Prediction of viscosity index of motor oils using FTIR spectral data and chemometrics

In order to ensure the quality of motor oils by measuring viscosity index (VI), regulatory agencies and producers need a more precise, easy and cost effective method for monitoring the qualities. Multivariate data analysis based on Fourier transform infrared (FTIR) spectroscopy was reported in this work as an alternative for measuring viscosity index of motor oils. 27 samples of motor oils of different brands were collected from different regions of Bangladesh. Viscosity index of the samples were first determined by the conventional technique. Savitzky-Golay (S-G), smoothing and mean normalization are the three distinct data preprocessing methods and these were assessed to measure their efficiencies by applying them in developing calibration procedures prior to modeling. Artificial neural network (ANN), principal component regression (PCR) and partial least-square regression (PLSR) were then developed using these processed data for determination of viscosity index of motor oils. Results showed that PCR performed best when it used Savitzky-Golay smoothing data. Performance of PLSR was slightly more than that of PCR (R2≈ 98%). PLSR (R2≈ 99%) had better predictive performance comparing to ANN (R2≈ 97%). Among the calibration techniques studied here, PLSR showed the best prediction results with Savitzky-Golay smoothed FTIR spectral data. The method proposed here to determine viscosity index of motor oils requires less staff dedication, shorter turnaround times and lower expenses than conventional approaches.