Data-Driven Safety Feedback as Part of Debrief for General Aviation Pilots
General Aviation (GA) is the foundation of most
flying activities and the training ground for civilian pilots, both
recreational and professional. However, the safety record for GA is lacking
compared to that of commercial aviation. Approximately 75% of accidents each
year involve personnel factors, that is, even if the pilot was not the cause of
the accident, they could have done something to either prevent it or improve
In this research, I aim to improve GA safety through safety-driven post-flight debrief that encourages pilots to consider the risk in their flights and identify behavioral changes that could make their flying safer. Providing pilots with a debrief tool that they can use with or without a flight instructor requires that we know both what to communicate, and how to communicate it. Risk communication heuristics and biases have not been researched in the context of aviation and flight training and we therefore do not know how pilots understand or respond to debrief.
To achieve the goals of this work, I used a three-step process: (1) identify events that may put the safe outcome of a flight at risk, (2) detect those events in flight data, and (3) inform the pilot in a way that helps them improve in their future flights. I use a state-based representation of historical aviation accidents to define a list of events or behaviors that need to be communicated to the pilots, in the form of states and triggers. I use flight data to retrospectively detect these behaviors upon completion of the flight, by mapping parameters or combinations of parameters that can be calculated and tracked in the flight data to the hazardous states and triggers defined. To present these events to pilots, I created a prototype interactive debrief tool with risk information that I use in a survey to evaluate the effectiveness of feedback in different representation formats. Specifically, I evaluate the impact of three factors: representation method (graphical and numerical), parameter type (safety and performance parameters), and framing language (risk-centric and safety-centric).
I disseminated the survey via aviation mailing lists, type groups, flying clubs, and flight training providers, end received 268 responses. The survey analysis showed that the feedback representation does affect its effectiveness in terms of risk perception, but not when it comes to pilots’ motivation to change. The lessons learnt from this survey can be used in creating additional surveys that delve further into risk communication biases and our understanding of how pilots perceive risk and feedback.