Development of a machine learning-based surrogate model for friction prediction in textured journal bearings

Surface textures in journal bearings offer significant potential for reducing friction and enhancing energy efficiency. However, the complexity of texture configurations necessitates an accurate and efficient performance prediction model to properly design textured journal bearings. To address this issue, this study develops a machine learning (ML)-based surrogate model to predict friction in textured journal bearings. First, computational fluid dynamics (CFD) models employing a dynamic mesh algorithm are developed to generate accurate data sets. Furthermore, three ML methods are trained and compared to select the most suitable prediction method: artificial neural network (ANN), support vector regression (SVR), and Gaussian process regression (GPR). Among these ML methods, ANN shows the best prediction performance. Given the high computational cost of CFD simulations, the prediction accuracy of the ANN-based surrogate model is further enhanced without the need for additional data sets. This enhancement is achieved through an architecture design based on cross-validation and further optimization utilizing the genetic algorithm. Eventually, the average prediction accuracy is improved to 98.81% from 95.89%, with the maximum error reduced to 3.25% from 13.17%. These findings demonstrate the potential of ML in the performance prediction in textured journal bearings and provide a promising approach for broader applications in developing highly efficient and accurate ML-based surrogate models, particularly in cases with limited available training data sets.