A student's loss is research's gain, as a jet crash launches one investigator on a quest to improve aviation safety and communications

Two years ago, a student came to computer engineering professor Krishna M. Kavi and spoke of losing friends on EgyptAir Flight 990, the New York-to-Cairo flight that was lost at sea in October 1999. U.S. investigators believe a suicidal co-pilot deliberately plunged the aircraft into the Atlantic; Egypt blames defective plane parts. But what the student wanted to know was why in-flight data, now stored on an aircraft's "black box" recorder and retrieved in the event of a crash, couldn't be transmitted to the ground and monitored in real time, thus providing the pilot and crew with an early warning system.

With the help of a National Science Foundation grant, Kavi - then in the Computer Engineering at the University of Alabama, Huntsville, now at University of North Texas - looked into his student's query. What he found was that, while such transmission was technologically feasible, early warning would probably be of little use, except in cases of engine stalls or hard landings. "Things happen so rapidly, like Flight 587, which took less than a minute, there's not much you can do," said Kavi, referring to the American Airlines plane that went down Nov. 12 over Queens, killing all 260 aboard and five on the ground. Kavi did develop tantalizing ideas about mining historical flight data to improve the accident investigation process and Federal Aviation Administration safety recommendations, but his government partner, the National Transportation Safety Board (NTSB), wasn't interested. First, maintenance scheduling was the airlines' worry, not theirs. And they couldn't spare the manpower to teach Kavi's team the processes they use to investigate accidents, Kavi remembered.

That was then, this is now. Earlier this year, a Navy contractor came to Kavi, interested in using his ideas on data-mining and machine learning for developing the next generation flight data recorder for future military aircraft. What really caught the contractor's eye was Kavi's idea of a "hijack button," to transfer control of an aircraft out of the cockpit to some remote site or automated system in case of a dire emergency. Since Sept. 11, when terrorists crashed commercial airliners into the World Trade Center and the Pentagon, the idea has been gaining ground, with no less a champion than President George W. Bush endorsing it in is first major post-911 speech on airline safety. "Real-time transmission of data, so you don't need to search for the flight data recorder, would work, the recorder is just a tradition," Kavi says. "The NTSB believes in the flight data recorder, they don't believe real-time transmission can be as reliable as black-boxes.

"The problem is most of those people are using old technology. They have a lot of knowledge; some of them can look at raw data, and with all their experience, say on the spot, 'This is the cause, or the pilot was left-handed'. We wanted to capture all that knowledge, if for nothing else but to train new investigators."

The problem of low government interest will not come as a surprise to many digital government researchers, who have learned the hard way that many IT departments think innovation comes out of the end of a box marked "Microsoft". Difficulty getting government agencies to assign staffers with both the time and commitment to work on joint projects has also been a challenge.

When the real-time monitoring project did not pan out, Kavi's research focus changed to developing an expert system and knowledge base to aid accident investigations. Kavi says the work indirectly led to new research directions on software architecture and extended UML for the design of intelligent agent-based systems; a PHD dissertation and journal publication are pending.

Kavi further proposed taking the NTSB's database containing 138,000 records of accident investigations since 1996, and developing an intelligent online searchable database. Another goal was to develop templates and processes to standardize and enhance the investigative process, and visualization tools to help investigators analyze information from the flight data recorders. Using machine learning approaches, Kavi thought he could mine the database and discover patterns leading to aircraft failures due to pilot error, weather, etc. The accident investigation process is extremely collaborative, involving 10 to 20 teams, each engaged in its own specialized activities and in producing a separate report. At present, these teams communicate only by phone, or at the end of the investigative process, leading to wholesale rethinking and report revisions in the eleventh hour.

Kavi believed he could create an intelligent assistant that could enable investigators to share partial report documents and data as they worked, rather than afterwards. Another idea Kavi had was to post NTSB flight data, and the tools investigators use to analyze it, on the Web for the public to see. "This would eliminate those conspiracy theories about accidents," he says.

"We believe that this project not only addresses the immediate needs of NTSB but also will lead to more effective and efficient ways of investigating accidents, design of more sophisticated FDRs, as well as new research on learning technologies, visualization tools, search engines and text-mining," Kavi wrote in a subsequent research grant proposal. "Enhancements derived from these technologies can contribute to reinvented, efficient and economical services provided by NTSB,FAA and other related agencies."

The proposal went nowhere, but Kavi expects to resume talks with the Navy and a major defense contractor in early 2002. "The timing is better this year, after the terrorism," says Kavi. "Now even President Bush is talking about a panic button."