Many tribological problems are expensive to solve due to the requirement of heavy and iterative computation. In this study, a direct mapping between design variables and merits is obtained using a neural network. Data from limited numerical simulations of the behaviors of fluid film of a slider in the thermohydrodynamic regime were used to train the network, which replaces further lengthy simulations. A balance between the modeling accuracy and generalization capability of the network is achieved by optimizing the network size using an efficient optimization scheme, DIviding RECTangles (DIRECT). Performance comparison of the optimization method is based on a hybrid learning strategy that combines the steepest descent and Levenberg-Marquardt method.

This content is only available via PDF.
You do not currently have access to this content.