Warning This site is not recommended for Internet Explorer browsers. Please use another web browser to get a better experience.

Machine Learning: key to optimizing aircraft manufacturing processes


During the lengthy process between the design of a plane and the end of its lifetime, several obstacles arise. Worn parts, grounded planes and passenger safety, among other things, cause major financial, economic and human problems. At Safran Engineering Services, machine learning is at the center of major progress that optimizes product lifetime management and equipment reliability.

Machine Learning: an opportunity for the aerospace industry

Machine learning has become an invaluable component of business productivity, and data is its fuel. Machine learning, or “automatic learning”, refers to concrete ways to use artificial intelligence in manufacturing. A program is implemented that enables automatic learning on a computer or machine, in order to carry out specific, highly complex operations. An algorithm generates the learning system, and the machine learns by automatically incorporating the outcome of these operations, in process of continuous enhancement. Sylvain Tanguy, Data Science Leader and Marc Lutrot, Manager of Data & Prognostics Health Monitoring at Safran Engineering Services explain the value chain for the use of this technology.

What’s happening right now in aerospace is very interesting. In “4.0” factories, machines and tools are now connected and processes are increasingly computer-based. They generate data that are major assets, from which we create algorithms which will enable the machine to identify models, enhance them and make optimized predictions. In flight, data production is something that’s always been around, but gathering and using this data is an expanding field. New aircraft programs are designed to be broadly implemented. Most of the parts have sensors and all data is collected, aggregated and used, mainly through machine learning, to better determine what the plane needs throughout its lifetime. This makes it possible to monitor all onboard equipment and considerably enrich the total data collected, which then leads to many opportunities,” explained Marc Lutrot.

So, machine learning not only helps increase the lifetime of aircraft parts, but also anticipates breakdowns and reduces operation interruptions. This technology is applied in three areas of manufacturing processes: sizing, production and operation monitoring.

Optimal Sizing

During aircraft design, sizing is the step where the best possible physical dimensions are determined for each part. “Today, traditional modelling math tools rely on a mainly theoretical approach. Since theory is not reality, and in order to ensure maximum safety, we take into account many conservative hypotheses, capable of covering any margin of error. This leads to some aircraft parts being oversized. Thanks to machine learning and the real data gathered on the device, we can create learning models that help us get closer to a practical reality. Of course, this method cannot replace classic methods. It adds to them, thereby ensuring better sizing and an improved life expectancy for the plane,” clarified Sylvain Tanguy. In this area, Safran Engineering Services is working, for example, on creating algorithms to optimize plane loads, starting with the fuselage design phase.

Production Process Monitoring

In addition to qualitative improvement of produced parts, production monitoring and supervision improves understanding of the part’s lifetime by combining its data with operation data, which also impacts the plane’s lifetime. “We are working mainly on digitizing our design plans. We extract digital and textual data from our paper plans through image recognition technology based on a neuron network. From this, we create digital databases which help us better monitor the product, anticipate breakdowns, analyze correlations, find root causes and more,” explained Sylvain Tanguy.

Predictive Maintenance Services

Finally, the operation cycle of a plane is punctuated by regular inspections, for checking the condition of onboard parts. But a grounded aircraft that leads to canceled flights is very expensive for the airline. Therefore, there is a major economic component. “The added value of this new technology is its ability to predict these breakdowns, enable delivery of parts by anticipating or processing breakdowns in advance, before they take place, thereby reducing operation costs,” added Marc Lutrot. That’s what we call predictive maintenance. “Today, Safran Engineering Services is succeeding in, for example, anticipating breakdowns in landing gear by using flight data,” he continued.

Still in the prospective stage, comprehensive monitoring — from production to operation — for each plane is yet to be achieved, but there is no doubt that it is an important short/mid-term goal for Safran Engineering Services. In this new 4.0 world where algorithms self-develop, produce increasing complexity and seem to surpass human skill, we could fear that people will end up relying too much on machines. This is not the case. Currently, machine learning still remains a decision-making tool for industry experts.