Prediction of Lubricant Service Life Using Data Mining to Improve Reliability of Water Injection Pumps in Crude Oil Production Facility

Abstract

Predicting service life of lubricant based on lubricant condition data can help identifying accelerated wear indicating reliability problem in machinery. Data mining has been implemented for prediction in various fields including in petroleum industries to support decision making during the reliability improvements of crude oil production facilities. This research aimed to predict the lubricant service life of water injection pumps using several data mining algorithms, such as Linear Regression, Multilayer Perceptron (MLP), Gradient Boosting, and Random Forest. As results, the predicted service life values have a good correlation with the actual service life. The Gradient Boosting and Random Forest algorithms show better performance, with R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.71 and 0.74, comparing with the Linear Regression and MLP algorithms ($R^{2}=0.3$ and 0.58), respectively. The RMSE values of Gradient Boosting and Random Forest algorithms (29.25 and 27.59) also show smaller errors than the two other algorithms (45.71 and 35.37). The results also confirmed that the Random Forest algorithm is slightly better than the Gradient Boosting algorithm. The decision tree of the prediction rule also can be shown by the Random Forest algorithm.