Deep learning-based approach to R-134a bubble detection and analysis for geothermal applications

M. Ahmed*, A. Habib, M. M. Nawal, M. M.H. Saikot, M. A.H. Chowdhury, M. A. Hoque, A. K.M. Asaduzzaman, H. Pálsson, P. Björnsson

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

This study utilized deep learning to analyse bubble images of R134a fluid produced on bare and novel oleophilic coated stainless steel heat exchanger plates to compare the heat transfer performance in terms of bubble parameters. This work was carried out as a part of the GeoHex project. The vapour bubbles of the boiling R134a fluid, produced on two types of heat exchanger plates, were captured with a high-speed camera during a boiling experiment. A deep learning algorithm based on the Mask R–CNN model was trained for bubble detection in the experimental videos. The algorithm achieved a bubble detection accuracy of 61.22% and 78.6% for the coated and uncoated plates, respectively. The lower accuracy for the coated plate could be attributed to the challenge of distinguishing bubbles from the complex coating pattern. The algorithm determined the number of detected bubbles, bubble sizes, and centroids per frame, tracking each bubble frame by frame. The tracking results were used to calculate the bubble parameters such as departure frequency, departure diameter, and active nucleation site density. The calculated departure diameter for both plates exhibited a difference of 3%, which corresponded closely to the experimental results showing a 4% variation in heat transfer coefficient, which was measured under identical experimental conditions utilizing the GeoHex rig. However, the calculated departure frequency and active nucleation site density exhibited a 17% disparity between the coated and uncoated plates, primarily stemming from the lower detection accuracy for the coated plate. Some limitations of the trained model were identified in accurately detecting very small, out of focus, and mutually occluded bubbles. Future research should focus on improving experimental rig designs to capture better side-view images and develop advanced machine learning or other techniques to address these limitations.

Original languageEnglish
Article number103377
JournalCase Studies in Thermal Engineering
Volume49
DOIs
Publication statusPublished - 1 Sept 2023

Bibliographical note

Publisher Copyright:
© 2023

Other keywords

  • Bubble detection
  • Bubble parameters
  • Deep learning
  • Geothermal energy
  • Heat exchanger
  • Image processing
  • Mask RCNN model
  • R134a bubble

Fingerprint

Dive into the research topics of 'Deep learning-based approach to R-134a bubble detection and analysis for geothermal applications'. Together they form a unique fingerprint.

Cite this