A Machine Learning Approach to Qualitatively Evaluate Different Granulation Phases by Acoustic Emissions

Area of Science:

• Pharmaceutical Engineering
• Process Analytical Technology
• Machine Learning Applications

Background:

• Wet granulation is a critical pharmaceutical process impacting final dosage form quality.
• Effective monitoring of granulation phases is essential for process control and economic efficiency.
• Current monitoring methods may not fully meet regulatory and quality standards.

Purpose of the Study:

• To develop and validate a machine learning approach for identifying distinct phases during wet granulation.
• To assess the efficacy of acoustic and vibration data for real-time process monitoring.
• To demonstrate a non-contact method compliant with pharmaceutical regulations.

Main Methods:

• Utilized microphones and an acceleration sensor to capture acoustic emissions and vibrations during granulation.
• Applied convolutional neural networks (CNNs) for phase classification based on sensor data.
• Transformed audio recordings were used as input for the CNN models.

Main Results:

• Achieved up to 90% classification accuracy using vibrational data.
• Reached up to 97% classification accuracy using audible microphone data.
• Demonstrated the correlation between identified phases and granule water content/processability.

Conclusions:

• Audible sound and machine learning are suitable for monitoring pharmaceutical wet granulation processes.
• Contactless acoustic monitoring aligns with regulatory requirements and Good Manufacturing Practices.
• This approach offers a promising tool for enhancing pharmaceutical manufacturing quality and control.