Presentation Profile
Optimizing NIR and Raman Performance
Main Author
Brian Rohrback - Infometrix, Inc.
Abstract:
Optical spectroscopy is the go-to technology to manage myriad applications both in the laboratory and in a process stream. It measures chemistry and physical properties essentially in real time and is non-destructive, quantitative, and requires less handling. The price to pay is that the spectrometer will require calibration to translate the spectral fingerprint into an analytical result. This distills down to three tasks.
· One is to create an optimized method for processing the data, typically setting appropriate preprocessing techniques and possibly selecting the optimal wavelengths. Even choice of algorithm is up for grabs in this method development step.
· Then the issue of maintaining the calibration as process conditions or input material change becomes the effort. Here, key is the selection of optimal rank (e.g., number of factors in a PLS model) along with choosing the best spectra to use in the calibration.
· Finally, a system to evaluate model quality (as in ASTM D6122) is critical to provide an objective evaluation of the current model’s quality.
All these tasks can be done autonomously, and the result is a vast simplification of the calibration process that is independent of technician training and consumes a negligeable amount of time to ensure success.
