Title Kompiuterinės regos ir mašininio mokymosi pritaikymo tyrimas pramoninės gamybos gaminių kokybės patikrai /
Translation of Title Research of computer vision and machine learning in industrial product quality control.
Authors Kartavičius, Algirdas
Full Text Download
Pages 65
Keywords [eng] Computer vision ; machine learning ; surface defect detection ; industrial manufacturing ; convolutional neural networks
Abstract [eng] The aim of the work is to identify defects in industrial products and to ensure the appropriate quality of manufactures using a product photo and machine learning algorithms. Because furniture production is likely the largest sector of Lithuanian industry, it is planned to delve into the quality assessment of furniture details. Product quality control is critical because it allows you to quickly identify problems, reduce production costs, avoid costly errors. The automated inspection allows you allows for objective evaluation of product quality and eliminates human error. The most important part of the work is the study of machine learning algorithms that are most suitable for defect identification. Because this is a real-time process, the system must be able to quickly determine the part's quality. Considerable attention is planned to the acceleration of data annotation, as this process is very time-consuming. The application of convolutional neural networks to defect detection was investigated. Several object detection algorithms were tested during the study. It was decided to use the “YOLOv4” method, which is known for its high speed and accuracy. Edge defect detection models were trained during the experiment. A part of the system has been developed that allows the integration of a surface defect detection model into the quality system of furniture parts. This system can be successfully implemented in furniture component factories with large-scale production.
Dissertation Institution Kauno technologijos universitetas.
Type Master thesis
Language Lithuanian
Publication date 2021