MRO AI Startups Encourage Data Sharing And Realistic Expectations

Aftermarket-focused artificial intelligence startups in Europe say their technology is helping MRO providers and airlines achieve significant efficiencies and savings, but challenges remain in data availability and public perception.

During the Aviation Week Network’s MRO Europe conference in Barcelona in October, several artificial intelligence (AI) companies discussed how their technologies have benefited early adopters.

Stargazr’s AI-powered software is focused on helping companies understand their financial data better, pinpoint where and why delays are happening and optimize their margins.

“What we have built is basically a simulation model that helps MRO companies understand what kind of impact their operations have on their financials,” co-founder and CEO Rafi Wadan said. For instance, when operational or financial leaders want to analyze where they are losing money, Stargazr’s software co-pilot could provide answers such as, “You have a problem with your overtime, or there are certain departments where you’re not performing well, and this has X effects on your profit and loss,” he says.

Stargazr customers include Lufthansa Technik and Hawker Pacific Aerospace. The company says its software reduces turnaround times by 20% and improves profits by 11% through smarter financial insights.

Amsterdam-based Aiir Innovations is using AI to improve the speed, accuracy and consistency of engine borescope inspections (AW&ST Jan. 13-26). Aiir says its software can analyze borescope video 75% faster than manual reviews , and it is working with such customers as KLM, MTU Maintenance and TAP Air Portugal.

“We have already prevented some very costly quality escapes . . . that would have cost north of $1 million if [the software] wouldn’t have found those defects,” Aiir Innovations co-founder and Chief Technology Officer Miriam Huijser said.

Huijser added that human inspectors tend to report 93% more findings with an AI assistant than without, and that has led to more comprehensive reporting, tracking and planning for defects. For instance, she said one customer has been able to cut inspection and reporting shifts to two from three, reducing turnaround time by 30%.

On the supply chain side, Dublin-based Armac Systems is using AI to help airlines optimize inventory and maximize spare parts availability. Armac is working with Cathay Pacific Airways, Finnair and Iberia Maintenance, and says that Iberia has reduced costs by 24% with the AI-powered software.

According to Armac Chief Experience Officer Mark Stacey, a major challenge the company encounters is sparse data in MRO supply chains.

“We have fast-changing supply chains,” Stacey said. “Particularly in the MRO space, we also have situations where the upcoming task cards that are being asked for are not always available, so we’ve actually worked with a customer to build a machine learning model that can do task card prediction.”

Stacey noted that the model uses historic data to predict task cards with about 75% accuracy, enabling companies to optimize inventory ahead of those checks.

It is tricky to forecast ordering for parts that might be needed only a few times a year, so Armac is using several techniques, such as long short-term memory modeling, to “make small gains and accuracies,” Stacey said. “We’re looking at improvement rather than perfection because we think that by making small gains and accuracies, we can improve the overall value delivered to industry optimizations and reduce cost,” he added.

Some companies are hesitant to jump headfirst into AI because they do not think they have enough data volume or quality, but Stargazr “figured out very early that you don’t have to have the perfect dataset,” Wadan noted.

“Sometimes it’s OK if you have a 50-60% optimized dataset that you can start with,” he said. “We always say improvements in data quality sometimes go hand in hand with visibility, so it’s always a chicken-and-egg problem. You can start with a narrow approach and then basically go from there. AI can help us connect all the data points.”

Huijser cautioned that users should not expect immediate perfection from AI systems, either. "Our biggest challenge right now is the hype that surrounds AI and the risk aversion in the market,” she said. “Because of this hype, people have the feeling that AI is kind of like magic and it should be able to do anything perfectly all the time. On the other hand, you have this disillusionment when AI systems cannot delivery constantly on that kind of expectation of perfection, and I think that’s a pity, because an AI system definitely does not have to be perfect to be implemented safely.”

Huijser pointed to a 2022 study by the University of Canterbury in Christchurch, New Zealand, that concluded humans found only about 68% of all engine defects during an inspection (Inside MRO October 2023). “An AI system paired with a human needs to be better than the alternative without the AI system,” she said. “If it adds value that way and if it can be implemented safely, . . . it’s pretty valuable.”

Stacey asserted that AI models can be improved through more industry cooperation on data sharing. “When we’re dealing with sparse data, the models can be improved by cooperating with other data points outside of that,” he said. “One of the things that we will be really interested in is working with our customers to agree on some sort of data pooling. Not in terms of sharing data directly—we don’t need to do that—but I think these collaborative models show potential for even better outcomes.”