Deep Learning Data-Driven Model for Stribeck Curve Prediction of Lubricated Tribo-Pairs

The prediction of tribological behavior in engineering systems based on output data streams is often considered empirical, as it is difficult to fully describe the underlying processes using closed-form mathematical models. For instance, accurately predicting the coefficient of friction (COF) across the three major lubrication regimes using a single model remains elusive. Machine learning (ML) offers a powerful data-driven approach for modeling complex tribological processes by learning nonlinear relationships between operating conditions and performance metrics. This work develops a high-accuracy deep learning model for predicting the COF in lubricated tribosystems across boundary, mixed, and hydrodynamic lubrication regimes. Extensive experimental data were collected using a ball-on-disk tribometer, which generated a large and diverse dataset spanning multiple materials, lubricants, temperatures, loads, and sliding speeds and covered the full Stribeck curve. These data were then used to train and optimize a neural network capable of accurately reproducing frictional transitions across lubrication regimes. The breadth of the dataset not only allows for the full Stribeck behavior to be captured, but also enables generalization across distinct tribosystems. The resulting model demonstrates strong predictive performance and is deployed using a graphical user interface as a practical tool for estimating COF in ball-on-disk lubricated sliding contacts.