Abstract
In any pharmaceutical packaging process, error rates must be kept very low to prevent defective or incorrect medicine from reaching the consumer. Published articles include a focus on the inspection of ampoule content for the presence of contaminants. Methods of inspection include optical cameras that record sequential images to detect the movement of floating objects via feature extraction and classification. Other methods employ spectral imaging and dying of ampoules to detect cracks. Thermal imaging is also used to inspect structural integrity and mechanical stress; however, this approach is marked by relatively high cost. This paper describes the enhancement of an optical vision-based inspection system for a glass ampoule packaging process. The developed solution involves a reconfiguration of the illumination system of the ampoule trays as well as reprogramming of a real-time optical camera. The camera is linked to the process controller and triggers the ejection of trays containing faulty ampoules. Eight defect types are of interest, namely: missing ampoules; ampoules without a label; crushed ampoules; ampoules with a broken tip - present or missing; ampoules with a broken base; cracked ampoules; empty ampoules. The original inspection system performance was inadequate and only detected 2 of the 8 defect types reliably. The improved system which is described in this paper has adequately detected all aforementioned defect types. Experiments show that the overall detection rates have improved from 55% to 99.6% on average. Average camera detection accuracy has improved from 38% to 99.5%. In addition, the false positives, i.e. rejection rates, were reduced from 45% to 7%.