Self-organizing maps-based generalized feature set selection for model adaption without reference data for batch...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Process Analytical Technology (PAT)

Background:

• Model adaptation is crucial for maintaining calibration model accuracy in batch processes using FTIR and multivariate calibration.
• Existing methods often require reference measurements or struggle with batch-to-batch variations.

Purpose of the Study:

• To introduce a novel, standard-free model adaptation method (VSSOM) for batch processes.
• To eliminate the need for reference measurements in adapting calibration models.
• To enhance the robustness and predictability of FTIR-based process monitoring.

Main Methods:

• Variable Selection Strategy with Self-Organizing Maps (VSSOM) classifies spectral variables using Self-Organizing Maps (SOM).
• It identifies optimal feature subsets by intersecting primary and secondary spectral variable classes.
• The method selects feature subsets that minimize Root Mean Square Error of Cross-Validation (RMSECV) on the primary calibration set.

Main Results:

• VSSOM demonstrated superior performance in adapting FTIR calibration models across different batches.
• The method achieved comparable prediction performance to existing techniques without requiring reference data.
• Comparative studies on γ-polyglutamic acid fermentation and paeoniflorin extraction datasets validated VSSOM's effectiveness.

Conclusions:

• VSSOM offers a robust and efficient standard-free approach for model adaptation in batch process monitoring.
• The method enhances the stability and consistency of feature subsets across varying batch conditions.
• VSSOM presents a significant advancement for real-time process analysis using FTIR spectroscopy.