So Finnair sent FCE data for the entire fleet for six months and did no bleed-system repairs so as not to ruin the experiment. FCE software spotted problems in the pneumatic system that Airman missed. “We changed it and the problem vanished,” says Skytta.
Often the real challenge in doing predictive maintenance, Skytta notes, is spotting real problems without too many false alerts. “The problem with OEM systems at the moment is you get too many warnings. So the troubleshooter does not trust them. And a solution is no good if no one uses it.”
Lack of data is usually not a problem: “We have thousands of parameters on Airbuses, and airlines have found ways to collect it,” Skytta notes. Carriers already use this data for flight-monitoring, as required by regulations, and some data also could be used for predictive maintenance. However, “pilot agreements might prevent sharing data for other purposes,” Skytta acknowledges. “We do not have a culture of sharing data.” That is unfortunate because the best predictions would be based on multiple-airline data.
Skytta says different data formatting is usually not a big issue, noting that Airbus manages to collect the data it needs mostly in standard format. Airman and other OEM tools can be expensive, especially for small operators, and Skytta believes consultants such as FCE might help them.
Skytta says FCE's statistical approach can look for abnormalities in any system for which there is data, including mechanical, hydro-mechanical and electro-mechanical systems, even generators of electrical power. “It would not work for electronic systems. We do not have metrics on the circuit board,” he says.
FCE has been using its Anomaly Detection Software since 2004, mostly in the power and rail industries. Aviation Project Manager Daniel Jaroszewski says the technique requires no knowledge of underlying physical processes but lots of data as it looks at the correlations among all signals from the system under study.
The aim is to minimize both false alarms and missed critical events. The software can “learn” without supervision by spotting data patterns in a mostly healthy operation of a component or system, as it did for Finnair. The longer an anomaly persists, the more likely it will turn into a critical event. If data on malfunctions is available, the software can “learn” with supervision.
Jaroszewski thinks his software outperformed Airman in spotting several bleed-valve problems because it looked at correlations of all the data, not just one signal. He says the approach is applicable to many aviation components, so long as data is plentiful, as it increasingly is with new aircraft. Apparently others agree: FCE has been talking to several aviation companies, including Lufthansa Technik, in recent months.
Other consultants are also active. “No one has all the tools,” stresses Vijitha Kaduwela, CEO of Kavi Associates. He says good analytical tools are available from SAS, IBM, SAP and others. “But you cannot just buy off the shelf and go; you must customize it.”
That is what Kavi consultants, drawn from United Airlines and GE, do. Kaduwela believes there is great value to be gained in spotting chronically defective units, minimizing aircraft downtime, assessing supplier quality and finding the “bad apples” among aircraft. “There are lots of opportunities to find the outliers and fix them,” he says.