Machine Learning model for tribological data extraction from experimental tests

Nowadays, increasing demand for performances and new environmental requirements have led to the need for new materials and lubricants in the fluid power field. The thesis is based on research carried out in the offices at the IFAS institute. Based on existing measurement data of a disc-on-disc tribometer, the shapes of the Stribeck curves were analysed. Furthermore, the essential characteristics of the curves were extracted, such as the minimum and maximum coefficient of friction, the speed at the curve's minimum point, the number of peaks, and the slopes of the curve. Some of the parameters and the experimental setup have the purpose of feeding a machine learning model. After tuning and comparing different models, the Random Forest regression model was chosen. Finally, predictions with new materials and lubricants were performed. In parallel, some significant comparisons between materials and lubricants were investigated. The latter does not need to create new specimens for the test bench, significantly saving time and money. Further research, increasing the sample size, is needed to improve the method and to get more accurate predictions.