Work

Intelligent anomaly detection in the manufacturing of kitchen cabinets

Year: 2024
Type: dissertation
Abstract: Anomaly detection within manufacturing processes is critical for ensuring product quality and customer satisfaction. In this study, we explore the possibility of employing an efficient anomaly detection model for manufacturing kitchen cabinets with improved performance and training time. Although several ML and DL models exist, developing a model that effectively combines the strengths of deep learning's automatic feature extraction with the reliability of explicit feature extraction, while mitigating the challenges of black-box nature and the high training time, could offer an efficient solution for industry applications. Through a comprehensive investigation, this work examines the impact of digital image processing methods on the performance of anomaly detection models when used for defect enhancement and segmentation. This research begins with a thorough analysis of defective characteristics, followed by the application of digital image processing techniques to enhance and segment defective regions within images. Subsequently, the performance of various machine learning and deep learning models is evaluated, including Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and YOLOv8n-cls, trained on defect enhanced, defect segmented, and unsegmented datasets. The findings reveal that ML models trained on enhanced data consistently outperform those trained on unsegmented data, exhibiting higher accuracy and precision in defect detection tasks. Furthermore, YOLOv8n-cls, a deep learning model, demonstrates superior convergence rates and robustness to variations in lighting conditions when trained on defect-enhanced datasets. These results underscore the importance of digital image processing techniques in enhancing anomaly detection performance. This study contributes to the advancement of quality control practices in manufacturing industries, paving the way for more efficient and reliable defect detection systems. The findings indicate future avenues for exploring the proposed technique in addressing defects across various domains.
Author: Lakshminarayanan, Sadhana
Language: English
Open Access status: green
OpenAlex ID: W7029414591
Cites:
Cited by: 1
Related to: 10