Operational Optimisation of Pharmaceutical Lyophilizers through Leak Detection Data Analysis

The freeze-drying process involves reducing pressure to extremely low levels within a sealed chamber to evaporate the product water. It is a vital process within the pharmaceutical industry, aimed at stabilizing, preserving, or extending the shelf life of drug products. In lyophilizers, external leaks are a major issue. Indeed, external inflows contaminate the lyophilization chamber and may cause the disposal of the current production batch. Given that a single batch often comprises thousands of product vials, leakages from freeze-dryers pose significant challenges throughout the production chain of lyophilized drugs. Early detection of leakages is critical to optimise the drug production process. This paper presents a novel leak detection data analysis based on Principal Component Analysis (PCA), alongside a comparison of two clustering algorithms. Through proper hyperparameter optimization, the obtained results show that K-Means Clustering exhibits superior performance in identifying anomalies and deterioration within historical data collected from leak test cycles. Additionally, a Health Indicator (HI) is assessed based on computed clusters to precisely identify trends where leaks exert the most substantial impact.