than 400 miles distant and 29 lived more than 1,000 miles away. The pilots lived so far away because they could not afford the cost of living in the Newark area.

• The average annual pay of a Dash 8-Q400 captain at Colgan was $67,000 a year. The pittance paid first officers was $24,000. The scant pay was a major reason pilots were commuting to work.

Interestingly, Colgan managers receive a cost-of-living adjustment in expensive areas. Pilots, though, were not on this program.

Of interest, the crew centers at some other carriers are divided into two rooms: a well-lit facility for flight planning, etc., and a dark room with beds where sleeping is permitted.

In the area of fatigue, there seems to be a devil’s compact. The pilots’ union, the Air Line Pilots Association (ALPA), does not want any restriction on commuting to work. The airline wants freedom to schedule crews for up to 16 hours of duty per day. ALPA ignores the airlines’ scheduling practices, and the airlines ignore commuting as a contributor to fatigue.

Of note, schoolteachers, firemen, police and other public servants are often expected to live in the city or county where they work. The precedent has been set, and it seems that pilots could be expected to live, say, within a 2-hour commute to their base. Salaries would have to be adjusted upward for high cost of living areas. The fatigue problem is amenable to a change in policies, it seems.

But endemic pilot fatigue is not recognized at Colgan. Daniel Morgan, the company VP for safety and regulatory compliance, flatly declared, “I don’t have serious issues with fatigue at this airline.”

By the way, the telephone hotline by which crews can report safety problems is not used. It’s basically non-functional. The airline has an Aviation Safety Action Program (ASAP), by which pilots, flight attendants and mechanics can anonymously report safety problems. Morgan said he receives about 12-15 ASAP reports a month, which he said is too few for trend analysis. There are 1,200 people at Colgan covered by ASAP, about 500 of whom are pilots. Assuming that all ASAP reports are submitted by pilots, the current participation rate is on the order of a miniscule 3%. This low participation rate suggests a lack of trust between pilots and management. Pilots appear fearful of disciplinary action based on what is disclosed through ASAP (which is defined as a nonpunitive means of reporting inadvertent safety problems).

Yet Morgan described the safety culture as “very, very good” at Colgan.

The company has “no serious issues with fatigue” because it isn’t reported through reporting systems as a threat to flight safety.

The third area in which comment is necessary is in the area of crew training, especially regarding wing stalls. Presently, Colgan pilots are exposed in the simulator to three kinds of approach to stall, a clean stall at 5,000 feet, a takeoff approach to stall and a landing approach to stall. As Douglas Lundgren, FAA’s principal operations inspector (POI) for Colgan said, “Pilots are expected to recover from the beginning of a stall.” (I.e., at the incipient stage, well before the manifestation of outright stall characteristics.)

They experience stick shaker, but they receive nothing more than a demonstration of the forceful stick pusher movement.

Moreover, approach to stall training in the simulator is highly scripted, along the lines of “Okay, we’re now going to practice approach to stall recovery.” The airplane is trimmed prior to stall training, and pilots are aware of what is expected. They are expected to recover with an altitude loss of no more than 100-200 feet (remember, they start at 5,000 feet). The element of surprise is written out of the training scenario.

The most outrageous aspect of approach to stall training at many airlines is that trainees are taught to hold their altitude throughout the maneuver. This is due to the FAA’s Practical Test Standards, which state that an applicant must recover to a reference airspeed, with minimal loss of altitude and heading deviation. The FAA never defines “minimal loss” of altitude. A great many instructors and check airmen have substituted their own standard, often 100 feet, as several Colgan instructors testified. Student pilots can be, and have been, failed for trading altitude for airspeed during a stick shaker recovery. Rather, they are taught to immediately go to full power, and use whatever yoke force is needed to keep the airplane level while it accelerates. This technique often involves “riding the shaker” for some time.

Pilots are being taught to stay in the shaker for longer than is necessary, rather than dropping the nose and unloading the airplane while adding power. Because pilots stop trimming at a higher airspeed, keeping the airplane level (for minimal altitude loss) can involve significant back force on the yoke until the airspeed increases. Therefore, simulator training is developing the exact motor memory that, if applied to a real world situation like Colgan flight 3407, will induce exactly the wrong control movements.

This particular airplane stalled during a turning approach to landing at lower altitude. In a banked attitude, one wing will stall before the other, heightening the element of surprise, as the airplane response is more dynamic, shall we say, then a plain wings level 1G stall.

The approach to landing situation is not covered in the simulator.

This thought occurs: approach to stall and stall recovery should be practiced in the simulator on a surprise basis. The trainer should be able to call for stall and recovery at any point in the flight envelope. This procedure would accustom crews to the element of surprise, and give them added confidence in recovery. As done now, pilots in simulator training know they are about to experience a stall, and which kind. Such an unanticipated and insidious scenario could be set up by the simulator instructor, winding in a tailwind during the base turn, forcing the crew to either steepen their turn onto final – or doing the good airmanship thing of leveling the wings and going around. The steepening turn to finals has been a feature of many accidents in the annals of aviation.

