Use of an Automatic System in the Preparation of Gas Mixtures for Multivariate Calibration: A Case Study Involving NIR Analysis of Natural Gas

This paper investigates the use of an automatic system for preparation of gas mixtures in a multivariate calibration problem involving near-infrared (NIR) spectrometric analysis of natural gas. The automatic system is used to prepare calibration mixtures according to a Brereton experimental design, in order to exploit a suitable range of gas concentrations and thus avoid extrapolations in the predictions. These mixtures were employed to build partial-least-squares models for NIR determination of methane, ethane and propane, which are the major components of natural gas. Prediction performance was evaluated by using a separate set of prepared mixtures and natural gas samples with composition analyzed by gas chromatography, as well as a group of certified mixtures. The resulting root-mean-square errors of prediction (RMSEP) values for methane, ethane and propane (3.0, 0.9 and 1.2% mol mol-1, respectively) were approximately 10 times smaller than the corresponding calibration ranges, with correlations of 0.91, 0.96 and 0.86 between the predicted and reference values.