Advanced deep learning models for comprehensive analysis and optimization of nucleate boiling heat transfer systems
|
Full Text |
Pdf
|
|
Author |
Mohammad Saraireh, Falah Al-Saraireh and Yassin Nimir
|
|
e-ISSN |
1819-6608 |
|
On Pages
|
912-927
|
|
Volume No. |
19
|
|
Issue No. |
14
|
|
Issue Date |
October 12, 2024
|
|
DOI |
https://doi.org/10.59018/072422
|
|
Keywords |
nucleate boiling, heat transfer coefficient, deep learning, computational fluid dynamics, critical heat flux, thermal management.
|
Abstract
For heat transfer systems used in nucleate boiling, deep learning-based solutions are required. This is a vital step to avoid the limitations of observational and experimental data. This study aims to combine Generative Adversarial Networks (GANs), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs) to enhance computational fluid dynamics (CFD) design. Better designs, such as Visual Geometry Group (VGG16), can extract hierarchical features, while CNNs can distinguish dynamic events such as bubble formation and growth in response to temperature changes. Because it uses current data, this strategy does not require large datasets. Images are processed to recover bubble statistics and physical data using mask R-CNN and other advanced object identification techniques. This approach ties bubble activity to heat flow parameters, resulting in a consistent sample for examination. To better understand boiling processes, the research suggests a dual method. This approach uses CNNs for feature extraction and Multi-Layer Perceptron (MLP) networks for data processing. This technique produces deep learning models and robust optimization tools while also advancing our knowledge of nucleate boiling. In the experimental setup, pool boiling is investigated using one of the precisely constructed heating tanks. This tank enables us to maintain consistent heat transfer when photographing with high-speed cameras. More improved imaging techniques are required for precise observations and analysis, as high-speed camera images demonstrate how minor variations in heat flow can impact bubble dynamics. The whole process of developing, training, and testing deep learning models includes refining data, segmenting instances using Mask R-CNN, and generating hybrid features by merging Mask R-CNN with CNN (VGG16). This strategy may lead to the creation of a regression model capable of reliably predicting heat flow in boiling water tests. The goal is to make the model easier to use and understand. It appears that the study of heat flow prediction and boiling dynamics covered a lot of ground. The average bubble size increases linearly from 0.5 to 5.0 mm when the heat flow increases from 10 to 100 kW/m2. This suggests an increase in heat flow. When bigger heat fluxes are present, the standard deviation increases, showing that bubble diameters might vary significantly. The recommended deep learning models proved to be very predictive. This study has greatly improved our knowledge and use of nucleate-boiling heat transport systems. The findings show that deep learning models may incorporate theoretical and practical components, resulting in more dependable and efficient thermal management systems.
Back