• Defect detection
• Glass tube production
• Real time inspection
• Image processing
• Inspection systems

Syringes, vials and carpules for pharmaceutical products are usually made of borosilicate glass. Such containers are made by glass converting companies starting from single glass tubes. These glass containers can suffer from inclusions, air bubbles, scratches and others issues, that can cause subsequent problems like product contamination with glass particulate or cracks in the glass. In recent years, more than 100 million units of drugs packaged in vials or syringes have been withdrawn from the market. As a consequence pharmaceutical companies are demanding an increased delivery of high quality products to manufacturers of glass containers and therefore of glass tubes.
An automatic, vision based, quality inspection system can be devoted to perform such task, but specific process features requires the introduction of ad-hoc solutions: in the production lines tubes significantly vibrate and rotate, and the cylindrical surface of the tube needs to be inspected at 360 degrees. The system must be designed and tested in such a way as to be effective in identifying defects and impurities of the order of tens of microns. Current solutions severely limit production capacity as they only support production lines up to a certain speed. Moreover, more sophisticated acquisition systems can guarantee a better definition of the smallest recognised defect but require software systems with reduced processing times. In this work we present software solutions, and in particular algorithms, that can replace the current techniques used by inspection systems used for quality control of glass tubes. These algorithms, being developed for the specific application domain, are more performing both from the point of view of processing time and accuracy in the defects detection. However, some of the proposed solutions, and in particular the ROI reduction technique, represent general methodologies that can be applied to accelerate the defect detection and classification phases. The reasoning that led to the definition of each algorithm was discussed and a strategy was presented for the automatic calculation of the parameters used by each algorithm that adapts to the current environmental and working conditions.