A Scalable Deep Learning Approach for Real-Time Multivariate Monitoring of Biopharmaceutical Processes With No Prior...

Area of Science:

• Biopharmaceutical Manufacturing
• Process Engineering
• Data Science

Background:

• Real-time multivariate statistical process monitoring (RT-MSPM) is critical for bio-pharmaceutical production.
• Existing methods like Batch Evolution Models (BEMs) have limitations, especially in multi-product facilities.
• Monitoring new products without historical data presents a significant challenge.

Purpose of the Study:

• To propose a novel deep learning framework for RT-MSPM in biopharmaceutical processes.
• To enable monitoring of cell culture manufacturing for new products without prior history.
• To develop a real-time root cause identification algorithm for anomalies.

Main Methods:

• Utilized Autoencoders (AEs) for anomaly detection.
• Implemented a multistage real-time data processing algorithm.
• Developed a novel algorithm for real-time root cause identification and contribution charting.

Main Results:

• The proposed AE-based framework demonstrated robust anomaly detection and root cause identification.
• AEs provided stronger evidence for anomalous patterns compared to conventional linear methods.
• The framework was successfully tested in a scalable software product for bioreactor monitoring.

Conclusions:

• The novel deep learning framework offers an effective solution for RT-MSPM in biopharmaceutical manufacturing.
• This approach enhances the ability to monitor and manage cell culture processes with no prior product history.
• The developed system contributes to improved fault detection, prevention, and root cause analysis.