|
|
 |
|
Christopher Johnson with the University of Glasgow |
|
Date: September 10, 2003
Time: 2:00pm
Location: NIA, Rm 404
Speaker: Christopher Johnson with the University of Glasgow
Subject:* "Computational Methods to Find Patterns of Human 'Error' and System 'Failure' in Mishap Reports"
Additional Information:Presentation (PPT)
Human `error' plays a significant role in causes and consequences of mishaps. This has resulted in the introduction of a number of training techniques that improve the `human factor' in aviation. For example, JAR OPS sub-part N, 1.945(a)(10) and 1.955(b)(6) and 1.965(e) requires that operators provide Crew Resource Management (CRM) training to flightcrews. CRM training helps personnel to avoid mishaps by introducing the problems of communication, situation awareness, team-based decision making etc that lead to operator `error'. The benefits of Crew Resource Management training have been validated by simulator studies. Crew performance can be shown to improve in simulated scenarios after thay have received this form of training. However, these simulator studies are controversial. Crews have been shown to react quite differently under the everyday pressures of commercial aviation. This talk, therefore, describes a range of novel computational techniques that have been used to identify whether CRM training is effective in altering crew behaviour. I have extended the US Navy's NaCoDAE conversational case based reasoning system with a probabilistic information retrieval tool to identify patterns in thousands of aviation incident reports published by several different countries. An important finding of my work is that there is no inverse correlation between CRM recency and the likelihood of involvement in CRM related incidents. In other words, crews that have recently gone through Crew Resource Management training are just as likely to be involved in CRM related incidents. The talk will go on to demonstrate the generic nature of the software by identifying a recent pattern of software and hardware failures in US telemedical applications. I will end the talk by identifying the limitations of the software that emmerged when it was used to analyse incidents involving the US military.
This work won a 'best paper' award at this year's International Systems Safety Society Conference in Ottawa.
|
|
|
Home