Drift correction in multivariate calibration models using on-line reference measurements

Area of Science:

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

Background:

• On-line spectral measurements can drift, compromising calibration models and analyte concentration predictions.
• Infrequently available on-line reference measurements are crucial for correcting these drifts.

Purpose of the Study:

• To develop and evaluate drift correction methods for on-line spectral calibration.
• To propose conditions for correct prediction using implicit correction methods (ICM) and explicit correction methods (ECM).

Main Methods:

• Utilized infrequent on-line reference measurements for drift correction.
• Proposed space-inclusion conditions, verifiable via Q-statistic monitoring, for valid ICM and ECM predictions.
• Investigated method performance under noise using Monte Carlo simulations and applied to fermentation data.

Main Results:

• Space-inclusion conditions provide necessary and sufficient criteria for correct prediction with ICM and ECM.
• ECM demonstrated slightly superior performance over ICM in the presence of noise, a statistically significant difference.
• Both ICM and ECM significantly reduced prediction errors for metabolite concentrations from infrared spectra.

Conclusions:

• Drift correction using ICM and ECM enhances the reliability of on-line spectral calibration.
• The proposed conditions offer a framework for monitoring calibration model validity and identifying the need for recalibration.
• ECM is recommended for applications with significant noise and rank-deficient data.