Pitch-up Illusion on Take-Off
On the 28th of April 2017, a single-engine turbo-prop flying at night under instruments crashed near Amarillo International Airport in Texas. The accident report had been available on the app.ntsb.gov site but appears to have been removed. You can download a copy using the CAROL query tool for the NTSB number CEN17FA168. This piece was written using my copy of the final report and the Report Docket.
The aircraft was a Pilatus PC-12, a single engine aircraft introduced in 1994 with a large payload and high cruise speed across long distances. The Pilatus PC-12 can be used in “rugged environment” with rough landing strips. The PC-12 is popular for corporate transport, as the “comfortable” configuration can take nine passengers. Flying magazine described it as “more economical to operate than any turbine airplane of similar size.”
This specific Pilatus PC-12, registration N933DC, was operated by Rico Aviation Ltd in the US under Part 135 (commercial, non-scheduled aircraft operations). The pilot was based at Amarillo International.
On that evening, the PC-12 was operating as an air ambulance flight, with the pilot and two medical crew members on board. The flight was scheduled to Clovis Municipal Airport in New Mexico to pick up a patient. The patient would then be transported to a medical facility in Lubbock, Texas.
The pilot did his flight preparation while waiting for the medical facility to be ready to receive the patient. He had almost 6,000 hours flight time, although just 73 on this make and model. He started his aviation career as a mechanic, progressing to repairman and flight engineer before qualifying as an Airline Transport Pilot license and Flight Instructor. His management and co-workers considered him to be a trustworthy pilot; the only negative point was a report from FlightSafety which described him as “behind the plane” during simulator training. Both the Chief Pilot and the Director of Operations had flown with the pilot recently and confirmed that they had no concerns or issues with the pilot’s flying skills.
His weight and balance submission for the flight from Amarillo included a 180-pound patient and 1,800 pounds of fuel, although he had no patient on board (yet) and normally should have departed with 2,000 pounds of fuel. The Chief Pilot of the operator believed that the pilot may have entered the weights that he was expecting to be departing from Clovis with.
At 23:32 local time, the pilot contacted Amarillo ground control to confirm he had ATIS (automatic terminal information service) information Oscar and requested IFR clearance to Clovis, New Mexico.
The ground controller issued a clearance “as filed”, meaning that the clearance was for the routing as submitted by the pilot for his flight plan, confirming that he had an initial climb to 8,000 feet above mean sea level and a transponder code of 4261. The pilot read back the clearance.
A few minutes later, the Rico Aviation medical dispatcher confirmed final acceptance of the mission. The pilot contacted Amarillo ground control again and requested taxi to runway 04 for departure. The flight was cleared for take off at 23:43.
As the aircraft climbed away from Amarillo, it started a right turn towards Clovis. The local air traffic controller noticed that the aircraft was broadcasting the wrong transponder code. He called the pilot to let him know and repeated the correct code: 4261
The aircraft levelled off at between 800 and 1,000 feet above ground level for about thirty seconds and the correct transponder code appeared on the controller’s radar. The controller confirmed this with the pilot and the PC-12 resumed its climb. The controller saw the aircraft disappear into the clouds and asked the pilot to contact departure.
The pilot changed frequencies to departure control and reported in as “with you at 6,000”. For his location, six thousand feet above mean sea level was about 2,400 feet above the ground. There was no sound of distress or mention of an issue.
A few seconds later, the PC-12 entered a rapid descent. Before any one had time to react, it disappeared from radar. The departure controller immediately attempted to contact the pilot but received no response.
About 1.5 miles south of the airport, the PC-12 crashed into a pasture and then burst into flames. There were no survivors.
Radar data showed that the aircraft had reached a speed of 17,000 feet per minute (descending at over 300km/hour or almost 200 mph) before crashing. A surveillance camera at a nearby truck stop recorded the lights of the aircraft descending at an angle of about 45° followed by an explosion.
The wreckage did not show any evidence of problems with the flight control system. The damage to the propeller hub and the bent blades made it clear that the engine was operating under high power when the aircraft impacted the filed. The ground scars and the damage to the aircraft, confirmed by the video footage, showed that the PC-12 was in a steep descent, nose low and wings level.
As far as anyone could see, the pilot literally flew a fully functioning aircraft into the ground. While the aircraft was climbing to 6,000 feet, the peak pitch angle was about 23° nose-up. But about 37 seconds before impact, the pitch angle decreased steadily until it reached 42° nose-down.
