Prediction of Lubrication Performance of Hyaluronic Acid Aqueous Solutions Using a Bayesian-Optimized BP Network

The present study proposes a Bayesian-optimized back-propagation (BP) neural network framework for predicting the tribological performance of hyaluronic acid (HA) aqueous solutions under hydrodynamic lubrication conditions. The model addresses the complex rheological behavior of HA and limitations of traditional trial-and-error methods. It integrates four operational parameters—applied load, sliding speed, fluid viscosity and contact surface inclination. These enable the simultaneous prediction of two critical lubrication characteristics: film thickness and load-carrying capacity. Bayesian optimization was used to automate hyperparameter tuning. This can significantly improve computational efficiency. The optimized model showed a coefficient of determination (R2) of 0.938 and a mean square error (MSE) of 0.0025 on the test dataset, indicating its ability for accurate prediction. The results indicated a significant positive correlation between HA viscosity and lubrication performance. This framework can be used as a screening tool for HA-based lubricants. The integration of machine learning with biotribology may offer opportunities to improve data-driven approaches to analyzing complex fluid behavior, where traditional models have limitations.