Research on multi-heat source arrangement optimization based on equivalent heat source method and reconstructed...

Area of Science:

• Thermal Management
• Computational Fluid Dynamics
• Artificial Intelligence

Background:

• Predictive image generation models like Variational Autoencoders (VAEs) show promise.
• Handling multi-heat source arrangements in thermal systems presents challenges.
• Traditional numerical solutions for conjugate convective heat transfer can be computationally intensive.

Purpose of the Study:

• To propose and validate a novel RFCNN-βVAE model for accurate temperature rise simulation and thermal management.
• To develop an effective deep learning approach for optimizing multi-heat source configurations.
• To provide a reliable alternative to traditional numerical methods for 2D conjugate convective heat transfer problems.

Main Methods:

• Developed a Residual-connected Fully Connected Neural Network with Variational Autoencoder (RFCNN-βVAE) model.
• Integrated analytical solutions, polynomial fitting, and temperature field superposition for simulation.
• Utilized multiple equivalent heat sources to model adiabatic boundary conditions.
• Adjusted the β parameter to balance reconstruction accuracy and latent space generalization.

Main Results:

• The model achieved high predictive accuracy in simulating temperature rise distribution.
• The RFCNN-βVAE model successfully optimized heat source coordinates, minimizing peak temperature rise.
• During optimization, 73.4% of results surpassed the dense dataset benchmark.
• Demonstrated the effectiveness of deep learning in thermal management system design.

Conclusions:

• The proposed RFCNN-βVAE model offers a feasible and effective deep learning solution for thermal management.
• This research provides valuable insights for optimizing complex thermal environments with multiple heat sources.
• The study validates the application of AI in designing efficient thermal systems, such as multi-beam lighting equalization.