So what exactly happened on the Colgan flight 3407? About 53 seconds before impact, everything is quite normal. The aircraft is level at 2,400 feet and 180 knots, intercepting the localizer for Runway 24 at Buffalo. The flight has been cleared for approach. The autopilot is engaged, and the flaps are extended to five degrees (Flaps 5). The Captain is the pilot flying and the FO is monitoring him. The aircraft has a moderate amount of ice on it. The crew has correctly turned the “Ref Speeds” switch on the ice protection panel to “Increase.” In icing conditions, this action decreases the angle of attack at which the stall protection system activates. During the preparation for the approach, the crew set the approach speed to 118 knots, which was correct for their weight and configuration. But, they did not include the mandatory 20 knot additive for flight in icing conditions. This failure is a small but important detail that got overlooked.

At time 22:16:00, the Captain reduced the power levers to flight idle, as necessary to slow for landing. A few seconds later he called “gear down, localizer’s alive,” and the FO extended the landing gear and moved the condition levers to their maximum position, 1,020 rpm. Both the landing gear and the flat-pitch propellers on the Dash 8-Q400 are quite draggy, and the rate of deceleration increased. In the next ten seconds, the airspeed slowed from 170 knots to 149 knots, a fairly slow speed for Flaps 5.

At 22:16:21 the FO noted “gear’s down,” and the Captain immediately called “Flaps 15, before landing checklist.” One second later, at 126 knots, the stick shaker began clattering away, almost immediately accompanied by the incessant blaring of the autopilot disconnect horn. The Captain advanced the power levers to 70% and left them there, well short of the 130% emergency torque available. Inexplicably, he hauled back on his yoke, pitching the aircraft to a 30 degree nose up attitude. These actions suggest an intent to go around, not the sensible methodology of stall recovery.

Dr. Robert Dismukes, a National Air & Space Agency (NASA) research scientist, said repetitive training in stall recovery is essential – but still not a guarantee of success.

“For unexpected events, pilot response is very dependent on training,” he said. “For a moment, our mind is discombobulated, this shouldn’t be happening. But if trained, the pilot may react correctly.”

Note the “may react correctly.” In a study of pilot reactions using an instrumented Learjet, Dismukes said even experienced pilots “flailed around” in upsets. Of eight different scenarios 40 pilots were put through, a nose high attitude and a nose low spiral dive were the two most failed events in upset recovery training. Note that in this Colgan air crash, the inexperienced crew experienced both nose high and nose low attitudes.

The NTSB chose not to call as witnesses the two Colgan pilots of another Dash 8-Q400 that experienced stick shaker later this year and successfully recovered the airplane. Here is a case where Colgan has a 50% success record on stall recovery: the two pilots who successfully recovered the airplane, and the two dead pilots who didn’t. Virtually nothing is known about the successful event: how experienced were the pilots, how long had they been awake, was icing a factor, was it at night (like the accident) and, perhaps most important, did simulator training positively affect the outcome?

There is also the question of whether the airplane’s systems gave due warning to the crew of the drop off in speed in the moments before stick shaker. Member Hersman observed that a chime sounds in the cockpit when the autopilot is within 1,000 feet of leveling out the airplane at a prescribed altitude.

“Maybe the crew needs an airspeed deteriorating alert before stick shaker,” Hersman suggested. A change in altitude is, for the most part, a low consequence event, she said, but failure to monitor airspeed, as in this case, is critical.

The NTSB will be digesting these and a host of other issues in coming weeks as the final accident report and recommendations are crafted.

Meanwhile, the three days of fact-finding testimony have already spurred interest in Congress. Senator Byron Dorgan (D-ND), Chairman of the Aviation Operations, Safety and Security Subcommittee, said the “stunning” revelation at the NTSB hearing will be the subject of his subcommittee in early June.

“The disclosures about crew rest, compensation, training, and many other issues demonstrate the urgent need for Congress and the FAA to take actions to make certain the same standards exist for both commuter airlines and the major carriers,” Dorgan said. “The NTSB investigation has disclosed some very serious problems that need to be corrected immediately.”

But, as one industry observer, reflecting on the role the aircraft and its systems played, said:

“How ironic. In the Buffalo case, an auto-throttle (if fitted) COULD have added power and prevented this crash. And yet, in the 25 February crash at Amsterdam of a Turkish B737, the auto throttle caused it.” (See Aviation Safety Journal, ‘Interim Crash Report Focuses on Auto Throttles and Crew Actions’)

The final arbiter of successful outcomes is always an alert (i.e., rested) crew who utilize, but don’t rely upon or trust, their automation.

Comments are closed.

Nolan Law Group