Rolls-Royce: Optimising jet engine maintenance with machine learning

Rolls-Royce is looking to machine learning to optimise how it maintains over 13,000 jet engines currently flying around the world

Improvements to jet engines have greatly improved aircraft range, efficiency and safety[1]. The traditional drivers for this improvement have been predominately related to design: optimised components, new materials and innovative manufacturing techniques. Machine learning offers jet engine manufactures like Rolls-Royce a new avenue to improve cost and engine performance by optimising the maintenance of engines in service.

 

Rolls-Royce sells a majority of engines on service contracts instead of an upfront purchase price. Under these contracts, the operator pays Rolls-Royce per hour of flying time and as a function of the engine’s efficiency. Therefore, operators and Rolls-Royces’ incentives are aligned to keep engines on wing as long as possible and at maximum efficiency. Engine efficiency degrades with age and component wear though, so the challenge is deciding when to take engines off-wing to perform maintenance, and which components to replace or repair. With 13,000 engines in service[2], small improvements to engine maintenance schedules would significantly boost earnings.

 

Currently, engineers monitor the level of wear and engine performance with regular inspections and real-time data. Judgements on when to take engines off-wing for maintenance are based on engineering models and previous experience. Although these methods have improved over time, machine learning offers a step change. This is because it can use a wealth of new data being made available from real-time sensors and 3D scanning technologies. In addition, newer engine models can use algorithms developed from older models, shortcutting the lengthy period to build service maintenance data.

 

To develop advanced machine learning algorithms requires Rolls-Royce to build appropriate datasets, and organisational capability.

 

The current CEO, Warren East, is looking to bolster Rolls-Royce’s digital capabilities through internal and external changes. The digital organisation has been strengthened as part of an internal restructuring. A partnership with Microsoft[3] in 2016 is helping to develop more advanced engine analytical tools. Numerous programs have been launched in tandem with partner research universities to build machine learning algorithms.

 

To improve engine monitoring, newer engine models are equipped with more real-time sensors that gather more data and relay it directly in real-time to Rolls-Royce. In addition, data from older engine models is being gathered and processed to further boost the engine performance datasets. This has been a challenge for older engines with more service history but data that is stored in legacy systems and structures. Training data is being further bolstered by Rolls-Royce’s growing 3D printing capabilities. Many types of damaged components can be manufactured quickly and cheaply. Specific datasets can then be generated by testing the components on rigs.

 

The EJ200 (powering the UK’s Eurofighter Typhoon) is proving to be a “goldilocks” engine to experiment with maintenance schedules tailored to individual engines. The engine first flew in 2003[4], and is equipped with cutting edge data gathering equipment that has generated a significant amount of in-service history data since. Comparisons in individual engine performance to historical data allows engineers to adjust maintenance to meet what the engine needs. To improve their judgement, engineers are working to marry the engine performance data with inspection data gathered from components. New 3D scanning inspection techniques are being used to build accurate dimension information, at low cost, on complex components (like aerofoils) that can help train machine learning algorithms to more accurately determine which components need replacing. These efforts are laying the groundwork to develop and roll out machine learning generated maintenance schedules.

 

There is a risk that these efforts are being pursued adjacent to other parts of the organisation and so wont maximise the benefits. System and component design teams are still composed of traditional aerospace engineers in silos separated from the digital capability teams, hindering the exchange of ideas. This is important as analysis from the machine learning algorithms could be used to drive the design of components and systems to maximise their lifetime value. Communication and the exchange of ideas between these teams should be improved through community of practices, joint teams and colocation where possible.

 

Another opportunity is for algorithms to provide feedback to the operator. Decisions on flight routings can impact engine degradation and performance. By combining flight data with engine performance and maintenance data, machine learning algorithms could be trained to analyse proposed flights. Operators would then be able to optimise flight routes to reduce fuel burn and engine degradation.

 

Rolls-Royce has succeeded over the past 100 years by building cutting edge hardware. Developing machine learning will require new skills. Can the organisation develop the digital capability to fully capitalise on this mega-trend? Should Rolls-Royce partner with external organisations?

