Lufthansa Technik: Keeping airlines flying optimally with AI-powered TechOps platform AVIATAR

Keeping global airlines flying high is a valuable business. Aircraft maintenance, repair, and overhaul (MRO) services are expected to be worth $56.6 billion globally by 2031. Unscheduled downtime is also costly, totaling around $62 billion annually prior to the pandemic. Additional grounded flights caused by COVID-19, coupled with a 60% drop in passenger demand, resulted in $371 billion in lost revenues for airlines in 2020.

The financial rewards for companies that can cut down costs and deliver more efficient and effective MR0 services are rapidly increasing, however, and will likely help to drive the industry’s post-pandemic recovery. That’s particularly true of the predictive maintenance sector, where advancements in technology are fueling a booming industry anticipated to be worth around $23.5 billion by 2024. And that’s to say nothing of the value to the airline industry of getting people and goods to their destinations reliably and on time.

Lufthansa Technik is a global leader in this arena through its flagship AVIATAR, an independent platform that manages the technical operations of fleets of aircraft. The digital solutions it employs combine aviation expertise with data-driven insights and engineering know-how to analyze the entire MRO value chain. In the case of predictive maintenance, this empowers airlines to switch from the purely reactive interventions of the past to adopting proactive measures that address component failure, improve operations, and lead to more effective management of resources. But the AVIATAR team recognized the leadership of Google Cloud analytics and infrastructure capabilities. The experts from Google Cloud and Lufthansa Technik teamed up to solve not only the technical challenges but to enable AVIATAR to be set up for future demands and recovery after COVID-19.

“Our product is aircraft availability,” explains Jan Stoevesand, Senior Director Digital Solutions Data and Analytics at Lufthansa Technik. “If an aircraft is not available, or it's delayed for a technical reason, this is a pain for passengers. That's why we are dedicated to reliable aircraft availability; the better the availability, the better the passenger experience. When it came to overhauling our analytics stack to meet the changing needs of airlines, we needed a reliable platform that was able to provide a secure, scalable, cost-efficient, real-time, event-driven architecture. For us, that was Google Cloud.”

Migrating to Google Cloud during COVID-19

AVIATAR examines huge amounts of data, including from the sensor logs of thousands of aircraft, to make recommendations and provide insights. It also analyzes data from maintenance shops and flight plans, examining factors such as where specific planes have flown from and to, and for how long. It uses this data to enable airlines to turn unscheduled maintenance issues into scheduled events, enabling efficient advanced planning for the preparation of replacement parts and alternative aircraft.

Lufthansa Technik previously used a self-managed platform, based on virtual machines to power this process, but found a number of drawbacks to this approach. That included a high investment in keeping the architecture running, both financially and in terms of working hours, as well as issues with stability, reliability, and scalability. The company made the decision to migrate to Google Cloud and started developing the project in Summer 2020. By January 2021, the team was ready to proceed and had completed the migration by the end of that month, all without any downtime or disruption to customers.

“That was vital to us,” adds Jasper Pult, Principal Architect of the AVIATAR Analytics Platform. “We already had many customers on the platform, and we needed to get all of their jobs across with zero downtime, all at the height of COVID-19. Everything else we were working on had been scaled down during the pandemic, but we knew that there were so many advantages to this platform, for us and our customers, that we decided to go ahead.”

Building a new serverless pipeline for more efficient operations

When it came to adopting new Google Cloud solutions, the AVIATAR team was keen to replace as many existing operations as possible with serverless managed services. The main components for its productivity pipeline now include Google Kubernetes Engine, Cloud Run, and Pub/Sub. Data enters the pipeline and automatically triggers ETL (extract, transform, and load) processes, as well as event-based jobs in Cloud Run. These activities then trigger data modeling instances running on Kubernetes clusters.

The amount of data entering the pipeline, numbering in the millions of data points each day, combined with the demand to reprocess large amounts of historical data (e.g. for model training) requires the highest flexibility in resource allocation and scalability, for both technical and economical reasons. Google Cloud has brought with it a number of key performance improvements. Of particular importance is a reduction in the time it takes between receiving data to delivering initial insights, thanks to the improved stability and on-demand scalability of the new pipeline configuration.

“Under the old setup, we had batch loads that were scheduled every few hours,” explains Pult. “Now what we have is completely event based, in near real time. It means we get updated predictions in the frontend within minutes of receiving the data. That means our development cycles are now shorter, and we can pass on the benefits to our customers.”

The new Google Cloud infrastructure has also led to significant economic efficiencies at Lufthansa Technik, reducing the infrastructure costs of its AVIATAR analytics platform by roughly half. Along with that, it also brings the benefit of greater financial transparency, something that was particularly important to Stoevesand. Now he can examine the price of scaling up—and the savings of scaling down—particular research and production tasks, as well as querying the financial implications of bringing on new customers, training additional data models, and taking specific pricing decisions.

Drilling down into the data for more effective insights

In addition to the pipeline, aircraft engineers are also using Notebooks and BigQuery to drill further down into the data before approaching the team’s data scientists to find a model to fit potential problems they’ve uncovered. The data team then works with various methods, such as data flow and platform training, to develop their model further before handing it over to the next team of data engineers who enable the data model to run in the team’s production architecture.

“Engineers typically spot the first signs of a problem and can say ‘this is something we should take a closer look at,’” Pult adds. “Notebooks enable them to establish correlations and prove their suspicions, before talking to the data engineers and data scientists. They can discover new things which they might not have otherwise found.”

One of the advantages of this approach has been to bring people together to work in a highly joined-up, interdisciplinary way. The whole team now has access to the same data environment, building a common understanding of what’s required between everyone involved. As a result of this, productivity within the team has also improved.

“The KPI that’s important to me is the time it takes to move from research to production. I wanted a single environment to improve this,” says Stoevesand. “There’s no difference now between the language and tools we use to approach a particular situation, which is a major step for us. It’s all about efficiency. It’s about being faster and it’s about being better at what we do.”

Machine learning models are also a big part of the research phase of the AVIATAR pipeline, and a collaborative approach informs this area of work. Lufthansa Technik uses Google’s AI platform to train its data models, and it has equipped data scientists and aircraft engineers with Notebooks, to experiment with, develop, and deploy machine learning models into production. “That was a big win for us,” adds Pult. “We’ve given engineers the tools they need to interact with the data, and get their hands dirty with it, so to speak, which they didn’t have before.”

Building back better after the pandemic

Lufthansa Technik has seen big changes in recent years, adapting to a global pandemic, and strives to emerge stronger as a result. It will build on the foundations it has laid with Google Cloud to focus more on product strategy. “We're finally where we wanted to be, with a stack that's able to grow with us,” Stoevesand concludes. “Google Cloud helped us create an infrastructure that allows us to scale fast, bring in new customers, and adapt more flexibly. Our people can now focus on their work, without having to spend their time keeping the technology running. They’re able to focus on the problem and give the customer first insights within days.”