An efficient ANN-FTSI model for predicting the lubrication performance of misaligned journal bearings under heavy-load and low-speed conditions

Abstract

The accurate and efficient prediction of misaligned journal bearing performance under heavy-load and low-speed operating conditions is of paramount importance for practical applications. However, the multiphysics simulation of fluid-thermal-structure interaction (FTSI) requires substantial computational time. This study focused on misaligned journal bearings under heavy-load and low-speed conditions and constructed an FTSI computational fluid dynamics model. Following this, a particle swarm optimization (PSO) algorithm was systematically implemented to enhance the predictive accuracy of an artificial neural network (ANN) through architectural refinement and hyperparameter tuning. The optimized ANN framework demonstrated robust capability in predicting critical tribological parameters, including the journal bearing shaft position, tilt angle, minimum oil film thickness, and maximum hydrodynamic pressure distribution within the contact interface. In terms of computational efficiency, ANN-FTSI model required only 10 s for prediction, which is in stark contrast to the 2–10 h needed by the FTSI model alone. By comparing the prediction outcomes of the ANN-FTSI model with FTSI simulation results, it was determined that the errors of all output parameters were below 5%, thereby validating the high accuracy of the ANN-FTSI model.