Building Human Profile by Aggregation of Activities Application to Aeronautics Safety

摘要

The work related here is devoted to the setup of a methodology regarding the study of polyvalent objects about which our knowledge is incomplete. It is concerned with the analysis and characterization of flights/flight maneuvers, considered from the standpoint of the involved human operators. The two following issues have been identified: 1) incompleteness, which comes from the second-hand nature of the recorded data that describes and situates the pilot's activity, and 2) variability of human sensations and reactions, as a result of which identical stimulations may cause different reactions and different observations may correspond to identical sensations/situations. Our aim is not to close up on the theoretical mechanisms of perception and preference but, based on these mechanisms, to obtain a behavioural model that will be used 1) to characterize observed patterns amongst the various recorded data, 2) to anticipate the patterns to be observed and to relate them to particular flight conditions. We introduce the three-step process of supervised aggregation, an aggregation driven by experts and expertise, which we successfully put into practice in the case of elementary turns. This process was developed aiming to convey characterizing and predictive power, notwithstanding the incompleteness and variability of observable data.