Multivariate Analysis as a Tool to Identify Concentrations from Strongly Overlapping Gas Spectra

Abstract

We applied a multivariate analysis (MVA) to spectroscopic data of gas mixtures in the mid-IR in order to calculate the concentrations of the single components which exhibit strongly overlapping absorption spectra. This is a common challenge in broadband spectroscopy. Photoacoustic (PA) measurements of different volatile organic compounds (VOCs) in the wavelength region of 3250 nm to 3550 nm served as the exemplary detection technique. Partial least squares regression (PLS) was used to calculate concentrations from the PA spectra. After calibration, the PLS model was able to determine concentrations of single VOCs with a relative accuracy of 2.60%.

Multivariate Analysis of Photoacoustic Spectra for the Detection of Short-Chained Hydrocarbon Isotopologues

“…In order to obtain the concentrations of individual substances of a measured spectrum containing multiple substances, multivariate analysis (MVA) [21,22] can be applied. This was also demonstrated in a recent work, where PA measurements of volatile organic compounds from strongly overlapping spectra in the mid IR region were evaluated [23]. Well-known multivariate regression models are the multiple linear regression, the principal component regression, and the partial least-square regression (PLSR), whereby the latter has become the standard method for spectroscopy [21,24].…”

“…As output values, loading and score matrix of X and Y, explained variance (in %) by the model by each LV, and the PLSR coefficients are computed. The PLSR coefficients are required to calculate target values (concentrations) out of initial data (spectra): where is the PLSR coefficient matrix, is the predictor matrix (spectra), the response matrix (concentrations), and the error matrix [ 23 , 24 ].…”

“…where B PLSR is the PLSR coefficient matrix, X pre is the predictor matrix (spectra), Y res the response matrix (concentrations), and G the error matrix [23,24]. The predictive accuracy, i.e., the standard/mean error of a multivariate calibration model, is expressed as the root-mean-square error (RMSE).…”

“…Well-known multivariate regression models are the multiple linear regression, the principal component regression, and the partial least-square regression (PLSR), whereby the latter has become the standard method for spectroscopy [21,24]. If the data are noisy and strongly correlated, PLSR is considered to be superior to the other techniques [23,24]. Here, a PLSR model was trained and verified with experimental data of single and multiple components in synthetic gas mixtures.…”

Multivariate Spectra Analysis: PLSR vs. PCA + MLR

“…If the spectral features of the single substances are broad and overlap strongly, the spectra evaluation requires a multivariate analysis. The general suitability of Partial Least Squares Regression (PLSR) to determine the absolute concentrations of different components of a mixture has been demonstrated [2,3]. However, the according study also revealed certain limitations of this evaluation method.…”