AI AND THE ART OF AIRCRAFT MAINTENANCE

It’s 6:20 p.m. and you finally got to the front of the line. You’ve been at the airport for over four hours waiting on a plane to take you home, and it’s your turn to ask a series of  questions to a kind, attentive customer service agent who has been on the receiving end of an hour of verbal abuse. Unfortunately, you’ll likely gain no new insights into your situation. It’s the tail end of a long trip through multiple airports, and half a dozen different flights. You have an early presentation in the morning to brief the executive team on the meetings of the last week. You’re tired, hungry, and ready to get home.

The plane that was supposed to take you home never made it. You overheard the airline crew talking about the minor maintenance issue which kept the plane grounded—a fault code which is a quick fix, but requires a very specific replacement part. Why is such a minor issue causing such major problems?

Flight delays are the bane of every traveler. However, as planes have essentially become flying computers and maintenance manuals have grown to the size of baby elephants, mechanics struggle to service aircraft within the 15- to 20-minute window allotted. There’s a lot at stake in that time period: A small team of mechanics must diagnose a complex system (an Airbus 380 has approximately 4 million parts produced by over 1,500 companies) to determine if passengers must be re-routed or a part must be sent for in-depth analysis. The pressure of servicing aircraft within a tight schedule is a major challenge faced by carriers who are focused on safety first while maintaining customer expectations of speed and efficiency.

Chad Kartchner, Director of Connected Aircraft Strategy at Honeywell, is looking for solutions to the maintenance challenge. The multi-billion dollar aerospace division of the Fortune 100 conglomerate has a long history in the industry, employing around 40,000 people and keeping a strong focus on maintenance software. “Who isn’t a Honeywell customer?” he explained. “Until I joined the company, I wasn’t aware that every time I flew on a plane, something on that plane was probably manufactured and supported by Honeywell Aerospace.”

Kartchner’s dedication to innovation is what attracted him to Honeywell. As a teen, he spent a few formative years in Taiwan, where he was exposed to engineering as a profession. He completed his undergraduate degree in Electrical Engineering at BYU, then moved on to graduate studies, earning an MBA from Kellogg School of Management. “I made a conscious decision to try to focus on where I felt like I could add the most value, which were the things that I was passionate about,” said Kartchner. “And for me, it was more about the process of integrating disparate pieces into an entire system.”

This led him to the growing services division of the Honeywell. “We invest a lot of our time now looking at what our customers’ experiences are and how we can make them better,” Kartchner explained. “We can look at the end-to-end process, and use that as a driver for innovation and growth.”

THE MAINTENANCE DILEMMA

Aircraft maintenance proved to be an area particularly ripe for improvement. In one of his early exposures to air- craft maintenance, Kartchner learned about an approach sometimes called “shotgun maintenance.” An experienced field engineer detailed this process to Kartchner: A mechanic has 30 minutes to fix an issue. After running diagnostic tests, a mechanic may make a judgment call and swap out the two or three parts most likely to have caused the problem, and hope for the best. This means a lot of wasted time and labor replacing parts that are perfectly fine. While it’s possible to send the parts back to Honeywell to test for faults, the parts often stay at the hangar while a director follows the plane’s status to see if the problem was fixed.

Kartchner spent a lot of time at hangars visiting with mechanics and directors of maintenance to understand this process. One suggested approach was to provide more training for the mechanics to diagnose the problem with higher accuracy. However, the complicated reality of the systems makes it difficult for mechanics to sufficiently analyze every possible issue.

Instead, inspired by IBM Watson’s ability to help a doctor make a diagnosis, he began to explore the benefits of augmented systems. Rather than relying upon a mechanic’s memory, a system with perfect recall would be able  to answer specific questions to diagnose a problem and provide just-in-time training if necessary.

THE AI TOOLBOX

Natural language processing (NLP) is the key to Honeywell’s success. Honeywell has partnered with SparkCognition to use the company’s DeepNLP® solution for centralizing maintenance information. Somewhat like a computer’s search function on steroids, DeepNLP can ingest massive quantities of data and consolidate it into a database that is is searchable using natural language—which means the user can have a more-or-less complete dialogue with the program. DeepNLP provides answers, not just links to keywords, and gives confidence scores as to the accuracy of those answers.

“SparkCognition has been a great partner,” added Kartchner. “I’ve been able to frame up some use cases and some problems, and asked them to go tackle those technology challenges.”

One significant advantage of NLP technology is the ability to preserve expertise in different forms—maintenance logs, handwritten notes, and data in tables. This aids maintenance across an enterprise, because it passes on “tribal knowledge” from shop books or other nontraditional sources. Operators can learn from others who have faced a similar issue.

“Once we create a system that is user-friendly enough that people want to use it, it’s generally better,” Kartchner adds. “I believe it will create this virtuous cycle, and be able to collect additional data that we can leverage to create even more value for the end user.” He specifically points to repair reports as a potentially rich source of information for planning purposes.

As the industry continues to innovate, smarter planes are also helping schedule maintenance. Onboard systems log fault codes that were captured during the flight, and transmit that information to a Honeywell Global Data Center, which packages this as an email sent to a director of maintenance. This enables mechanics to be better prepared when the plane lands, both in terms of knowing what to expect (and doing additional research if needed) and planning the maintenance schedule.

In the future, more analysis can be done by an augmented system before an email even gets to the mechanic, who will know exactly what’s wrong with the plane and know how to fix it. This means less time grounded by maintenance issues.

THE BUSINESS ADVANTAGE

“AI is more expensive than paper manuals,” Kartchner pointed out. “This technology does come at a price. But we are in the right place for trying this out is because of what’s at stake.” He references a recent EU law that requires passengers to be reimbursed for delays, which cost one airline $25M in six years. 

Kartchner believes AI blends case-based and model-based problem-solving approaches to create the most effective maintenance system. “We are able to take the official records and create models from things like maintenance manuals and have a starting point that is really good,” he explains. “Then we can augment those model-based approaches with knowledge that’s gained over time. Using AI, we can make the system question and answer and back and forth, so you end up with not only the blending of the two approaches but the ability to go off- script and ask for additional information and get that information quickly.”

“For this to work, it’s all about scale,” Kartchner elaborates.“ We have to be able to automate the process of ingesting all of that information and learn quickly. The fact that we are doing that with great accuracy and not a lot of manual training is a tribute to the technology really getting to the place where what we are trying to do is feasible.”

THE FUTURE OF FLYING

Kartchner doesn’t see the future of maintenance in fancy gadgets— though he does think there’s potential in hands-free augmented reality. He’s more excited about enhancing user experience, like enabling dialogue with a computer to get assistance: “Being able to have a case and talk to a computer almost like a coach, and that coach has the ability to recall everything about that aircraft,  and not only recall it but make smart recommendations based on all of that information.”

The ultimate goal, for Kartchner? “It would be really cool to be flying somewhere and seeing a mechanic in a vest on a plane with the technology that we have created, and not only just using it but enjoying it.”

As Kartchner and Honeywell continue to innovate and push the boundaries of possibility by applying AI technology to aviation maintenance issues, they’re working toward making the grounded plane a thing of the past, and getting you home on schedule.

Last modified: February 16, 2018

Leave a Reply

Your email address will not be published. Required fields are marked *