Aviation Communication: A study of communication problems for EL2 pilots in the Australian Aviation Environment

Experimental setup Flight simulator at UNSW, with XPlane 6 for a Cessna 172. The pilots were given instructions for each flight containing a weather forecast and a flight plan. The audio stimuli were played through a headset and the voice responses recorded with Audacity. Recording of the pilot’s actions through XPlane software on one computer. Audio recordings of the pilot’s responses through Audacity software on another computer. Stimuli Stimuli were designed for 8 flights, with 2 versions for 4 different flights, as follows: • Rate of speech o Flight 1 A: Slow (w/ pauses) / Flight 1 B: Fast (no pauses) • Amount of information o Flight 2 A: Low Info (< 3 items per com) / Flight 2 B: High Info (> 4 items) • Workload o Flight 3 A: Low workload (quiet Nav) / Flight 4 B: High workload (Nav w/ fuel recalculation) • Environment o Flight 4 A: Clear environment (1 AC in CCT) / Flight 3 B: Congested environment (>5 AC in CCT) Flight plans and weather forecasts were constructed for each flight pair. Each flight was about 8 mns, giving 15 opportunities for error (every 30 seconds) per flight. The audio stimuli were designed by Dominique Estival and recorded by David Maddock, Chief Flying Instructor at Gostner Aviation. The individual recorded stimuli were concatenated into a single file for each flight, with time between the stimuli for the pilot/participant to respond and react. The order in which the 8 flights were presented to the pilots was determined by a Latin Square design. Participants A total of 18 pilots from the Sydney Basin participated between November 2012 and March 2013. The first 13 participants were given $50 for travel expenses, the last 5 were given $50 book vouchers. Some pilots found the experiments too challenging, with the first pilot expressing frustration and another pilot abandoning the task. In total, 17 pilots completed the tasks for a total of 136 flights. Data analysis The analysis of the data is two-fold: 1) Audio: a spreadsheet provides the expected answers for each of the stimuli in the 8 flights. Each call, or transmission, consists of a number of items, e.g. Heading, Altitude, Call sign, which must be read back accurately by the pilot. Some variants are acceptable (e.g. “(Maintain) 3500”). At the call level, calls are coded as either correct or incorrect and the results present the number of correct/incorrect calls per flight relative to the number of expected calls in that flight. At the item level, the number of missing (i.e. not read back) and incorrect items (wrong information, i.e. 3500 instead of 4500) are counted for each call. The results present the total number of missing and incorrect items for the flight. At a lower level of analysis, we could report the number of correct/missing/wrong items for each call. In a later study, we will further categorise the correct/missing/wrong items according to the type of information they convey, e.g. heading, altitude, radio frequency, call sign, etc. 2) Flight responses: automatic analysis of the altitude, heading and position at each point in flight; comparison with expected behaviour for that stage of the flight. Due to technical issues, audio recordings for 6 flights were either lost or incomplete, with 85 possible pilot transmissions missing out of a total of 2,142 transmission opportunities; the XPlane output for one pilot is missing. Access to the data is mediated by the authors Dominique Estival ORCID 0000-0002-6178-3825 or Brett Molesworth UNSW [email protected].