The quality control of products in the pharmaceutical industry is extremely important since various defects made during production, packaging or transportation can affect the effcacy and safety of medicines. Quality assurance in pharmaceutical manufacturing is thus a very complex problem, where quality control of end products alone is insuffcient. Namely, it requires monitoring and control of all the factors that can influence the end quality. These can include everything from the design of the manufacturing facility, individual processes or projects to the control of services and materials. In such systems, the quality control of end products serves only as the final confirmation of the quality.
Established methods of quality assurance therefore include not only the quality control of end products, but also routine quality control of raw materials and intermediate products by sample analysis in analytical laboratories. However, due to the variability of raw materials and individual processing phases, it is often impossible to reliably evaluate the quality of the entire batch of raw materials or intermediate products. Besides, the analytical procedures are usually very time-consuming. Consequently, in-line and on-line monitoring of process critical parameters are currently being enforced. Additionally, we can exploit the gathered measurements for an advanced optimization of energy effciency and process yield.
Some of the crucial elements of quality assurance of medicinal products are monitoring and
quality control of visual characteristics of materials, intermediate products and end products. Machine vision is a promising technique that has already proven useful or is being promoted at various stages of pharmaceutical manufacturing processes. For example, machine vision systems for visual inspection of tablets and capsules have proven superior compared to manual visual inspection in terms of reliability and speed, which allows for the inspection of every single product in a batch. Likewise, various laboratory machine vision systems for fully automated analysis of raw materials and intermediate products are routinely employed. Moreover, in recent years, machine vision systems are increasingly being adopted for in-line and on-line monitoring of manufacturing processes.
This dissertation studies machine vision systems for both quality control of end products as
well as in-line monitoring and analysis of manufacturing processes. In the first part, we propose and validate an image analysis method for print region detection on images of transparent pharmaceutical capsules for visual quality control. In the second part, we describe a novel method for in-line monitoring of the agglomeration degree of pharmaceutical pellets during the coating process using machine vision. Furthermore, we propose a machine learning approach that improves the accuracy and robustness of in-line agglomeration degree estimation.