Using Artificial Neural Networks to Accelerate Thermo-elastohydrodynamic Lubrication Simulations

Thermo-elastohydrodynamic lubrication (TEHL) simulations have long been used to predict the lubrication performance of various engineering systems. The aim of this study is to test the performance of artificial neural networks (ANNs) in predicting important parameters in tribological systems and to test the speed of hybrid simulation frameworks that combine ANNs with classical finite volume method (FVM) approaches to solve TEHL problems for point contacts. Six different ANNs were trained to predict values of rigid separation between the contacting bodies, maximum fluid temperature, mid-film friction coefficient, as well as profiles of hydrodynamic pressure, liquid film friction and fluid temperature. In total, 600 classical TEHL simulations were used as training and testing data for the ANNs. The ANN predictions were very accurate. The rescaled R^2 values (in the range of 0 to 1), indicating the correlation between the ANN predictions and classical TEHL results, were 0.997 for the rigid separation, 0.985 for the maximum fluid temperature and 0.939 for the friction coefficient. The mean rescaled R^2 values across all testing samples were 0.997 for both pressure and liquid film fraction profiles and 0.873 for the fluid temperature profiles. Using ANNs to initialise the rigid separation, pressure and liquid film fraction profiles and the temperature profile in each simulation results to significant reduction in simulation times with the mean simulation time across all testing samples being about 40% less than the mean time of the classical TEHL simulations, and for individual simulations, the hybrid frameworks being up to three times faster.