Threshold concentration monitoring based on pattern recognition analysis of differential near-infrared spectra

Abstract

A threshold concentration monitoring procedure based on near-infrared (near-IR) spectroscopy is described for use in continuous process monitoring applications. The method is based on collecting an off-line reference sample and obtaining a near-IR spectrum and corresponding reference concentration at the start of the monitoring period. Subsequently, spectra are collected continuously and ratios are taken with respect to the reference spectrum. The resulting spectra in absorbance units are differential spectra whose effective analyte concentration (termed the differential concentration) is the difference in concentration relative to the concentration in the reference sample. By knowing the reference concentration and a user-specified threshold, a critical concentration can be defined that specifies the threshold in terms of differential concentrations. Determining whether the analyte concentration is within specification can then be addressed as a pattern classification problem and a qualitative classification model can be used to discriminate differential spectra that reflect the two possible states: within or outside of specification. A simulated biological process is used to test the methodology in which a dynamic system of glucose, lactate, urea, and triacetin in the mM range in phosphate buffer is monitored continuously to detect occurrences when the glucose concentration drops below a threshold of 3.0 mM. With the use of three sets of prediction data, one of which was collected 2.5 years after the calibration data, the monitoring algorithm is implemented with 100% successful detections and no false detections.