For many years, aircraft engine manufacturers have had access to inflight performance data, being warned of actual or impending failures. This has even allowed them to take pre-emptive action, having engineers with the correct spare parts waiting at the arrival gate. This was a consequence of powerplants being the most heavily instrumented systems on the aircraft, as well as the most flight critical. Unfortunately, many other components and systems on the aircraft were passive, unable to communicate their status as they were never considered to be important enough to justify the investment required, or that their failure would generate major problems such as delays, diversions and cancellations.
This changed with the latest generation of e-enabled aircraft, such as the Airbus A350 and Boeing 787, with many new components and systems having been designed from scratch to be able to record their performance. In addition, developments in IT and telecoms made it much easier to transmit and analyze the data. As a result, not only is there a better awareness of more faults as they happen, huge amounts of routine data are generated from every flight, which can now be downloaded after landing and made available to OEMs, MROs and airline departments.
Of course, this is a massive exercise and, with every flight, the pool of information, or data lake, gets deeper. For example, the Skywise open data platform from Airbus, which was launched in 2017, had accumulated 12 petabytes of data by August last year, the latest date for which information is available because of the pandemic. At that point, 130 airlines had signed up to Skywise, with more than 9,000 aircraft in operation. As well as the airframer, there were also more than 10 suppliers involved along with four certified partners; 15,000 internal and 2,000 external users; and 700 data analysts trained by Airbus.
Several other open data platforms have since been launched, all with the aim of providing a neutral space in which data can be analyzed. This because the lake is now so deep that it is impossible for a single airline to navigate solo. Indeed, the trick is to convert raw data into useful information that has a direct effect on operations. That means each airline needs a program specifically tailored to its own unique operating environment as well as the assistance of outside specialists.
Many components and systems have fixed service intervals, usually defined by flight hours or cycles. Often, performance will gradually deteriorate with use. Using existing technical records, thresholds for each item can be established that trigger an alarm when a fault is likely to occur. A decision can then be made whether to remove the item prematurely, with the expectation that repair will cheaper than replacement after failure. Hence the term ‘predictive maintenance’.
While this sounds great, it is not straightforward. If an airline has a power by the hour contract, with fixed monthly payments, is it reasonable to expect a discount or refund if repair costs are reduced? An even bigger issue is that, to derive maximum advantage from the data lake, input really needs to come from across the worldwide fleet. This could show general failure trends for components as well as regional variations caused by climatic conditions, for example, or allow an airline to benchmark itself against industry averages. The platform builders always say the data remains the property of the airline and that it is completely anonymized when incorporated for wider analysis, but cut throat competition means some operators are always nervous about giving something away. Something not given away, of course, is the data processing, which is a subscription service.
A good example is provided by Etihad Aviation Group, which was not only an early adopter of Skywise but assisted Airbus in its development, having started work on prognostics, data analytics and text mining algorithms in 2012, using the Intelligent Operations service from Taleris, a joint venture technology company between Accenture and GE Aviation. In 2013, it started working closely with Boeing using Airplane Health Management Gen3 Prognostics on the 777, focusing on ATA Chapters 21 (Air Conditioning), 30 (Ice & Rain Protection) and 36 (Pneumatic). These reflect sandy conditions in its home in Abu Dhabi, where a local university has helped with machine learning, data analytics and text mining. The Group has also worked with other industrial partners.
Bernhard Randerath, vice president Design, Engineering and Innovation, Etihad Aviation Group, says the aim has been to develop simple and verifiable monitoring algorithms, with failures being predicted 500 flight hours in advance. Condition monitoring should be available online and offline and adaptable to aircraft configuration changes. The number of new and existing sensors should be low and not only limited to the aircraft domain — passenger preferences/profiles and improved cabin reconfiguration have also been under study. This is typical of data mining, as airlines suddenly recognize the potential for other applications. After all, high value passengers are just as likely to be annoyed by a blank monitor as a delay caused by an engine problem.
Etihad defined six steps for nominating, isolating and predicting failures:
Step 1 – Choose for the right maintenance strategies
This is divided into three sections:
Improvement: reliability driven and includes modification, retrofit, redesign and change orders
Preventative: divided between equipment driven (self-scheduled, machine cued, control limits, when deficient and as required); predictive (statistical analysis, trends, vibration monitoring, tribology, thermography, ultrasonics and NDT); and time driven (periodic, fixed intervals, hard time limits, specific time)
Corrective: event driven and includes breakdowns, emergencies, remedial, repairs and rebuilds.
Step 2 – Choose the right relation between cost and value
In order of ascending value creation, this involves primitive (fix it when it breaks), preventative (preform time-based tasks), predictive (collect data, assess condition, repair as needed) and proactive (solve root cause of chronic problems)
Step 3 – Integrate operational data and isolate real problem makers
This can use general statistics, pilot reports, component removal reports and shop reports. This has been augmented by a dedicated reliability report, which better assists in identifying chronic problems.
Step 4 – How are predictions integrated in the maintenance process?
This involves breaking down the work orders costs that are included in the maintenance budget (reactive, periodic and non-periodic) and those that are excluded (production support, capital projects, expense projects and R&D/product testing/demonstrations)
Step 5 – Process and train in the right way
This includes condition monitoring and condition prediction processes. The condition prediction process has now been updated with certification information (temperature, HALT and HASS) as well as human factors. The result is more accurate predictive information in the case of operation in hot temperature conditions, like the home base.
Step 6 – Understand failures and integrate correction codes
This uses correction codes to achieve a Flat Local maximum and introduces local search algorithms with Hill Climbing functions.
This should produce an end-to-end intelligence platform, that is an autonomous data analytics system for prediction validation. This can be displayed on a dashboard tailored for use by the various departments in the airline, with MRO functions such as planning and electronic task cards having been added recently, although overall progress has been slowed by COVID-19 restrictions.
Another early adopter and developer of Skywise was easyJet. It has long experience in this area, having started manual entry trend engine monitoring in 1990. In the 2000s, this switched to using ACARS. From 2015, it worked closely with Airbus to identify the top 100 technical issues affecting its operations as part of early Skywise development while 2016 saw the start of a project to analyze three years of data to try and spot trends. Flight trials in that year with equipment on 85 aircraft focused on three specific technical issues, with 14 impending failures being successfully identified.
Despite all this work, it took a rather different approach from Etihad as predictive maintenance is integrated into its Operational Resilience Program, a suite of solutions that are used to keep day to day operations running smoothly and when there are problems. For example, schedule design essentially puts the right sized aircraft at the right airport at the right time to match demand. Making sure the first wave of flights departs on time makes it easier to protect the schedule if something comes up later in the day. If this happens, there are revenue, customer and crew consequences that have to be resolved. That means the predictive analytics suite needs to anticipate weather, ATC, crew, aircraft and airport challenges so personnel can accurately assess schedule, aircraft, crew, customer, airport and cost impacts in response. Some of these other solutions include the Amadeus SkySYM flight network simulation solution, produced by Optym in partnership with Amadeus; a crew pairings analyzer; standby aircraft tracker and optimizer; and a claims forecaster.
As the industry begins to recover, any cost efficiencies that can be generated will be useful and predictive maintenance will play an important part.