On a winter night in February 2009, Colgan Air Flight 3407 crashed just outside of Buffalo, N.Y., killing all 49 passengers and crew aboard and one person on the ground. An investigation by the National Transportation Safety Board determined that the turbo-prop aircraft experienced many factors contributing to the crash, including an aerodynamic stall from which the aircraft did not recover.
“In most cases, a lot of things go wrong before a plane actually crashes,” says Peter Grant, a professor at the University of Toronto Institute for Aerospace Studies (UTIAS) and a leading expert on flight simulation. “Part of the challenge is that pilots are often trained on simulations that take an aircraft right up to the point of aerodynamic stall, but not past it.”
Aerodynamic stall occurs when the stream of air flowing over an aircraft’s wings becomes separated from the curve of the wing and no longer generates enough lift to counteract the plane’s weight. This is caused by reaching too steep an angle of attack — when the nose of the plane aims up too sharply. When an airplane reaches aerodynamic stall it often begins to roll and becomes harder to control, further complicating the situation for a pilot.
The correct response is to lower the aircraft’s nose, decreasing the angle of attack and re-establishing lift. This also makes the airplane easier to control, giving pilots a chance to correct even a severe roll. Commercial aircraft have various safeguards in place to prevent stall, such as alarms, a “shaker” mechanism or built-in “pusher” system that prompts the pilot to direct the nose down to lower the angle of attack when a plane approaches stall. Some large jets are equipped with “envelope protection” measures designed to keep the plane flying within safe parameters.
But even with these preventative measures, catastrophes such as the June 2009 crash of Air France Flight 447 — for which envelope protection had deactived when it hit aerodynamic stall in a storm and crashed into the Atlantic Ocean, killing all 228 aboard — reveal that recognizing and manoeuvering out of a stall may not be intuitive to pilots in an extremely chaotic situation.
“Automation in aviation has really reduced the incidence of crashes and made flying a lot safer, but currently, it can’t do everything,” says Grant. “Besides take-off and landing, most commercial flight is on autopilot, until unless the situation becomes critical. Suddenly, the autopilot switches off and we’re putting pilots in a position where they need to take over under the worst possible circumstances.”
Following the Air France and Colgan Air disasters, and several similar incidents worldwide, the U.S. Federal Aviation Administration called for researchers from industry and academia to design a new methodology for developing representative models of aerodynamic stall. These models will be used to create new simulations to improve pilot training regimens.
Grant and his team answered the call.
Their first challenge was a scarcity of data on aircraft behaviour past the point of stall that could be used to build their model. To overcome this challenge, Grant and his team worked with Bombardier Aerospace to source data from wind-tunnel testing, and from the test flights aircraft manufacturers conduct to have their aircraft certified. These flights are flown by trained test pilots and push aircraft outside of their typical flight envelopes, including stall, to ensure the aircraft post-stall behaviour is manageable.
Once they had developed a representative model, they extrapolated two alternate versions: one that simulated a version of aerodynamic stall where the effects came on more quickly and affected the plane more severely than the average predicted by the data, and one that minimized those effects. The goal was to determine whether their representative model would be sufficient for training pilots past stall.
Grant and his team then recruited a pool of 15 pilots — all volunteers from Air Canada, Jazz and WestJet — and divided them into three groups. Five pilots were trained on the most representative model, five were tested on the more sensitive version, and five on the minimized version. All were trained to recover from four different stalls in the UTIAS flight simulator. After training, all 15 were tested on the most representative version to study how well their respective training regimens had prepared them.
“We found no statistically significant difference between the three groups,” says Grant. “Once we had trained them on what to look for, and how to respond, all 15 were equally capable of performing under stall conditions — this suggests representative modelling is sufficient for full-stall recovery training.”
The FAA requires updated upset recovery training for pilots to roll out starting in 2019, and new simulations for stall recovery will be incorporated in these new programs.
“Per mile flown, air travel is incredibly safe,” says Grant. “But as demand for commercial pilots continues to increase, we need to keep engineering improved simulations that allow for better training.”