Inverse Heat Convection Methods Using Physics-Informed Neural Networks

Professor Introduction

 X. X | Ph.D. in Engineering

Home Institute：University of Melbourne

[ Research Interests ] Application of machine learning in scientific computing、Turbulence modeling、Structural wind engineering
[ Additional Experience ] Member of the Australian Fluid Mechanics Society、Member of the Chinese Society of Mechanics、Participation in multiple domestic and international research projects

Project Description

This research project focuses on the development of inverse methods for heat convection using cutting-edge deep learning algorithms, specifically Physics-Informed Neural Networks (PINNs). The aim is to construct partial differential equations (PDEs) governing fluid flow and heat transfer, and subsequently perform inverse analysis to derive flow and temperature fields from given input data.

Heat convection, a fundamental process in fluid dynamics and thermal sciences, involves the transfer of heat through fluid motion. Accurate modeling and inversion of heat convection are crucial for various applications, including climate modeling, industrial processes, and energy systems.

PINNs are a novel class of neural networks that incorporate physical laws, expressed as PDEs, directly into the learning process. By embedding these physical constraints, PINNs can effectively solve forward and inverse problems in fluid dynamics and heat transfer with greater accuracy and efficiency compared to traditional methods.

Project Keywords

Project Outline

Part 1 :  Introduction to Heat Convection and PINNs
• Overview of heat convection and its significance in fluid dynamics.
• Introduction to Physics-Informed Neural Networks (PINNs).
• Comparison of PINNs with traditional neural networks and numerical methods.

Part 2 : Governing Equations and Physical Constraints
• Detailed exploration of the partial differential equations (PDEs) governing fluid flow and heat transfer.
• Discussion on the physical constraints and boundary conditions relevant to heat convection.
• Formulation of the inverse problem for heat convection.

Part 3 : Data Collection and Preprocessing
• Identification of relevant datasets for flow and temperature fields.
• Data preprocessing techniques, including normalization and feature extraction.
• Handling of noisy and incomplete data.

Part 4 : Construction and Training of PINNs
• Development of PINNs for solving the inverse problem of heat convection.
• Training and validation of PINNs using collected datasets.
• Hyperparameter tuning and model optimization.

Part 5 : Model Evaluation and Analysis
• Evaluation of PINN performance using standard metrics.
• Comparison of PINNs with traditional inverse methods.
• Analysis of model predictions for various heat convection scenarios.

Part 6 : Case Studies and Practical Applications
• Application of the developed PINNs to real-world heat convection problems.
• Case studies demonstrating the improved accuracy and efficiency of PINNs.
• Discussion of practical implications and potential areas for further research.

Suitable for

• High School Students interested in the application of deep learning in scientific research.
• Undergraduate and graduate students in fluid dynamics, heat transfer, and computational mechanics