(75c) Data-Driven Approaches Towards Equipment Health-Classification and Predictive Monitoring in Drug Product Manufacturing

Digitalization is the central pillar towards the introduction of Industry 4.0 in pharmaceutical manufacturing. However, sophisticated data-driven approaches are yet to be explored to fully take advantage of the abundance of recorded process data available. Drug manufacturing is a highly regulated field. Thus, regulators require manufacturers to store process data over many years for backtracking purposes. This work explores the potential to use such data for predictive maintenance purposes.

This work presents a data-driven approach for equipment monitoring and the implementation of equipment classification and predictive maintenance in an aseptic filling line in drug product manufacturing. The goal is to leverage stored short-term process monitoring data for a secondary application in long-term equipment monitoring. Multiple years’ worth of industrial process data are used, which include process monitoring data and logbook information regarding each process run. Process monitoring data is mainly acquired through Supervisory Control and Data Acquisition (SCADA). Monitored variables include pressure, temperature, or process step duration. Furthermore, logbook information records details regarding equipment and process failures as well as maintenance data including measures taken and maintenance intervals.

In a first step, process data was pre-processed and different classification algorithms were applied. Thereby, equipment failures were shown to be clearly linked with process monitoring data through, either, observable shifts in principal curves or peaks representing events. Short-term noise in the process data through variation caused by manual operation and batch-wise production were successfully minimized and long-term based effects due to equipment deterioration were identified. An analysis of results obtained from different classification algorithms for equipment monitoring is presented.

In a next step, different machine learning algorithms are applied to the process data and compared in their ability to predict equipment deterioration and failure. Additionally, optimized maintenance strategies for equipment are formulated. Finally, a decision-support framework is being developed, where expert knowledge will be included for defining counter measures taken in the case of alarming changes in equipment status.