A Machine Vision Approach for Defect Inspection of Vials

Manufacturing defects in vials are prevalent in pharmaceutical industries during the production of drug fluids. These defects can jeopardize product integrity and profitability. Various imperfections, such as cracks, black spots, bubbles, and scratches, can occur across different regions of the vial, including the surface, neck, mouth, and base. Effective quality control systems are crucial for identifying and rejecting these defects before they tarnish the company's reputation and incur additional costs. Pharmaceutical companies require dedicated inspection lines equipped with advanced processors, memory, and integrated machine assemblies. However, existing systems often suffer from accuracy and speed limitations and lack a unified approach for detecting all types of defects. This study proposes a realtime inspection system to address these challenges. The proposed system utilizes industrial cameras to capture vial images, which undergo preprocessing stages. During preprocessing, noise in the vial image is mitigated using discrete wavelet transform. Subsequently, thresholding techniques are applied to identify defects in the region of interest (ROI) on vials, utilizing the Otsu technique to generate optimal threshold values. To detect defects on the mouth and base regions, ROI identification for the base region is performed using the Hough Transform. The implemented system has demonstrated superior accuracy compared to existing solutions, inspecting all regions of the vial. The proposed method has achieved a precision of 97.86%, recall of 98.56%, accuracy of 98.13%, F1-score of 98.21 %, AVC-ROC of 98.14%, and AVC-PR of 98.77%.