Emuowhochere, Oghenevwegba and Covenant University, Theses (2022) DEVELOPMENT OF AN ANN-BASED DEFECT DETECTION SYSTEM FOR PROCESS QUALITY OPTIMISATION IN BOTTLING INDUSTRY. Masters thesis, Covenant University Ota.
![[img]](http://eprints.covenantuniversity.edu.ng/style/images/fileicons/application_pdf.png) |
PDF
Download (182kB) |
Abstract
The quality of a manufactured product is a major factor that contributes to the growth, productivity and sustainability of a manufacturing firm. This is the major reason a lot of resources both human and machines in the area of quality control and quality assurance. In bottling industries, (that is firms that make use of bottles as raw material, and firms that produce bottles as end product), quality control in the production floor is also vital part of the manufacturing process and here in Africa, human labour are mostly used for the quality control process and also to save cost. The use of these human factors in tracking possible defected in the production line has led to loss of material due to fatigue, which in turn affects the productivity of the manufacturing firm. The aim of this study is to develop an CNN based defect detection system that is capable of detecting possible defected bottles in the production line. The Convolutional Neural Network model was developed with; bottles with bad labels, bottles with bad caps and, bottle that met the quality requirement. These are the defects the model is expected to detect on the production line. The materials selected for this development of this system were; a server computer, raspberry pi 3A+, raspberry camera, and the CNN trained model. The raspberry pi serves as the client side of the setup which takes photos of bottles in the production line and sends them to the server running the CNN model. The training model of the CNN model was divided into two part which were; training of the CNN model with 100-pixels dataset and training of the CNN model with 300-pixel dataset. From the result obtained, the prediction accuracy of the trained model with 100-pixel dataset was 83.333% and that of the 300-pixel dataset was 90.81654%. Running the CNN model on the raspberry pi solo leads to an average response time of 15 minutes per prediction and this is not ideal for an industrial space. Running the raspberry pi as a client-side image collector reduced the response time of the setup to an average response time of 8 sec. This response time is ideal and better than the human factor response time.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | ANN, CNN, Artificial Intelligence, Defects, bottle, Manufacturing, Industries |
| Subjects: | T Technology > T Technology (General) T Technology > TJ Mechanical engineering and machinery |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Engineering Sciences |
| Depositing User: | nwokealisi |
| Date Deposited: | 12 Sep 2022 11:39 |
| Last Modified: | 12 Sep 2022 11:39 |
| URI: | http://eprints.covenantuniversity.edu.ng/id/eprint/16165 |
Actions (login required)
 |
View Item |
