Abstract:The purpose of the present work is to examine two possibilities; firstly, predicting equivalent sand-grain roughness size $k_s$ based on the roughness height probability density function and power spectrum leveraging machine learning as a regression tool, and secondly, extracting information about relevance of different roughness scales to skin-friction drag by interpreting the output of the trained data-driven model. The model is an ensemble neural network consisting of 50 deep neural networks. The data for the training of the model is obtained from direct numerical simulations (DNSs) of turbulent flow in plane channels over 85 irregular multi-scale roughness samples at friction Reynolds number Re$_\tau=800$. The 85 roughness samples are selected from a repository of 4200 samples, covering a wide parameter space, through an active learning (AL) framework. The selection is made in several iterations, based on the informativeness of samples in the repository, quantified by the variance of ENN predictions. This AL framework aims to maximize the generalizability of the predictions with a certain amount of data. This is examined using three different testing data sets with different types of roughness, including 21 surfaces from the literature. The model yields an overall mean error of 5\% to 10\% on different testing data sets. Subsequently, a data interpretation technique, known as layer-wise relevance propagation, is applied to measure the contributions of different roughness wave-lengths to the predicted $k_s$. High-pass filtering is then applied to the roughness PS to exclude the wave-numbers identified as drag-irrelevant. The filtered rough surfaces are investigated using DNS, and it is demonstrated that, despite significant impact of filtering on the roughness topographical appearance and statistics, the skin-friction coefficient of the original roughness is successfully preserved.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to predict the frictional drag caused by any rough surface under turbulent conditions. Specifically, there are two research objectives: 1. **Predict the equivalent sand - grain roughness size**: Based on the probability density function (p.d.f.) of roughness height and the power spectrum (PS), use machine - learning methods (especially regression tools) to predict the equivalent sand - grain roughness size $ k_s $. The equivalent sand - grain roughness size $ k_s $ refers to the sand - grain size that produces the same frictional drag coefficient as a specific rough surface in a fully rough state. 2. **Identify the correlation of different roughness scales to frictional drag**: By interpreting the output of the trained data - driven model, extract information about the influence of different roughness scales on frictional drag. This includes using the layer - wise relevance propagation technique to measure the contribution of different rough wavelengths to the prediction of $ k_s $, and applying a high - pass filter to exclude irrelevant wavenumbers. ### Background and Motivation Surface roughness is an important factor in engineering applications, especially in applications involving fluid flow. For example, the roughness of a ship's hull due to bio - fouling can significantly increase frictional drag, resulting in billions of dollars of energy waste each year. Therefore, studying the frictional drag of rough surfaces under turbulent conditions has important practical significance. ### Research Methods 1. **Data Generation**: Constructed an unlabeled roughness "library" $ U $ containing 4,200 irregular multi - scale rough surfaces. These surfaces are generated by mathematical methods, and their probability density function (p.d.f.) and power spectrum (PS) can be controlled. 2. **Direct Numerical Simulation (DNS)**: Select 85 rough surfaces from the library $ U $ for DNS simulation to determine the equivalent sand - grain roughness size $ k_s $ of each surface. These data are used to train the machine - learning model. 3. **Active Learning (AL) Framework**: Adopt an active learning method to select training data to maximize the generalization ability of the model. In the initial stage, 30 samples are randomly selected for labeling and training, and then in each iteration, 20 new samples with the highest prediction variance are selected for labeling and training. 4. **Machine - Learning Model**: Construct an ensemble neural network (ENN) model consisting of 50 independent neural networks, each with the same structure. The input vector includes the discretized roughness p.d.f. and PS, as well as three additional features ($ k_t/k_{99} $, the normalized maximum and minimum rough wavelengths $ \lambda_0^*=\lambda_0/k_{99} $ and $ \lambda_1^*=\lambda_1/k_{99} $). ### Results and Evaluation 1. **Evaluation of the Active Learning Framework**: By comparing the model trained with data points selected by active learning and the model trained with randomly selected data points, the effectiveness of the active learning framework is verified. The results show that the data points selected by active learning can better improve the performance and generalization ability of the model. 2. **Model Performance**: The model is evaluated on different test data sets, including unseen rough surfaces, natural rough surface scan data, and artificially generated irregular rough surface data. The overall average error of the model is between 5% and 10%. 3. **Roughness - Scale Analysis**: Through the layer - wise relevance propagation technique, the contribution of different rough wavelengths to the prediction of $ k_s $ is measured. The results show that although high - pass filtering has a significant impact on the topological appearance and statistical properties of the rough surface, the frictional drag coefficient of the original rough surface is still successfully retained. ### Conclusion This study shows that by using data - driven methods, combined with active learning and machine - learning techniques, the equivalent sand - grain roughness size $ k_s $ of any rough surface can be effectively predicted, and the influence of different roughness scales on frictional drag can be identified. This provides new tools and methods for understanding and predicting the frictional drag of complex rough surfaces.

Prediction of equivalent sand-grain size and identification of drag-relevant scales of roughness -- a data driven approach

Data-driven prediction of the equivalent sand-grain height in rough-wall turbulent flows

Direct Numerical Simulation of Channel Flow with Real Surface Roughness Using a Ghost Cell Immersed Boundary Method

Drag prediction of rough-wall turbulent flow using data-driven regression

On predictive models for the equivalent sand grain roughness for wall-bounded turbulent flows

Data-driven prediction of the equivalent sand-grain roughness

A machine learning model for reconstructing skin-friction drag over ocean surface waves

Analysis and Development of Novel Data-Driven Drag Models Based on Direct Numerical Simulations of Fluidized Beds

Drag Model from Interface-Resolved Simulations of Particle Sedimentation in a Periodic Domain and Vertical Turbulent Channel Flows

Rough surfaces in under-explored surface morphology space and their implications on roughness modelling

A machine learning-based drag model for sand particles in transition flow aided by spherical harmonic analysis and resolved CFD-DEM

An assessment of effective slope as a parameter for turbulent drag prediction over multi-scaled roughness

Developing Extended and Unscented Kalman Filter-Based Neural Networks to Predict Cluster-Induced Roughness in Gravel Bed Rivers

Development of Data-Driven Filtered Drag Model for Industrial-Scale Fluidized Beds

Neural network-augmented SED-SL modeling of turbulent flows over airfoils

Data-driven discovery of drag-inducing elements on a rough surface through convolutional neural networks

Construction and Analysis of the Mesoscale Drag Force Model Based on Machine Learning Methods

Development of a Filtered CFD-DEM Drag Model with Multiscale Markers Using an Artificial Neural Network and Nonlinear Regression

kεNet: discovering the turbulence model and applying for low Reynolds number turbulent channel flow

High Reynolds number airfoil turbulence modeling method based on machine learning technique

Application of Deep Learning Method to Reynolds Stress Models of Channel Flow Based on Reduced-Order Modeling of DNS Data