It's better to be safe than sorry. Behind this common saying, which is easier said than done, hides a major big data usage strategy, which is starting to make its way in MRO. Air France Industries KLM Engineering & Maintenance (AFI KLM E&M) has jumped in with its own predictive analysis tool, called Prognos. Developed in the context of the MRO Lab innovations programme, it must enable the wear on various elements to be better identified and understood, to schedule their replacement before they break down and so avoid costly, unplanned maintenance operations. The company has rolled out two variants: one for airframe, Prognos for aircraft and the other for engines, Prognos for engine.
"As an airline and a maintenance company, we have both flight and workshop data", explains Rodolphe Parisot, AFI KLM E&M's Digital Innovation director. "We have developed concrete tools to use it, which is not true for everyone. Prognos lets us drill deep down into the systems and sub-systems, which is quite unique on the market. "
More precisely, the Airbus A380, for which certain elements were subject to breakdowns, launched this approach. AFI KLM E&M was aware of the super jumbo's capacity to generate much more data than previous generation aircraft, and so its teams looked for a way to increase visibility in operating and maintaining aircraft. A structure dedicated to big data was therefore created at AFI. It enabled the completely in-house development of this predictive analysis solution. A first proof of concept (PoC) was produced in 2015, then industrialisation started for the project at the beginning of 2016.
Using big data
Data is transmitted as soon as the aircraft lands at the airport. It is transmitted over Wi-Fi or 4G connection depending on the type of aircraft. Engines benefit from real-time data transmission thanks to the Acars messages which are sent to maintenance centres. They are then enhanced with additional information collected on the ground (when the aircraft type allows). This in-flight transmission is not yet on the agenda for Pognos for aircraft, but is part of the plans for the coming years. The increase in the number of onboard connectivity tools could speed things up.
This raw information is collected, cleaned up, processed and analysed to build up a structured database. This enables the changes in a component to be monitored throughout its operational lifetime and the variations to be observed. With the use of large series of data, it is then possible to consolidate this component's parameters with those of the rest of the fleet and use algorithms to see if there is any correlation. These models may be refined according to the flight conditions, the type of rotation made, the destinations served and so on.
"Initially, we carried out enhanced troubleshooting", explains David Vazquez, Data & Connectivity project manager at AFI KLM E&M. "This enabled us to identify the source of the breakdown more quickly, particularly by distinguishing between the failure of a sensor and the failure of a system. ". AFI KLM E&M has been able gradually to establish models able to detect the advance warning signs that an element is about to break down. This provides the possibility of carrying out predictive analysis and anticipating repair operations by integrating them into programmed maintenance phases.
"Prognos enables us to detect a failure between 10 and 20 flights before the breakdown actually occurs. This gives us about a week to change the element", says a happy David Vazquez. He also highlights how reliable the prediction is: "All the elements replaced for preventive reasons were confirmed as being at fault by the manufacturer ". He also quotes the example of a fuel pump group on A380. Of the four pumps of the same group, only one is qualified as critical for flight. When Pognos detected a failure on it, the maintenance teams decided to swap it with a non-critical pump. The pump in question did actually break down, but thanks to the swap the plane avoided being grounded in Singapore for two days, the time it takes to send out the spare part.
Extending the scope
New aircraft and systems are being integrated gradually into Prognos for aircraft, according to needs and requests. Before the system can start being used for a particular plane or piece of equipment, a first data collection phase is needed to build up a sufficient baseline and be able to deduce the first models from it. This takes around two to three months to generate effective algorithms, which will then be improved through constant enhancement of the database.
The first elements to be processed in Prognos were the fuel systems, with their 28 sub-systems, of Air France's A380s. The portfolio was then expanded with the integration of the front and main gear. KLM then joined the project with the Boeing 747-400.
At the moment around ten systems - which each have around 10 to 20 sub-systems, are monitored on the A380. On the 787, AFI KLM E&M has set up Prognos on a first system, and others should follow. Work is just beginning on the A350, but deployment will be easier thanks to a system architecture which is close to that of the A380. This is also true for the 777 in relation to the 787. AFI KLM E&M is also targeting the A320 and A330, but the planes have different data structures. The speed of these successive integrations will depend mainly on the number of operators who opt for Prognos: the larger the working base, the quicker the models will be to build.
For now, only around thirty planes at Air France and KLM benefit from Prognos. This figure could increase with the marketing of the solution to other operators than the two parent airlines. For engines, Prognos for engine is quickly becoming part of the landscape. AFI KLM E&M has 1 500 engines in its portfolio. It must be said that the solution is compatible with all types of engine and fleet.