Guest editorial: Data-driven methods for heat transfer and fluid flow

Citation

Ransing, R.S. (2024), "Guest editorial: Data-driven methods for heat transfer and fluid flow", International Journal of Numerical Methods for Heat & Fluid Flow, Vol. 34 No. 8, pp. 2833-2835. https://doi.org/10.1108/HFF-08-2024-946

In recent years, data-driven approaches have become increasingly popular for predicting heat transfer and fluid flow behaviour in complex physical systems. These methods leverage vast amounts of training data to learn patterns and relationships, which can then be applied to new, unseen scenarios. This special issue on Data-Driven Methods for Heat Transfer and Fluid Flow highlights innovative research that expands the boundaries of this evolving field, showcasing advancements in methodologies and their applications across various domains. The 16 papers featured in this special issue are introduced here without any specific order:

1. Two-phase flow regime identification using multi-method feature extraction and explainable kernel fisher discriminant analysis. This paper introduces a novel flow regime identification method that combines dynamic pressure measurements with advanced machine learning techniques. The proposed DWT+KFDA method excels in performance and offers explainability, enhancing our understanding of flow regimes.

2. Prediction of multi-physics field distribution on gas turbine endwall using an optimized surrogate model with various deep learning frames. By improving typical deep learning models such as multi-scale Pix2Pix and Swin-Transformer, this study develops a surrogate model to predict physical fields in gas turbines. The connection between U-Net layers and sample scales is explored, and the models' predictions are experimentally validated.

3. Swirl-induced motion prediction with physics-guided machine learning utilizing spatiotemporal flow field structure. This research proposes a framework for predicting flow fields, generalisable to various crank angles, cycles and swirl ratios. The framework aims to enhance real-time flow control and reduce the need for extensive in-cylinder flow measurements.

4. Machine learning-based temperature prediction model for two-phase immersion cooling. The R-LSTM-ED model outperforms traditional LSTM models across various forecast ranges, demonstrating significant improvements in temperature prediction for two-phase immersion cooling. The study also identifies optimal sampling periods and window sizes, while noting the model’s sensitivity to noise.

5. Permeability estimation for deformable porous media with physics-informed convolutional neural network. Using PICNN, this study accurately estimates the permeability of deformable porous media, achieving high correlation coefficients. The method significantly reduces processing time while maintaining high prediction accuracy under varying deformation strains.

6. Embedded data distribution on the accuracy of reconstructing the flow field around a train in crosswinds using physics-informed neural network. The application of PINN in reconstructing flow fields around trains demonstrates minimal discrepancy from numerical simulations. The study identifies optimal spatial data density thresholds, ensuring high reconstruction accuracy.

7. Turbo-RANS: straightforward and efficient Bayesian optimization of turbulence model coefficients. This work introduces a novel black-box optimisation procedure for turbulence model coefficients. The turbo-RANS framework is versatile, incorporating various solvers and offering a flexible objective function for calibration.

8. A neural based modelling approach for predicting effective thermal conductivity of Brewer’s spent grain for biomass applications. This paper presents a neural model that outperforms multiple linear regression in predicting the thermal conductivity of brewer’s spent grain. The findings highlight the effects of particle diameter and heating rate on thermal conductivity.

9. Artificial intelligence based droplet size prediction for microfluidic system. A predictive model for droplet size in microfluidic systems is developed using AI, demonstrating readiness for automated closed-loop control. The model is built on experimental data, accounting for variations in viscosity and flow rates.

10. Applying artificial neural networks and conventional approaches to predict fluidization design parameters of biomass particles. This study leverages machine learning techniques to enhance neural model performance in predicting fluidisation parameters of biomass particles, comparing results with conventional correlations and multiple regression equations.

11. An artificial intelligence approach for the estimation of conduction heat transfer using deep neural networks. By developing novel DNNs and loss functions, this paper presents a unique approach to estimating conduction heat transfer, shifting from traditional models to machine learning, with promising results for heat transfer physics and beyond.

12. Physics-informed neural networks (PINNs): application categories, trends, and impact. This comprehensive review categorises PINN applications across seven domains, with fluid dynamics and CFD as the primary focus. The paper traces the exponential growth of PINN publications, highlighting their impact on various scientific fields.

13. Data-driven wall modeling for LES involving non-equilibrium boundary layer effects. This research demonstrates the efficacy of data-driven models in predicting both equilibrium and non-equilibrium boundary layers. By incorporating data from transitional regions, the model accurately captures non-equilibrium physics, making precise predictions. This motivates further research to include more non-equilibrium phenomena in the training data to better capture recirculation effects.

14. Physically consistent temperature fields for geophysical inversion based on the parametrized location of an isotherm. Introducing a novel numerical procedure for geophysical inversions, this study ensures physically sound temperature field solutions, tailored for scenarios involving known immersed conditions and energy balance problems.

15. Reinforcement learning for cooling rate control during quenching. Pioneering the application of DRL in controlling complex heat and mass transfer during quenching, this paper's findings have significant implications for steel cooling processes, potentially reducing irregular residual stresses in critical machinery.

16. Flow control by a hybrid use of machine learning and control theory. Using CNN-AE for fluid flow linearisation, this study designs a feedback control law, marking the first attempt at integrating machine learning with control theory for transient fluid flow management.

I extend my sincere gratitude to all the authors who submitted their work for this special issue. The selection process was challenging, and many commendable works could not be included due to space constraints. I also wish to thank the anonymous reviewers whose diligent efforts and insightful feedback were crucial in shaping this issue. Without their contributions, this endeavour would not have been possible.

This special issue highlights the potential of data-driven methods in heat transfer and fluid flow, paving the way for more efficient, accurate and innovative solutions in engineering and scientific applications. The contributions represent a significant advancement in using machine learning and neural networks to address complex physical phenomena.