An Intelligent Visual Analysis Scheme for Automatic Disassembly Processes in the Recycling Industry

von Erenus Yildiz
Dissertation
Datum der mündl. Prüfung:2021-04-20
Erschienen:2021-04-30
Betreuer:Prof. Dr. Florentin Wörgötter
Gutachter:Prof. Dr. Jens Grabowski
Gutachter:Prof. Dr. Ramin Yahyapour
Gutachter:Prof. Dr. Carsten Damm
Gutachter:Prof. Dr. Babette Dellen
Gutachter:Prof. Dr. Alexander Ecker
Zum Verlinken/Zitieren: http://dx.doi.org/10.53846/goediss-8573

 

 

Dateien

Name:Thesis.pdf
Size:45.6Mb
Format:PDF
Description:An Intelligent Visual Analysis Scheme for Automatic Disassembly Processes in the Recycling Industry

Lizenzbestimmungen:

Zusammenfassung

Englisch

This thesis aims to develop and deploy a visually intelligent disassembly scheme to automate the recycling routines for end-of-life products. The recent developments in artificial intelligence and computer vision are yet to be utilized in the E-Waste recycling industry, a shortcoming this thesis addresses. We ask to what extent and in what ways state-of-the-art deep learning methods could constitute an intelligent and generalizing scheme that can be used to disassemble commonly found computer parts such as hard drives and graphical processing units, that are known to contain valuable metals. Using relevant metrics to evaluate the accuracy and performance of individual components and the entire system altogether, we empirically show that methods based on deep learning and computer vision are well suited for estimating the state of disassembly and inferring the required visual parameters for possible action executions. The significance of this study is that it introduces an industry-oriented scheme that only requires off-the-shelf sensors to operate, and can be repurposed to work with new products. The work has also been part of Horizon 2020 project "IMAGINE", aimed to develop a fully automated disassembly robot to be used in recycling plants. Therefore, the results obtained and presented in this thesis have also been used in the IMAGINE project.
Keywords: Deep learning; object detection; object classification; image segmentation; computer vision
 
Statistik