Development of a Modelling Framework for NIR Spectroscopy Based on-line Analyzers using Dimensional Reduction Techniques and Genetic Programming

Abstract

The quality of fuels is determined by several properties (e.g. aromatic components, cloud point, flash point, density etc.). Measurements of these properties are mostly costly and time consuming that makes real-time control infeasible. An affordable option to perform real-time quality control is the application of Near Infra-Red (NIR) spectroscopy based on-line analyzers. Using this tool multiple measurements can be substituted by one measurement combined with a complex prediction model. The two dimensional mapping of the spectral space can be used for monitoring the operation of the production and for the validation of the analyzer. The mapping is based on two aggregates which are mathematical functions combining absorbance values at several wavelengths. We developed a genetic programming based approach to design these aggregates. Results related to the monitoring of diesel fuel blending process illustrate the applicability of the method.