Real-time detection of plastic part surface defects using deep learning- based object detection model

• Published in: Measurement : journal of the International Measurement Confederation Vol. 235; p. 114975
• Main Authors: Çelik, Miraç Tuba, Arslankaya, Seher, Yildiz, Aytaç
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-08-2024
• Subjects: Artificial intelligence; Deep learning; Defect detection; Quality control; You-Only-Look-Once version 8; Deep learning; You-Only-Look-Once version 8; Artificial intelligence; Quality control; Defect detection
• DOI: 10.1016/j.measurement.2024.114975
• EISSN: 1873-412X
• ISSN: 0263-2241

•High accuracy and efficiency have been achieved using the YOLOv8 object detection model.•Real-time detection of surface defects on plastic parts has been implemented.•The potential for improving quality control in plastic part production processes has been increased.•Real-time defect detection will make the company's processes more efficient, reduce human errors, and lower operational costs.
In this study, it was aimed to detect defects in plastic parts produced in a company operating in the automotive sub-industry using the YOLOv8 object detection model. The defect types seen in plastic parts were evaluated with the help of Pareto analysis, and scratches, stains and shine were selected as the most common defect types, and data on the three defect types were collected. YOLOv8 models were trained using faulty part images. As a result of the training, the highest mean average precision value of 0.990 was obtained in the YOLOv8s model, and the shortest training time was obtained in the YOLOv8n model. In the YOLOv8s model, which gave the highest mAP value, hyperparameter adjustment was made according to the batch size and learning rate values. The testing phase was carried out with the hyperparameter values that gave the best results and the mAP value was obtained as 0.902.