Using multiple linear regression to predict engine oil life.

This paper deals with the use of multiple linear regression to predict the viscosity of engine oil at 100 °C based on the analysis of selected parameters obtained by Fourier transform infrared spectroscopy (FTIR). The spectral range (4000-650 cm⁻¹), resolution (4 cm⁻¹), and key pre-processing steps such as baseline correction, normalization, and noise filtering applied prior to modeling. A standardized laboratory method was used to analyze 221 samples of used motor oils. The prediction model was built based on the values of Total Base Number (TBN), fuel content, oxidation, sulphation and Anti-wear Particles (APP). Given the large number of potential predictors, stepwise regression was first used to select relevant variables, followed by Bayesian Model Averaging (BMA) to optimize model selection. Based on these methods, a regression relationship was developed for the prediction of viscosity at 100 °C. The calibration model was subsequently validated, and its accuracy was determined using the Root Mean Squared Error (RMSE) metric, it was 0.287. Finally, the obtained model was used to predict the lifetime of engine oil in diesel engines operating under severe conditions.