Pharmaceutical Manufacturing Process Anomalies Detection And Diagnosis Using Machine Learning

Marco Vicentini, Analyst IDS Digital Solutions And Data Engineering, Eli Lilly Italia

Giuseppe Salerno, Sr Associate IDS Digital Solutions And Data Engineering, Eli Lilly Italia

Repeatable and stable manufacturing processes are essential for ensuring a reliable supply. In the realm of pharmaceutical continuous batch processes, maintaining stability amidst the intricate web of multivariate time series variables presents a significant challenge.
This presentation introduces a solution tailored to address this challenge by harnessing the combined capabilities of Seeq and Power BI. The integration facilitates anomaly detection within continuous batch processing, streamlining the diagnostic process to pinpoint and mitigate sources of process variability.

Built on a data-driven foundation, the solution employs time series distance computation between a golden standard series and those generated during production cycles. These data inputs fuel a neural network tasked with classifying cycles as anomalous or within expected parameters.
What distinguishes this solution is its level of abstraction and adaptability, allowing for easy replication across different production processes with very limited configuration effort. Moreover, its operational versatility via Python embedded in Seeq Data Lab promotes scalability across various manufacturing environments.
This presentation not only showcases the effectiveness of the solution in ensuring process stability but also emphasizes its adaptability and scalability, highlighting its potential as a versatile tool for enhancing reliability in pharmaceutical manufacturing.