Quantitative analysis of open-path FTIR spectra by using artificial neural networks

Abstract

Quantitative analysis of absorbance spectra to retrieve gas concentrations in open-path FTIR air monitoring is not always a straightforward task. Most of commercial software use classical-least-squared algorithms to retrieve the unknown concentrations. These codes usually work in real time and give appropriate results. However, sometimes these codes fail when the background reference spectrum presents absorption lines of the gas to be monitorized. This effect is frequent in some applications. Line-by-line approaches give satisfactory results because these codes solve the problem associated to the reference spectrum generating a synthetic reference background. The main drawback is that these algorithms do not work in real time, and need a skilled operator. In this work, we propose the use of artificial neural networks to analyze absorbance spectra in real time to retrieve the unknown concentrations in a simultaneous way. In addition, capabilities of the method to solve spectral overlapping will be studied. In this sense, simultaneous analysis of four atmospheric gases (CO↓2, CO, H↓2O and N↓2O) will be included in this first version. The effectiveness of the method will be evaluated from the experimental point of view. Experimental open-path FTIR spectra (0.5 cm↑-1 of spectral resolution) will be analyzed with the proposed method, as well as with CLS and LBL codes for comparison purposes. Moreover, in these experiments CO concentration has been measured by using standard extractive equipment and can be compared with the values provided by our method. Finally, some indications will be pointed to extend the method to other gases and spectral regions.