On-line product quality and process failure monitoring in freeze-drying of pharmaceutical products

In this work the information provided by a non-invasive imaging sensor was used to develop two algorithms for real time fault detection and product quality monitoring during the Vacuum Freeze-Drying of single dose pharmaceuticals. Two algorithms based on multivariate statistical techniques, namely Principal Component Analysis (PCA) and Partial Least Square Regression (PLS), were developed and compared. Five batches obtained under Normal Operating Conditions (NOC) were used to train a reference model of the process; the classification abilities of these algorithms were tested on five more batches simulating different kind of faults. Good classification performances have been obtained with both algorithms. Coupling the information obtained from an infrared camera with that of other variables obtained from the PLC of the equipment, and from the textural analysis performed on the RGB images of the product, strongly improves the performances of the algorithms. The proposed algorithms can account for the heterogeneity of the batch and aim to reduce the off-specification products.