Wear volume estimation for a journal bearing dataset

Abstract

Abstract To develop a system for predicting the remaining useful lifetime of a journal bearing, it is necessary to monitor the progressive wear quantitatively. For this purpose, we create a dataset where the wear volume is tracked throughout several experiments. The roundness profile is used to determine the wear volume over the entire life of the journal bearing. Therefore, a procedure for tracking the wear volume is described. The uncertainty of the procedure is analyzed. It is shown that the procedure has good accuracy and that the uncertainty is induced by the manual setting of the measuring positions. It has been shown that acoustic emission can be used to classify different friction states and identify defects in journal bearings. In addition, it has been demonstrated in experimental setups that it can be used to estimate the wear volume of sliding lubricated metallic contacts. Several experiments were carried out under different operating conditions for the dataset’s creation. Finally, the root mean square value of the acquired acoustic emission signal is used for estimation. Linear regression, random forest regressor, multilayer perceptron, and recurrent neuronal network are applied. The wear volume can be estimated with a root mean square error of 0.32 mm3 and a coefficient of determination of 93 %. Neural networks have the distinct advantage of being able to estimate wear at any point during an experiment.