The pilot was the victim of a relatively rare but often fatal spatial disorientation known as the “pitch-up illusion” or more correctly, somatogravic illusion.
We have an inbuilt spatial orientation system which allows us to perceive motion and our three-dimensional position in relation to our environment. Our sense of orientation allows us to instinctively know where our limbs are and which way is up.
Generally speaking, we rely on visual cues to determine our orientation: we know where we are based on what we see around us. The visual system is reinforced by additional information supplied by the vestibular system, the balance organs located in our inner ears, and the proprioceptive system, also known as the “seat of the pants”, where tendons, muscles and joints detect control forces and pressure. The information collected by these three systems is integrated into a single model of orientation and is usually amazingly accurate.
However, our internal systems were designed for walking on the ground, not flying through the air in a Pilates PC-12.
Visual information supplies some 80% of the data used to define our orientation, with the remaining 20% split between the vestibular system and the proprioceptive system. A pilot who cannot see the horizon or other points of reference has a wildly less accurate understanding of his or her orientation. The vestibular system and the proprioceptive system are much less reliable and can lead us to experience a number of spatial illusions. Worse, because we are used to our sense of orientation happening subconsciously and very reliably, we are instinctively reluctant to consider outside information which might help to correct the illusion.
One of the side-effects of the way the vestibular system works is the somatogravic illusion. The vestibular system consists of the semi-circular canals and the otolith organs. The otolith organs are actually a clever biological linear accelerometer: small grains of calcium carbonate embedded in a gelatinous substances move forward and backward which tickles the end of small hairs, sending a signal to the central nervous system.
If you tilt your head up, the grains shift backwards. If you accelerate, the grains also shift backwards, bending the small hairs in a very similar way. Because our bodies have not evolved for fast acceleration, the central nervous system can misinterpret this data to mean that your position is pitching up as opposed to accelerating. If there is no other useful data from the other systems, there is nothing to correct this.
When an aircraft is taking off, you are accelerating and pitching up. In the dark with no visual references, the central nervous system is prone to interpreting sustained acceleration as tilting, so that you feel that you (or in this case the aircraft) are pitching up more than you actually are. In one example that I read, an aircraft accelerating from 170 to 200 knots over a period of ten seconds feels the same as pitching 9° nose-up.
So taking off in the dark, the central nervous system is quite clear that the aircraft is pitching up more excessively than it actually is. In response, the pilot presses forward on the control column to return to the normal climb. At the point where the pitch feels correct to the pilot, the aircraft is actually pitching down.
Researchers have explored somatogravic illusion since the 1940s, after a startling number of pilots taking off at night ended up pitching the aircraft nose-down and crashing into terrain. In modern times, it is most commonly associated with go-arounds at night, a high stress situation where it is easy to get distracted. Because it takes place while climbing away in non-visial conditions, it is often fatal before the pilot has a chance to consider the instruments and realise that he or she has fallen prey to an illusion.
On that night in April 2017, the pilot seemed to be lightly distracted or perhaps fatigued: broadcasting the wrong squawk and then levelling briefly as he updated his transponder. This would make him that much more susceptible to spatial illusions. He changed frequencies to say “with you at 6,000” although he had been cleared to 8,000 feet, and should have reported climbing through 6,000 feet. It seems likely that he was levelling out at this point in the climb.
But a bigger clue turned up in the maintenance logs, which showed that there had been recurring issues with the autopilot, a Honeywell KFC 325. There was a specific fault which affected a number of PC-12, in which the autopilot lost the ability to adjust the horizontal stabiliser trim. In the cockpit, the master warning light would illuminate accompanied by the continuous beeping of the autopilot-trim warning tone. In addition, a red TRIM light on the autopilot mode control panel and a red A/P TRIM light on the Crew Alerting and Warning panel would illuminate. The autopilot would automatically disengage.
This fault had been logged twice in the previous five months, once by the accident pilot and once by another pilot. The other pilot said that in order the stop the continuous beeping and extinguish the warning lots, he would press the autopilot test button on the autopilot mode control panel.
The Chief Pilot confirmed that there was a continuing issue with the autopilot and mentioned the same fix.
It would often disconnect unexpectedly, triggering a master warning tone. It would require the pilot to reset the system by pushing the autopilot test button, then re-engaging the autopilot.
