Data-driven equipment condition monitoring and reliability assessment for sterile drug product manufacturing: Method and application for an operating facility

Abstract

Data-driven equipment condition monitoring is difficult to apply in operating pharmaceutical manufacturing plants. The required revalidation costs and efforts hinder the installation of dedicated sensors. This work proposes a three-step methodology to exploit the existing process monitoring infrastructure to infer equipment condition-related information. The developed methodology addresses the challenges imposed by using process monitoring data for a secondary application. In the detection step of the methodology, batch variations are separated from changes due to underlying equipment deterioration. Principal component analysis is conducted followed by slow feature analysis to extract long-term equipment variations. After identifying relevant shifts in the slow feature baselines, the sources of variations in equipment conditions are traced back to sets of installed sensors in the localization step. The identified sensor sets are used to determine the location and physical type of cause of the changes. Last, in the prevention step, tailored maintenance actions are proposed to avoid potential equipment-related failures. Two case studies are presented to demonstrate the application of the methodology with real industrial data. Data from historical maintenance records in the facility were used to verify the findings from the case study. The successful application provides lessons-learned to include maintenance as part of holistic system design and is an important step towards data-driven equipment reliability assessment and maintenance planning in pharmaceutical manufacturing.