Data-driven methods for heat and fluid flow

Closes:

Submit your paper here!

Introduction 

With advancement in Artificial Intelligence, machine learning and inverse modelling methods, data-driven approaches have become increasingly popular for predicting heat and fluid flow behaviour in complex physical systems. The goal is to transfer knowledge learnt from training data on a particular scenario and apply it to new situations which the model had not seen before. Three prominent data-driven methods for fluid flow modelling are Physics-Informed Neural Networks (PINNs), Graph Network Simulators (GNSs) and Convolutional Neural Networks (CNNs).

PINNs use traditional feed-forward neural network architecture coupled with experimental data and applying loss function that obeys laws of physics and constraints to represent(mimic) real world system. PINNs are particularly useful for problems with sparse or noisy data, where-in traditional physics-based models may struggle. By incorporating physical laws constraints on training data, PINNs can provide accurate predictions even with limited data. On the other hand, GNSs use a different approach where-in a graph structure is used to represent the fluid flow interaction (dynamics). Each node in the graph corresponds to a physical entity (such as a fluid or a boundary particle) with its position as node feature. Physical laws and constraints such as velocity or acceleration can be represented as edge features in the graph structure. GNS can simulate fluid flow behaviour from large datasets of particle motion and are particularly useful for problems with complex interactions. Continuous convolutions have also been proposed on particles instead of using graphs as the underlying representation. CNNs use a set of time sliced velocity field images to predict future frames of velocity fields. It is based on the deep learning ability of convolutional layers to recognise patterns in images. 

All three approaches have shown great potential for accurately simulating heat transfer and fluid flow phenomena. These interactions between different parts of the fluid system, such as turbulent flows or multi-phase flows can be dynamically learned by the model - and produce accurate inference results. However, there are still challenges to be overcome in using these models from generalisation ability to accurately capture complex fluid flow phenomena such as free surface flows and fluid splashing. The GNS models have shown promise in areas such as particle motion data and free surface flow capture however, they require large amount of training data to achieve accurate predictions. PINNs are limited by the quality and quantity of the data used for training and may not generalize well to new or complex situations. CNNs predict fluid flow by visualising it as a time series of images rather than learning physical laws that govern particle interaction.  

Therefore, further research is needed to enhance the performance of data-driven methods to overcome these limitations and improve the accuracy and reliability of predictions for heat and fluid flow applications. 

Starting with this special issue, we aim to create a forum for discussions on the latest research trends and future directions in data-driven methods for heat transfer and fluid flow. We invite researchers to submit their original research and review articles on the following topics:

  • Novel data-driven methods for heat and fluid flow that can capture complex phenomena such as free surface flows, multi-phase flows, fluid-structure interactions, and turbulent flows.
  • Hybrid approaches that combine physics-based models with data-driven methods to enhance the accuracy and efficiency of simulations.
  • Explainable AI techniques that can improve the interpretability of data-driven models for heat and fluid flow.
  • Uncertainty quantification and sensitivity analysis of data-driven models to enhance the reliability and robustness of predictions.

We welcome contributions from researchers in academia and industry, as well as interdisciplinary work that combines data-driven methods with other areas of research including inverse modelling.

 

Guest Editor


Dr Rajesh Ransing,
Swansea University, UK,
[email protected]

 

Submissions Information

Submissions are made using ScholarOne Manuscripts. Registration and access are available by clicking the button below.

Submit your paper here!

Author guidelines must be strictly followed.

Authors should select (from the drop-down menu) the special issue title at the appropriate step in the submission process, i.e. in response to “Please select the issue you are submitting to”. 

Submitted articles must not have been previously published, nor should they be under consideration for publication anywhere else, while under review for this journal.

 

Key deadlines


Opening date for manuscripts submissions: 1 July 2023
Closing date for manuscripts submission: 31 October 2023