Klaus Dierolf, group head of project management at Schubert Packaging Systems, and Daniel Greb, head of image processing at Gerhard Schubert look at camera-based clearance in pharmaceutical production.
Schubert Packaging Systems
When it comes to pharmaceutical manufacturing, a batch is not immediately followed by the next. Before new medicines can pass through a production line, manufacturers must first check that it is free of incorrect products or potential residues – this is the principle behind line clearance. Line design and image processing systems help ensure that production lines are clean, correctly prepared and free of risks before a new manufacturing process can begin.
Thousands of products a day, often processed at very high speeds: day-to-day pharmaceutical production occupies a unique position in terms of efficiency – and places high demands on hygienic purity, monitoring and documentation. When numerous individual products pass through a line every day, certain risks are inevitable: mix-ups or cross-contamination can occur when there are rapid changes between products on the same line. In such cases, residues from the previous product may end up in the new batch. If active ingredients mix, they can compromise the quality of the end product – with potentially serious implications for patients. Materials or labelling can also become ‘mixed up’: if product names or packaging designs are similar, the likelihood increases that incorrect labels, package leaflets or packaging will remain in the production area – and, in the worst-case scenario, lead to mislabelled products that may fail to fulfil their therapeutic purpose.
To prevent such issues and regulatory breaches, the pharmaceutical industry’s Good Manufacturing Practice (GMP) stipulates a line clearance procedure. It involves several control steps, ranging from the removal of all residues or incorrect or non-compliant products, through a visual inspection for purity, to a thorough check of the entire line to ensure correct settings and the right tools. As soon as the line is entirely clean and correctly set up, qualified personnel document the results of each line clearance and issue written authorisation for the next batch. Predefined criteria provide guidance on a case-by-case basis.
Systematic clearance
Personnel need to rely on more than their own powers of observation: a regulatory-compliant, efficient line clearance is based on specially designed equipment and supporting inspection systems – also with the aim of ensuring that processes do not remain idle between two batches for any longer than necessary. Pharmaceutical manufacturers depend on support solutions that ensure rapid clearance and ultimately the highest possible line availability. Closely coordinated interaction between the individual machines on a line and image-processing systems is critical: cameras can detect anomalies at defined points in the process, and the mechanical design of the machines prevents individual products from falling out of sight or even leaving the process.
Even with state-of-the-art technology, deviations from the process cannot be entirely avoided. For example, when packaging vials, defects in gripping tools or imprecise product handling can result in the glass containers not being placed in the designated tray and remaining within the line. To prevent cross-contamination or mix-ups with subsequent batches, operators need to be able to easily locate and collect the affected products. With this in mind, machine manufacturers such as Schubert equip their pharmaceutical packaging machines with image processing systems as well as seamlessly integrated protective features at critical points such as transfer areas. These ensure that misplaced products or those that have strayed from the process remain visible within the machines.
Tracking deviations
Any product that strays from the process represents an anomaly that can be detected at an early stage using advanced area cameras. Although the products remain within the respective line thanks to the integrated protective features, depending on the size and layout of the design, it may not be possible to view every area directly. Extensive lines, in particular, can quickly develop several potentially critical points where products that have deviated from the process could accumulate – and there could be too many for human staff to inspect continuously. This makes automated vision systems that take on this task all the more important: area scan cameras with sensors are increasingly being used for line clearance – monitoring even hard-to-reach areas and reliably detecting foreign objects.
Two approaches have proven highly effective in practice. Classification systems recognise objects and categorise them as acceptable or unacceptable; the respective categories are defined in advance in accordance with very specific line clearance requirements. In classifying camera systems, neural networks are ‘trained’ in advance using synthetic datasets of the undesirable objects so that the area sensors can recognise them consistently. If a camera detects a product classified as unacceptable, it sends a corresponding visual alert to the operator via interfaces such as Human Machine Interfaces (HMI). Visual and acoustic signals, such as lights or warning tones, can additionally signal the presence of a foreign object and the need for action.
Anomaly detection offers a widely used method that can be rapidly deployed, whereby the camera simply captures and recognises the environment in which it is located. As the underlying neural network is only familiar with the normal state, it classifies foreign objects within this range as deviations from the norm without describing them in detail. Unlike classification systems, anomaly detection does not require extensive training of neural networks using multiple datasets, yet it accurately detects unwanted objects in the process and immediately informs the operating personnel.
Manufacturers who use flexible foreign object detection solutions benefit from a significant advantage. They can switch between detection methods as required. If foreign objects are varied and unknown, making classification impossible, anomaly detection is the most suitable method. For known and therefore clearly definable types of defects, classification is used, provided there is sufficient training data for neural networks. In state-of-the-art vision systems, such as those developed and provided by Schubert, hybrid approaches are standard and offer manufacturers considerable flexibility by allowing them to choose between high sensitivity (anomaly detection) and high specificity (classification) – thereby enabling foreign object detection that is precisely tailored to the situation. Furthermore, vision systems that interact closely with the line management systems are conceivable in the future: here, the camera-based technology sends images to the Learning Management System (LMS), which in turn integrates them into the batch report, thereby enabling even more precise batch documentation.
Which detection method manufacturers opt for depends on several factors – ranging from GMP criteria to data quality and the types of errors that may occur. Regardless of the chosen approach to foreign body detection, camera-based systems pave the way for efficient, time-saving line clearance, allowing even areas that are difficult for operating staff to access to be thoroughly inspected – ensuring the safe, GMP-compliant production of vital medicines.