How AI Developments Could Aid Engine Inspections

Artificial intelligence tools and programs are beginning to make their mark on the commercial aftermarket as more shops collaborate with specialists in the field to find ways to utilize the technology.

For engine maintenance, MRO providers are tapping into artificial intelligence (AI) tools to aid tasks such as inspections. Yet harnessing the data effectively and improving the overall quality of what is produced remain near-term obstacles.

Three-year-old AI startup Amygda, based in Derby, England, is working to tap large quantities of data generated through engines as well as some of their components such as line replaceable units, CEO and co-founder Faizan Patankar said at Aviation Week Network’s Aero-Engines Europe in Amsterdam in September.

“We look to help some of the people closest to those engines and parts, such as the maintenance engineers and technicians, make sense of that data,” Patankar said. “AI is one technology that’s helping us bring that reliability and efficiency faster to the people who most need it.”

Discussing applications to engine maintenance, Patankar described AI as the “parent category.” “Underneath that we have machine learning,” he said. “But then again we have machine learning on different systems.” Given the multiplying data volumes used in aviation, he said engines are increasingly producing “unstructured information.”

“These are subjects like, what is in your maintenance manuals and [enterprise resource planning] systems?” he said. “Bringing that data together, whether that’s text or numbers, and making valuable insights from it is one set.”

Another startup is Amsterdam-based Aiir Innovations. CEO and co-founder Bart Vredebregt said the company has worked with MROs, lessors and OEMs for several years with a remit to “get AI into the hands of as many people doing engine inspections as possible.” Admitting the concept is so broad that it often leads to confusion about its application, Vredebregt said Aiir’s software has become more commonly used in aviation businesses in recent years.

For engine maintenance, Vredebregt identified one effective MRO use case with computer vision, which enables analysis of images and videos to identify and learn objects and people. “We will look at something and try to see what’s happening in the image for a task like borescope inspections,” he said. “It’s dividing cracks or dents into really something easy to understand visually.”

When rolling out AI in an MRO shop, Vredebregt said, some studies have shown maintenance-related findings increase significantly. “Part of that is we make it easier to report, so that explains maybe some of the increases,” he noted, adding that some of the findings were likely overlooked by suppliers or other third parties.

Arjan de Jong, principal of maintenance and engineering at the Netherlands aerospace center (NLR), said the research institute is using AI platforms for aerospace vehicles, operations and systems. De Jong said the NLR sees AI as something that can “perceive, reason and act.”

De Jong noted that the NLR is heavily involved in projects related to robotics and robotic inspections. “In that case, we like to perceive what the condition of a particular part of an aircraft is, reason what is wrong with the part and the anomalies, and then act and give advice about what to do with the part based on the findings,” he explained.

De Jong cited the ability of AI to reason as one of its specific advantages. “This makes it also possible to automate certain tasks and even make systems autonomous,” he said. “That is an interesting concept because it can also mean that you can do work by machine and that you can leave certain work to the machine, meaning improvement in working conditions for the mechanic, for example.”

Perfect Fit

Wouter Kalfsbeek, unit leader of big data engineering at AFI KLM E&M, said that since he began overseeing AI utilization across the unit 10 years ago, his team has grown from a handful of people to around 40. “The level at which we are using AI today is quite extensive,” he said. “In the area of maintenance, it is used for predictions—an area where you can use a lot of AI and also for optimization where we try to make the best use of our assets.”

Kalfsbeek described the maintenance industry as a perfect fit for AI predictions. “You have to deal with an enormous number of uncertainties and unknowns,” he said, adding that drawing on historical data to start and then real-time data to make better predictions will ultimately lead to improved efficiencies.

One of the major predictive tools AFI KLM E&M has developed in the past 10 years is its Prognos platform for aircraft and engines. “Data is taken from the engines and the systems, and we try to translate that into the understanding of the health of these actual systems and then try to predict how much time that they’ll have left until they run into failure,” Kalfsbeek said. “Ten years ago we would have thought this impossible, but now we’re doing it on an almost daily basis.”

Kalfsbeek sees the number of use cases for AI rising and expects this to be even more dramatic over the coming years. Still, he considers relying on data over physical inspections or validations—to the point of using it to extend or even relieve maintenance inspection intervals and initiate maintenance actions—as further off. “That’s the Holy Grail, but we’re not there yet,” he said.

Kalfsbeek expressed concern about the quality of the data available. “This is a problem when working with AI, but it depends on what is being worked on,” he said. “If you’re taking data from sensors on an engine or an aircraft, these sensors are fantastic measuring systems and will ensure high-quality data. But compared to data collected from maintenance processes or transactions moving around the world from different owners—for example, from a lessor to an airline and then to a repair station—the data quality in these instances is often not as good.

“AI can help brush up the data quality, but most of that can be lost,” Kalfsbeek continued. “So if a user doesn’t invest in the right measurement system for this, it’s still going to be quite hard to get the right quality of data to perform AI.” 

New technology such as blockchain is providing some solutions, he added. AFI KLM E&M uses blockchain to collect data over the life cycles of aircraft components and engine parts, enabling it to validate the data from other parties.

De Jong of the NLR also believes that innovation in the sector will continue to accelerate. “If we look at the scientific competitions that are going on in AI and their technologies, then it’s amazing, and there’s a lot more to come,” he said. However, he identified one hurdle in MRO inspections with robotics.

“When we build a robot that’s going to inspect the aircraft part, we need to make sure that’s done properly,” he said. “One of the things that we need to provide in the process is that the observation and the diagnosis are indeed correct. We have to prove that a robot can do the same work as a human inspector technician.”

In the long term, Aiir Innovations’ Vredebregt believes the key to improving the quality of AI systems is to keep training them with more data inputs to make them smarter.

Broader concerns regarding AI “hallucinations” have been exacerbated over the past two years, and those naturally have trickled into the MRO segment.

“The word 'hallucination’ comes from the fact that people can use ChatGPT, which sometimes gives answers that are made up, hence the word 'hallucination’ is now everyday language,” Patankar said. “Having looked at some of the big predictive maintenance platforms, there are hundreds of people who manually review what an alert or prediction is saying before it gets into an organization.

“ChatGPT democratized AI into the hands of the end users,” he noted. “And aerospace is now in that phase where the thought needs to be the people using these tools will be the end users—whether that’s engineers or technicians. Prediction errors have existed for years, but you just haven’t seen them because you’ve been shielded.”