Detection of Defects on Pharmaceutical Tablets
Title: Detection of Defects on Pharmaceutical Tablets
Speaker: Dr. Erica Donnelly-Swift, UCD School of Mathematical Sciences
Date: Thursday, March 5th
Time: 4.00PM
Location: Science East, Room E0.01
Abstract:
Identification of defects is a high priority within the pharmaceutical industry. If a defect issue has been identified in the manufacturing process, it is often necessary to have visual inspections of a large number of batches to determine the defect was present in one particular batch or if it was in a number of different batches. Visual inspections are a time consuming tedious process which are often prone to human error. This research applies image processing filters and morphological techniques to pre-process the images of tablets and statistical techniques are used to identify broken/scratched tablets. From existing literature in the area of image processing, it has been shown that the Weibull distribution adequately fits image data and recent research has implemented user defined thresholds to detect defects. In epidemiological literature, scan statistics are widely used to identify cluster locations - thus his research utilizes the two parameter Weibull of gradient magnitudes and a variable window scan statistic to identify defects. This approach offers a non-supervised approach to defect detection that does not require training data, is independent of tablet texture and does not require a user defined threshold. Results indicate that this approach is successful in the identification of defects.
Series: Statistics
Social Media Links