
Andrew StarrProfessor of Maintenance Systems, Cranfield University
“Automated health assessment for railways – challenges in robotic plant”
Europe’s railways include civil structures which have been in use for almost 200 years, and are inspected for continued use or repair. They are very expensive to replace. A vast number of structures and systems are past their original design life, but in daily use. Inspection of infrastructure is as old as the railway, because some parts and features degrade quickly. The track itself is robust, but the precise geometry for fast moving trains remains a challenge. Inspection tasks include the rails; fixings; isolation pads; sleepers or bearers; ballast; the underpinning earthworks or formation; and drainage. System issues such as geometry, access and site containment (fences, trees) are an on-going issue. For most of history, these have been manual tasks conducted outdoors, in all weathers, and often at night.
There is a strong desire to reduce co-workers’ exposure to the hazards on the railway. Mechanisation has progressed since the 1940s, but true automation is quite new. Automatic track geometry recording is mature; video inspection generates a lot of data, but is less mature for high quality inspection. The mechanical handling of heavy tools accelerates the maintenance work, but usually requires the railway to be closed. Fully autonomous devices have the prospect to operate independently at night, when there is more track availability.
The robotic vehicles, and physical technologies for inspection and repair, pose interesting challenges. Sensing and tooling may be robust but will continue to develop rapidly. Location estimation e.g. GPS poses systemic challenges. Software poses obsolescence issues
Thinking of the future, we can foresee benefits in safety, skills and productivity which will encourage further development of expansion of automation. A long term plan in systems design must recognise the potential for disruptive technology in the maintenance of transport systems.

Darren MacerSenior Technical Fellow, Predictive Maintenance and Health Management, Boeing
“Prognostics and health management in aviation”
Data has become the lifeblood to the industry and we continually work to enable, aircraft, their systems, and their components to provide increasingly more data which has ultimately allowed us to provide increasingly accurate and meaningful insights to operators, OEM’s, regulators and service providers. This has been further enhanced by the ability to combine this aircraft data with other data sources about the production, operation and maintenance of the aircraft to generate and maintain operational digital twins.
As Machine Learning and Artificial Intelligence, combined with high power computing, have grown from areas of study and niche fields to everyday aspects of aviation operation. This has allowed predictive maintenance and health management to become keystones of support and services, and is gaining regulatory approval and informing new designs. As we continue along this journey, there is increasing focus on obtaining feedback of how predictive models are performing, how accurate the predictions are, and the actions taken as a result of predictive alerts, this focus is, in part, driving how the industry thinks about feedback loops and data sharing. This requires collaboration across the industry between Airlines, Airframe and Component OEM’s, Suppliers, MRO’s, and parts providers to ensure that as a discipline, actionable insights can be achieved at an even greater success rate.
This session will share how Boeing is helping maintenance move from reactive, through prognostics, to a predictive approach where unscheduled maintenance is being turned into scheduled maintenance and where maintenance is completed based on the condition of a component, system, or aircraft and not fixed schedules based on flight hours or cycles.

Jerome LacailleSafran Emeritus Expert in Algorithms, Safran Aircraft Engines
“Mathematical approaches in artificial intelligence for the digital clone of an aircraft engine”
The presentation begins by the introduction of Safran Aircraft Engines company, CFM engines and our initial work in PHM. It continues with the different approaches that lead us today to offer artificial intelligence solutions based on recurrent neural networks and generative models. These artificial intelligence models make it possible to anticipate maintenance operations and simulate with great precision the continuous measurements obtained during flights. It thus becomes possible to compare engines by observing their behavior on virtual operations.
This work is carried out in part by our DataLab founded in 2015, stimulated by a small group of doctoral students. Their research allows us to offer the company modern statistical elements that break with usual technologies.