Triboinformatic Modelling for Friction Prediction using Machine Learning Algorithms /
Muhammad Talha Manzoor
- 101p. Soft Copy 30cm
The tribology is the study of surface roughness, adhesion, friction, wear, and lubrication of solid surfaces in relative motion. The integration of Tribology with Machine Learning (ML) and other data-driven/informatics methodologies is commonly referred to as Triboinformatics. This dissertation employs triboinformatic approaches for the prediction of coefficient of Friction. This dissertation presents the development of machine learning models, including artificial neural network (ANN), support vector machine (SVM) and gradient boosting machine (GBM) to correlate the coefficient of friction with various tribological parameters. Machine learning investigation indicates that the instantaneous entrainment velocity has the most substantial impact on the coefficient of friction (COF). For enhancement of energy efficiency in automobiles, the reduction of frictional forces is vital. The key areas of focus for friction reduction have been lubricant chemistry, surface coatings, and surface changes. All experiments were performed on a modified reciprocating tribometer encompassing actual cam and tappet shim under actual engine running conditions. The three tappet shim samples, were engraved with texture densities of 5%, 8% and 10%. These textured shims were analyzed in comparison with the original shim and the friction data was obtained at different temperatures. The best friction reduction effectiveness of 18.33% was achieved by 8% textured shim at 90°C. After the experimentations on textured shims, experimentations on various friction modifiers were done using the same tribometer setup. The four different types of friction modifiers including Organic, Moly A, Polymeric and Moly B Friction Modifier were used along with base oil and the trends were observed. For application of ML models, it is required to acquire the suitable dataset. Two types of experimentations are performed to obtain the required dataset and the dataset of experimentations with different friction modifiers is used to train and test the used ML models. The ANN, SVM, and GBM models for coefficient of friction (COF) for camshim contact under lubricated conditions are developed. The most effective predictive xvi performance for COF has been demonstrated by the GBM model. The lubricant entrainment velocity is recognized as the primary variable for predicting coefficient of friction (COF). The values of mean squared error (MSE), mean absolute error (MAE) and coefficient of determination (R2) are obtained and compared for all of the three models. This dissertation illustrates that the Triboinformatic methodologies can be effectively applied in tribology, yielding new insights into structure-property correlations across diverse material classes.