Evaluation of the influence of sample variability on the calibration of a NIR photometer for quantification of ethanol in gasoline

The calibration of a NIR photometer for analysis of liquid substances aims to find a curve that relates the concentration of some analyte with spectral data collected from the samples. However, the measured optical signals depend not only on the concentration of the analyte of interest, so that the quality of the calibration also depends on the natural variability of samples and on the way the optical data are processed and modeled. This paper presents a study on the influence of these two factors on the calibration of a LED-NIR photometer, which was applied to the quantification of ethanol in gasoline, and compares the results obtained by the prototype with those obtained by a commercial spectrophotometer. We used two distinct procedures for selecting the calibration samples, in order to evaluate the influence of the sample matrix, considering two different methods of determining the absorbances from the optical data. Experimental results showed that the calibration samples must be prepared and selected with uniform distribution in relation to the different sources of variance, in order to make the equipment able to accurately predict future samples. The performance of the prototype was very close to the one of the commercial equipment, with a difference in root mean square errors of less than 0.8% in the concentration (v/v) of ethanol.