Flow-field-structure-informed prediction of churning torque in dip-lubricated gear

The gear-churning flow field significantly affects gearbox lubrication and efficiency. Yet the evolution mechanisms across wide temperature ranges remain poorly understood, limiting lubrication design for underwater vehicles and renewable energy systems. The study links vision-based flow fields to churning torque through a structure-to-torque pathway. A temperature-adjustable visualization test rig for oil flow is built to capture flow field evolution and measure churning torque. The reverse hill-shaped oil domain is parameterized by width c ₁ and height h _1, and the oil domain volume is approximated as Φ ≈ κ _s ·b·c ₁ ·h ₁. A mechanistic-data hybrid model is constructed to predict ( c ₁, h ₁) based on ( v , μ , σ , θ ). The model integrates analytical formulations classified by viscosity regime within a physics-informed neural network (PINN) framework. A shape factor is introduced into Changenet’s classical churning torque calculation theory. Consequently, an improved prediction strategy is developed to predict churning torque based on the flow field structure. Prediction accuracy comparable to that of a standalone PINN is achieved by the hybrid model. Physical consistency is preserved under extrapolation to −38 °C and 100 °C. The predicted c ₁ and h ₁ values can be used as inputs for the improved churning-torque prediction model. The proposed strategy links input parameters, flow-field structures, and churning torque. The proposed strategy can reduce the dependence on costly experiments and laborious numerical simulations, providing a methodological foundation for torque prediction in engineering transmissions.