[1] Airinsight.com, “How about an A330neo”, https://airinsight.com/how-about-an-a330neo-3/, accessed November 2018

[2] Rolls-Royce, “Annual report 2017”, https://www.rolls-royce.com/~/media/Files/R/Rolls-Royce/documents/annual-report/2017/2017-full-annual-report.pdf, accessed November 2018

[3] Rolls-Royce, “Rolls-Royce takes TotalCare digital with Microsoft and Singapore Airlines”, https://www.rolls-royce.com/media/press-releases/2016/11-07-2016-rr-takes-totalcare-digital-with-microsoft-and-singapore-airlines.aspx, accessed November 2018

[4] Rolls-Royce plc website, “EJ200”, https://www.rolls-royce.com/products-and-services/defence-aerospace/combat-jets/ej200.aspx#/, accessed November 2018

 

Word count: 793

Previous:

“What do you mean, Doctor?!”

Next:

Block by Block: Harnessing Open Innovation at The LEGO Group

3 thoughts on “Rolls-Royce: Optimising jet engine maintenance with machine learning

  1. Chris, thanks for sharing this look into how data analysis is driving an improvement both in efficiency and profit for both operators and Rolls Royce themselves. Having spent nearly 1800 hours held aloft by jet turbines (the Rolls Royce Adour is near and dear to my heart), I can tell you that operators also put great stock in the safety and reliability of their motors. Monitoring engine performance in near real time helps both operators and manufacturers to detect the degradation of an engine and pull that engine off the wing before subtle vibrations can lead to fatigue or a catastrophic blade failure. These advancements drive not only the excellent safety record of modern engines, but also improve the economics of long-haul overwater flying for airlines, ultimately allowing consumers to fly for less money and reduce their carbon footprint by flying aboard more efficient two-engine jets rather than less efficiency four engine jets. Thanks for a look into this high-tech world!

  2. Will the type of data being collected change over time? Will the existing fleet need to be equipped with updated sensors? Challenges of having a separate digital team – will this slow adoption across other product-driven (rather than tehcnology-driven businesses?). Given alignment of incentives with airlines, any additional data that can be collected from the aircraft, pilot or airline?

    Thank you Chris for sharing – very interesting to see how a manufacturing-based business is leveraging machine learning, a trend I originally thought was generally more limited to technology and software businesses. I found the point you raised on digital teams being separated from product teams particularly interesting. Given lack of information sharing, you effectively point out that this can slow adoption and innovation around how ML can be leveraged to extend duration between maintenance cycles. An interesting question comes to mind – given the organizational structure of manufacturing and product-based businesses, how easily can ML be adopted and innovated? It seems there are more barriers to adoption than in a typical software business. Perhaps the answer is in outsourcing, but this of course can be limiting, and the decision of whether or not to build these capabilities internally is especially interesting as you point out at the end of the essay.

    One other question – is Rolls Royce using ML in its manufacturing processes? In a world where IoT is becoming commonplace in manufacturing facilities and where sensors are increasingly abundant, would be curious if these capabilities are being used at the plant level to monitor equipment maintenance to maximize uptime.

  3. Chris

    Thanks for sharing a very interesting use case for machine learning that is actually being applied in real life today compared to so many of the machine learning stories we read about that are more “marketing fairytales” of benefits to come in the future. It is easy to see how machine learning can significantly improve maintenance schedules, and I wonder if machine learning can’t also be used to identify real time flight conditions and over time analyse how weather or other aspects could be analysed to even more precisely predict/model maintenance schedules.

    I believe the biggest challenge for RR to reap the benefits of machine learning lies in your question reagirding RR’s actual ability to adapt, develop and apply new cutting age software and data analytics innovations. Having seen first hand the challenge of attempting to transition a hardware based company to a software driven approach (although admittedly from an investor perspective rather than engineer…), I believe RR would be well served to start this journey in partnership with an external party experienced in machine learning technology and application. Done correctly, I wonder if such a partner could potentially significantly speed up the adoption curve for RR in machine learning while they develop their own internal systems and teams to spread the learning across the organisation. Regardless of the approach RR chooses to take going forward, I have no doubt that effectively adopting machine learning in their jetty engine development and maintenance process will be crucial for RR to remain at the cutting edge of jet engine development.

Leave a comment