Human visual quality control of lyo cakes is dependent on the individual operator’s perception. Alternatively, if a camera inspection is used it is dependent on the imaging analysis technology which has also proved to be unreliable. This is often the cause of a high, unnecessary reject rate. Therefore, NIR provides an accurate, efficient, environmentally and operator friendly, quality control application which is 100% automated.

Our customer was experiencing a problem with their product which happened during the freeze-drying process. This is a risk to product reliability. Cake defect often occurs because of a problem with humidity. It can be an indication that the lyo process for those containers has not worked properly. Last but not least, a cosmetic defect can occur so that the cake looks different and can cause customer complaints.

To detect the cake defect in the process two sets of product samples were needed, one good and the other bad. It was important that our customer gave us the exact properties of a cake defect type for the evaluation of the efficient classification of the model. A chemometric model was developed for the classification of defects. The model was validated using independent samples but the properties of these samples were known.   

The validation is based on the model properly predicting the characteristics / features of these known samples.

The engineering and method development studies resulted in an automated inspection platform that can 100% test multiple product sizes at speeds of up to 600 vials per minute. The cake defect detection can be measured non-destructively and at a high throughput.