An award-winning professor at MIT is working to develop a system to spot accident trends in airline flying before the accident happens. MIT’s John Hansman along with colleagues at MIT and in Spain are working on a data analysis detection tool that uses cluster analysis, which is a form of data mining that breaks flights down into series of common patterns and then looks for anomalies in those patterns. Once those outliers are flagged, analysts can further study the data to see if the unusual data is of any real concern.
In a limited form, such analysis is already in use, through Flight Operations Quality Assurance (FOQA) programs in which certain airlines participate. Through FOQA, the airlines gather data, generally mechanical position and performance data, to spot potential problems so they can be prevented before they cause an accident.
MIT’s plan goes beyond that, as it encompasses many more parameters than the 88 that the airlines monitor through FOQA.
Testing was done on Boeing 777s being flown by an international operator, says MIT, that is out of business. While you’d think that the testing, which took place over the course of a month on more than 300 flights, would have yielded few results, you’d be wrong. Analysis found one pilot who regularly took off with reduced thrust and identified one landing in which the airplane’s flaps were misconfigured, requiring more power than normal and resulting in a too-low approach.
Hansman is working with several airlines to get access to more data to test, though he says that the process can be very tricky, as union agreements regulate the dissemination of flight data. Still, he is confident that accommodations can be reached that will allow his team to do further analysis toward a model that will help reduce accidents in a day and age when airline mishaps are rare and increasingly difficult to spot ahead of time.