Two days before the accident, the Chief Pilot had recorded a video of the autopilot disconnecting during a flight. Analysis of the video after the event concluded that the autopilot’s pitch trim adapter likely lost power for a moment, which caused the autopilot to disconnect. In the video, the Chief Pilot recovered by using the test button.
According to the manual, in this situation, the the pilot should disengage the autopilot and then remove power by pulling out the autopilot circuit breaker. There was no reference to using the TEST button or reinstating the autopilot.
On the night of the crash, the aircraft was in a climbing right turn when ATC contacted the pilot about the transponder code. The aircraft levelled off briefly while the pilot updated the code, thus the autopilot was not likely to have been engaged at that moment.
As the aircraft climbed through 1,000 feet, the pilot almost certainly engaged the autopilot, as was standard procedure at the airline.
The wind was from the north, 21 knots gusting 28, and moderate turbulance was reported in the area. The turbulance might have stopped the pilot from noticing that the autopilot had disconnected. It certainly would have contributed towards a feeling of spatial disorientation.
The aircraft continued to climb, reaching a peak pitch angle of about 23° nose-up at 37 seconds before impact.
A few seconds later, the pilot reported to ATC that he’d reached 6,000 feet.
At 31 seconds before impact, just after the report to ATC, the aircraft pitched down. At the same time, the roll angle increased to the left. It is clear that the autopilot was not connected, or else it would have correct the roll.
It seems clear that the pilot was pitching down based on the feedback from his spatial orientation systems which, in the dark, was misinterpreting the continued acceleration as a pitch up event. The aircraft’s pitch decreased steadily to 42° nose-down with the roll reaching 78° to the left. The pilot had just seconds to realise what was happening and to correct it.
Finally, the light bulb filament showed that the “autopilot disengage” caution indicator was lit at the point of impact.
The amber A/P DISENG caution message will illuminate 3 seconds after the signal input to the CAWS changes from 28V (A/P engaged) to 0V (A/P disengaged) and the Control Wheel Steer (CWS) button is not pressed. The caption will remain illuminated for about 26 to 27 seconds; it extinguishes at a maximum of 30 seconds from the initial time of the autopilot disconnect.
The autopilot disengage caution indicator remains illuminated for thirty seconds after the autopilot disengages, thus we know that it must have been disengaged within 30 seconds of impact. We also know that it was not engaged at the point when the aircraft started its descent and roll to the left.
Thus, at some point after the pilot pitched nose-down, he attempted to engage the autopilot. This was almost certainly an attempt to use the instruments and to correct the unusual attitude that the pilot realised that the aircraft was in.
What we’ll never know is whether the pitch trim adapter lost power in the last 30 seconds of the flight, which would have disengaged the autopilot, or whether the autopilot was engaged too late to make a difference. At nine seconds before impact, the aircraft’s limit load factor was exceeded which would automatically disengage the autopilot. If the pilot had only turned it on a few seconds earlier, there would not have been time for the autopilot to correct the flight attitude.
From the report:
The apparent pitch and roll angles, which represent the attitude a pilot would “feel” the airplane to be in based on his vestibular and kinesthetic perception of the components of the load factor vector in his own body coordinate system, were calculated. The apparent pitch angle ranged from 0° to 15° as the real pitch angle steadily decreased to -42°, and the apparent roll angle ranged from 0° to -4° as the real roll angle increased to -78°. This suggests that even when the airplane was in a steeply banked descent, conditions were present that could have produced a somatogravic illusion of level flight and resulted in spatial disorientation of the pilot.
The pilot’s loss of airplane control due to spatial disorientation during the initial climb after takeoff in night instrument meteorological conditions and moderate turbulence.
You can download a copy of the final report by searching the CAROL query tool for NTSB number CEN17FA168.
It is easy for us to say that a pilot must always rely on instruments and not rely on his or her own perception. This dismisses the point that we have complex systems which define our sense of time and place; it isn’t something that we consider or conclude. Because we instinctively know where we are in relation to the earth and each other, we resist taking in new information which proves that our sense of our position or motion is wrong, a particular risk in three-dimensional space.
Studies have shown that pilots with more experience are just as susceptible to spatial disorientation and that somatogravic illusion is most often suffered by pilots who hold instrument ratings. This seems obvious, in that instrument rated pilots would be more likely to fly at night and, more importantly, would more likely have survived to tell the tale.
One study concluded that pilots fall into two categories: those who have suffered from spatial disorientation and those that haven’t …yet.