Assessment of artificial neural network for thermohydrodynamic lubrication analysis

In this study, artificial neural networks (ANNs) are constructed and validated by using the bearing data generated numerically from a thermohydrodynamic (THD) lubrication model. In many tribological simulations, a surrogate model (meta-model) for obtaining a fast solution with sufficient accuracy is highly desired.

The THD model is represented by two coupled partial differential equations, a simplified generalized Reynolds equation, considering the viscosity variation across the film thickness direction and a transient energy equation for the 3-D film temperature distribution. The ANNs tested are having a single- or dual-hidden-layer with two inputs and one output. The root-mean-square error and maximum/minimum absolute errors of validation points, when comparing with the THD solutions, were used to evaluate the prediction accuracy of the ANNs.

It is demonstrated that a properly constructed ANN surrogate model can predict the THD lubrication performance almost instantly with accuracy adequately retained.

This study extends the use of ANNs to the applications other than the analyses dealing with experimental data. A similar procedure can be used to build a surrogate model for computationally intensive tribological models to have fast results. One of such applications is conducting extensive optimum design of tribological components or systems.

The peer review history for this article is available at: https://publons.com/publon/10.1108/ILT-03-2020-0109/