Data-driven prediction of the equivalent sand-grain roughness.

Area of Science:

• Fluid Dynamics
• Computational Fluid Dynamics (CFD)
• Surface Science

Background:

• Surface roughness significantly impacts near-wall fluid velocity and drag.
• Equivalent sand-grain roughness (k_s) is crucial for fluid dynamics modeling.
• Existing k_s prediction formulas lack universality and are often limited to specific roughness types.

Purpose of the Study:

• To develop a more accurate and universal method for predicting equivalent sand-grain roughness (k_s).
• To evaluate the performance of a Particle Swarm Optimized Backpropagation (PSO-BP) model against traditional methods.
• To provide improved solutions for computational fluid dynamics (CFD) and engineering applications.

Main Methods:

• Utilized a Particle Swarm Optimized Backpropagation (PSO-BP) model to predict k_s.
• Trained the model using surface parameters from DNS, LES, and experimental data for diverse roughness.
• Compared PSO-BP performance against existing roughness correlation formulas and a standard Backpropagation (BP) model.

Main Results:

• The PSO-BP method achieved superior performance with a Mean Absolute Error (MAE) of 0.0390, Mean Squared Error (MSE) of 0.0026, and Mean Absolute Percentage Error (MAPE) of 28.12%.
• PSO-BP outperformed traditional BP models and existing roughness correction formulas in prediction accuracy.
• An optimized polynomial function was also proposed as a transparent predictive model.

Conclusions:

• The PSO-BP approach offers a more precise and efficient method for estimating equivalent sand-grain roughness.
• This novel approach enhances the accuracy of CFD simulations and engineering calculations.
• The findings promise more effective solutions for fluid dynamics problems involving surface roughness.