Projection to latent structures regression and its application to Mach–Zehnder interferometer optical fiber sensors for acetone detection

The techniques of multivariate analysis methods as prediction models open a new vision to the analysis of applied data. The method used in this work was projection to latent structures regression, which is a regression method that uses the latent variables generated by calculating the linear relationship between the dependent and independent variables giving them equal importance and thus obtaining the maximum variance and correlation of these variables. This method was applied to optical fiber sensors for acetone detection. It is important to detect acetone because it is a biomarker of diabetes mellitus. The sensor was fabricated with two cascaded long-period fiber gratings (LPFG) with a 515 µm grating period and separated 1 cm to form a Mach–Zehnder interferometer (MZI). Clinical studies for the diagnosis of diabetes are usually invasive, the development of these sensors proposes a new non-invasive option through the detection of acetone in human breath whose concentrations are in the order of 1.25–2.5 ppm. To study the response of the sensors to acetone, different sensing films, such as polydimethylsiloxane (PDMS), polymethyl methacrylate, Apiezon L and Apiezon T were used, which have a good affinity to this compound. Spectral changes in the transmission spectrum of the MZI were obtained due to the modes interference together with an increment of the sensitivity, since the interaction between the acetone concentration and the sensing film provokes a change in the effective index of the cladding, which in turn is amplified by the LPFGs separation. The analysis showed that the best results were obtained for the sensor with PDMS as sensing film, with the lowest limit of detection, 1.7 ppm using 4 latent structures.