Regression Models in Lubricant Oil Quality Assessment

After a suitable data analysis, we adjust two types of regression models to support decision-making when monitoring lubricants in an industrial environment. The target variable is the remaining useful life of a lubricant, as it is an essential process in preventing damage to associated equipment. The main objective of this work is to predict at an early stage how long will an oil load still perform correctly. The first model is a multiple linear regression, adjusting coefficients using cross-validation on several subgroups of the data set. Though the performance of this model is satisfactory, the corresponding adjustment is compromised by the fact that the observations are not independent, as they are grouped taking into account the oil loads to which they belonged. For this reason, a second type of regression was adjusted: a linear regression model with mixed effects. After adjusting the models, a Python algorithm was created to determine an optimal interval regarding the oil service hours after which the lubricant should be replaced.