Neural network based prediction of lemon bore journal bearing performance with micropolar lubricants

Abstract

The research paper aims to predict the static characteristics of lemon bore journal bearings lubricated with micropolar lubricant using an artificial neural network (ANN) and compare the results using traditional numerical methods. It commences by employing the Finite Difference Method (FDM) to solve the Reynolds equation for micropolar lubricants. Swift-Stieber boundary conditions were functional to evaluate pressure distribution, which enables the examination of the static characteristics by varying the eccentricity ratio (ε) = 0.25, 0.3, and 0.35, length of micropolar (l_m) = 1 to 50, and coupling number (N²) = 0.1 to 0.9. Thereafter, the results from the FDM approach are used to train artificial neural network models. The study utilizes a dataset comprised of over 1560 data points with corresponding inputs and outputs. Three distinct feed-forward ANN architectures were considered: Levenberg-Marquardt optimization technique, Bayesian regularization technique, and Scaled Conjugate Gradient, and their performance was compared with results obtained through FDM. Results indicate a substantial reduction in computational effort and time when employing the trained ANN models compared to the FDM approach. Furthermore, it is found that the difference between the results given by the ANN models and the FDM results is not above 5%; the agreement between these two approaches thus appears good. This study reveals the effectiveness of ANNs in reducing computational time and effort while maintaining accuracy in evaluating the performance of lemon bore journal bearings lubricated with micropolar lubricant, and provides a new approach to predict the performance of fluid film bearings.