Classification of boiling regimes, fluids, and heating surfaces through deep learning algorithms and image analysis
Year: 2025
Type: article
Abstract: Accurate monitoring and prediction of boiling phenomena is crucial for reliable operations near critical conditions in thermal systems. In this regard, this study focuses on applying Convolutional Neural Network (CNN) algorithms for image classification to identify boiling regimes. Specifically, images used for training the Deep Learning models were acquired based on a Flow-Boiling experimental setup with Joule effect heating. Four distinct CNN architectures were employed to classify boiling phenomena under different operating conditions, covering single-phase flow, nucleate boiling, and pre-critical heat flux states. The study also examines the robustness of the models in classifying images according to the working fluid and the heating plate (non-textured and micro-textured). The performance of each CNN was evaluated using precision, recall, and F-score metrics and compared to a previously reported methodology involving downsampling, Principal Component Analysis, and a Support Vector Machine. The findings highlight the effectiveness of CNN algorithms in accurately classifying boiling regimes from images, outperforming the previous methodology. Notably, InceptionNet achieved an F-score of around 98% for plate type classification. AlexNet showed the best performance for both boiling regime and fluid type classification, with F-scores over 95%. In contrast, the previous methodology achieved a maximum F-score of around 88% for these classifications.
Institution: Universidade de Vigo
Language: English
Subfield: Mechanical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,220
Award: G4693101584
OpenAlex ID: W4407765763