The goal of this PhD project is to investigate existing computer-vision-based quality inspection methods for textile yarns, as well as sheet metal parts, and to improve on them in innovative ways. This project is in collaboration with VIA University College (Herning, Denmark) and the Department of Materials and Production at Aalborg University.
In the subject of textile yarn inspection, the objective is to visually assess the quality of yarns that are spun from recycled materials/fibers. To accomplish this, a system for re-winding yarn from one bobbin to another needs to be acquired or developed. With this system, color images of the yarn can be taken at specified intervals. These images will allow for the identification of yarn characteristics, such as yarn evenness and hairiness, as well as of surface defects such as neps. Currently, the suggested approach for this project is the development of a publicly available yarn image dataset and an investigation of potential deep-learning approaches that utilize the dataset to achieve SOTA (State-Of-The-Art) results.
In the subject of sheet metal part inspection, the objective of this project is to develop a flexible method that automatically assesses the degree of similarity between manufactured sheet metal parts and the CAD models that were used to produce them. By scanning a given part and acquiring a point-cloud data structure that represents it, it is possible to determine whether its manufacturing features, such as holes, curves, and edges, are within the allowed tolerances that are established during its design. The suggested approach for this project is the development of a tailored approach for automatically detecting such features specifically in sheet metal parts. Generalization to other types of parts is encouraged, but not the goal.
This PhD project is funded by Innovation Fund Denmark