Machine learning assisted in-line calibration models for near-infrared spectroscopy in dry granulation.

Area of Science:

• Pharmaceutical Manufacturing
• Process Analytical Technology (PAT)
• Spectroscopy

Background:

• Dry granulation is crucial for solid oral dosage forms, requiring precise control due to variable particle properties.
• Near-infrared (NIR) Spectroscopy offers fast, non-invasive characterization essential for pharmaceutical process monitoring.
• Existing methods struggle with real-time monitoring of continuous granular flow.

Purpose of the Study:

• To develop and evaluate an in-line NIR spectroscopy system for continuous monitoring of dry granulation.
• To compare the performance of various nonlinear calibration techniques against a linear baseline.
• To assess the system's ability to predict key granule physical attributes in real-time.

Main Methods:

• An in-line NIR spectroscopy system with a specialized SpiderWheel sampling device was implemented.
• Calibration models were developed using artificial neural networks (ANN), Bayesian neural networks via Monte Carlo Dropout (BNN-MCD), and least-squares support vector machines (LS-SVM), with partial least squares (PLS) as a baseline.
• Spectral preprocessing techniques, including Savitzky-Golay filtering and standard scaling, were optimized.

Main Results:

• The BNN-MCD model demonstrated superior predictive accuracy for granule size distribution (R² = 0.993).
• Accurate predictions were achieved for porosity (R² = 0.984), specific pore volume (R² = 0.984), and envelope density (R² = 0.980).
• Optimized preprocessing significantly enhanced nonlinear model performance.

Conclusions:

• The developed BNN-MCD assisted NIR spectroscopy system provides accurate, real-time characterization of pharmaceutical granules.
• This advancement supports Quality by Design (QbD) principles and enhances process monitoring in continuous manufacturing.
• The system represents a significant step forward in Process Analytical Technology (PAT) for pharmaceutical production.