Artificial Intelligence Techniques for Enhanced Sorting

The Ph.D aims to establishing a robust effective sorting technique in the AUSOM project investigating how geometrical information from cameras and composition dat a derived from LIBS sensors can be combined to improve both, the classification, the localization of objects and the control of the sorting of scrap metal. The technical feasibility of integrating computer vision based on deep learning and data fusion technologies to distinguish different alloys of metal scrap, such as different aluminum and stainless-steel alloys, manganese, copper, brass and lead, will be investigated. Also, techniques will be developed that exploit both classification and geometric data to enable selective removal of components (e.g. via controlled pneumatic ejection). The Ph.D. thesis will contribute to the development of an automated sorting technology for metal alloys based on laser-induced breakdown spectroscopy (LIBS) and computer vision. The research goals are: • To apply computer vision based on deep learning to replace and even improve the sorting of non-ferrous metals that are sorted based on visual inspection. • To increase the capacity of a point and shoot LIBS based system by reducing the number of elements to be analyzed by only analyzing by LIBS these metal parts for which the confidence of the computer vision result is unacceptably low. • To create a frame for integrating computer vision and LIBS data to improve the accuracy of the classification. • To improve the control for a multiway pneumatic ejection which takes the geometric and material properties of the object into account to control the ejection system.