Project R-15954

Vision AI for Real-Time Assembly Quality Inspection of Large / Complex Parts (VARY_IRVA) (Research)

The inspection of large products (> 1m) in an assembly environment is today largely manual. An operator goes over a checklist, typically at the end of the line, to visually inspect the product. This process is not only time-consuming, but it is also prone to human errors and interpretation of what is a defect. Furthermore, there is no visual proof of the (non)quality. Finally, the inspection is often done at the end of the line whereas the detection of defects & miss-assembled parts in-line could prevent expensive repairs. The goal of this project is to largely automate this entire process including iterating over the various inspection steps, using vision sensors to register the inspection, and finally automatically detecting potential defects. While tactile sensing research is out of scope in this project, existing tactile data can be used where available in combination with vision data. The focus of the project will not only be on end-of-line inspection but also on inspection in the line. Defects that are targeted are assembly errors, surface defects, and deviations from nominal assemblies. These defects include, loose fasteners (that may lead to leakage), improper parts fit, missing or incorrect components, over or under application of adhesive, incorrect orientation, improper installation of electrical / mechanical connections, surface quality issues, etc. We aim to keep the configuration time for a new product under 1 day and for a new product variant under 1 hour. The strategy of the project is to make the inspection in assembly lines highly flexible, to deal with large variability of assemblies and their variants (models, options), as well as to make the inspection design fast by virtualizing most of the inspection design, including robotization and vision, and using the 3D CAD file of the part to be inspected. The project will provide a high TRL Vision AI solution for assembly quality checks for Large & Complex Assemblies (Figure 1), which pushes the state of the arts, providing so far only solutions limited in assemblies' complexity and size.

01 September 2025 - 31 August 2027