Introducing AVIA
Acubed’s Advanced Digital Design and Manufacturing team (ADAM) is leveraging artificial intelligence to optimize aviation and aerospace manufacturing. The approach, known as Advanced Visual Intelligence for Assembly, or AVIA, is the latest effort to utilize machine learning solutions – in this case, computer vision – to better assess and improve processes, quality and safety in commercial aircraft production.
Rising Demand
Notwithstanding lingering COVID supply chain delays and existing material shortages, air travel is rebounding, with experts projecting a continued rise. In the face of these challenges, innovative AI models like Acubed’s AVIA can support teams' abilities to safely and reliably increase production to meet rising demands.
Computer Vision
Utilizing computer vision – collecting and interpreting information from images and video – AVIA gathers data on manual processes as they happen, enabling teams to better understand how they are progressing in the production schedule. Despite innovations in the design and manufacture of aircraft, an aircraft production line relies on manual work by skilled production workers – processes like installing a wiring harness, for example – at certain stages of production. These manual processes require workers to manually report their activities. With computer vision, this process can be automatically and accurately captured so workers never have to report back to the computer. While some systems exist to report finished tasks, the resulting data can be inconsistent. Unlike systems simply addressing task completion or supply chain visibility, AVIA reliably informs teams of their current production state and, ultimately, what they can do if there are delays.
For a given process, AVIA is designed to operate based on relatively straightforward inputs. Prompts like “pick up bracket” or “turn screwdriver” as well as trained understanding of the relevant component and tool enable AVIA to examine the action on the assembly line and determine what manual process is being performed. Having identified the process, teams can easily verify the work performed aligns with quality standards based on factors such as the steps taken and time to completion. With this information, teams can focus on their work knowing that their progress is being captured and verified by the system; if there are delays, they will better understand how to resolve them, resulting in a more flexible, resilient production system. Once applied in a production setting, AVIA can also enable teams to compare production data to planning data in order to determine whether planning assumptions were accurate or if they need to be updated. For example, if over time AVIA observes that a process consistently takes longer than designers planned, teams can adjust their expectations and make decisions that are ultimately more accommodating to production line workers.
AI Evolution
The aviation and aerospace industry has been exploring computer and machine vision in a variety of applications – and Wayfinder has been instrumental in leveraging machine vision to make autonomy a commercial flight reality – but this is a first in the production space. As AI has become increasingly more sophisticated, systems like AVIA now require fewer or simpler backend inputs. This level of automation – AVIA’s capability to understand and interpret straightforward inputs – eases the burden on engineers, ultimately making it easier to apply the model in a production setting.
Current projections anticipate AVIA improving manual task efficiency by several minutes per worker per day or more. Across all Airbus’ production lines, that could result in tens of millions of euros saved yearly.
Continued Development
The team is working full speed to ensure the quality of AVIA’s detection, and more cameras may be necessary to improve the system’s ability to detect the breadth of activities occurring on Airbus’ production lines, but the build-out has undoubtedly been a success. Acubed is planning to test AVIA in our Alabama-based Final Assembly Line in a representative production environment by the middle of this year, and expand the test cases throughout 2025.