One BARC team is analyzing big data to help standardize aircraft brackets. Modern aircraft have thousands of custom brackets. “Brackets are everywhere, from critical structural components to those used to guide conduits and other electric and airflow systems throughout the aircraft,” says Steve Brunton, professor of mechanical engineering and adjunct professor of applied mathematics and computer science at the University of Washington.

Brunton, doctoral student Emily Clark, her co-advisor Nathan Kutz and Boeing project engineer Angelie Vincent spent three years developing and deploying algorithms to sift through data from Boeing and its subcontractors and identify brackets that are similar enough to be standardized. Brunton says that defining a similarity metric to reveal how close two brackets are in design and functionality was surprisingly challenging. While the results are proprietary, Brunton says their work demonstrates that many brackets can be standardized, which will reduce manufacturing and inventory costs and streamline the design process.

Another machine learning research team developed an automatic process for detecting tow boundaries in parts fabricated with automated fiber placement (AFP), which is crucial to ensuring acceptable structural properties. Currently, most composite aircraft parts, including fuselage sections, fairings and empennage components, are visually inspected with the aid of laser projection. Manual tow boundary inspections are extremely time-consuming. For instance, it might take several days to inspect an entire composite wing skin. Ashis Banerjee, associate professor of industrial and systems engineering and mechanical engineering, says that for a part with 100,000 tow ends, even state-of-the-art semi-automated inspection requires three to six hours of additional manual inspection.

Banerjee is leading a BARC team that’s using machine learning to develop an automated tow boundary detection system that would minimize the need for extra manual inspection. Collaborators include two doctoral students at the university, Wei Guo and Ekta Samani, and Agnes Blom-Schieber, technical fellow at Boeing Commercial Airplanes – Product Development/Structures Division. For two years, the researchers gathered thousands of tow boundary inspection images, but they found it difficult to generate a sufficiently large training dataset for their machine learning model.

“While we knew that the inspection images would be challenging to analyze, we were still surprised at the extent of the challenge due to variations in illuminating conditions, material processing parameters and ply geometry,” says Banerjee. “In fact, it was virtually impossible for non-experts like us to discern the tow boundaries visually in many images. Even traditional edge detection-based image processing methods failed miserably.”