Robust principal component analysis-multivariate adaptive regression splines (rPCA-MARS) model for determining total acid number (TAN) and total base number (TBN) of crude oil samples using attenuated total reflectance fourier transform infrared (ATR-FTIR) spectroscopy

Abstract

Rapid assessments of Total Acid Number (TAN) and Total Base Number (TBN) in crude oil samples are significant in the oil industry. The analysis of TAN and TBN was conducted using the standard methods of the American Society for Testing and Materials (ASTM). However, the standard methods are costly and require a large number of samples for analysis. Therefore, providing analytical methods for the rapid crude oil analysis is very essential. In the current research, we report an application of Attenuated Total Reflection Fourier-Transform Infrared (ATR-FTIR) spectroscopy, based on robust Principal Component Analysis-Multivariate Adaptive Regression Splines (rPCA-MARS), for crude oil analysis. In the rPCA-MARS model, the ATR-FTIR spectral matrix data was decomposed into PCs scores using the rPCA method as an input data for the MARS model. The multivariate calibration models were applied to the analysis of crude oil samples based on the quantitative determination of total acid number (TAN) and total base number (TBN). An analytical method was developed for determination of TAN and TBN of crude oil samples. The ATR-FTIR spectroscopy coupled with multivariate calibration methods can be applied as a novel analytical method for crude oil samples. The result of partial least square regression (PLS-R), principle component analysis (PCR), Piecewise-Linear rPCA-MARS and Piecewise-cubic rPCA-MARS models were compared for analysis of total acid and base number of crude oil samples. The squared correlation coefficient (R2) and root mean square error (RMSE) for calibration and prediction sets were calculated to evaluate multivariate calibration models. The importance of methods was studied and discussed. The mean square error (MSE) values of Piecewise-Linear rPCA-MARS for TAN and TBN were 1.587 * 10−4 and 0.009, respectively. The Piecewise-Linear rPCA-MARS model can be successfully applied for the analysis of TAN and TBN of crude oil samples based on the obtained results. The fast and easy prediction capability of the proposed Piecewise-Linear rPCA-MARS model, as compared to standard methods, for determining TAN and TBN without any sample preparation step is important in the oil industry