End‐To‐End Deep Learning Temperature Prediction Algorithms of a Phase Change Materials From Experimental Photos
ABSTRACT A Phase‐change material (PCM) experiences irregular shape and nonlinear temperature changes at different locations during the melting process; these parameters provide valuable information on the characteristics of the PCM. Traditional explicit image processing, statistics, and mathematical...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2025-06-01
|
| Series: | Energy Science & Engineering |
| Subjects: | |
| Online Access: | https://doi.org/10.1002/ese3.70110 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | ABSTRACT A Phase‐change material (PCM) experiences irregular shape and nonlinear temperature changes at different locations during the melting process; these parameters provide valuable information on the characteristics of the PCM. Traditional explicit image processing, statistics, and mathematical techniques may be used to estimate the temperature of the PCM photos, but these methods have limitations such as high inaccuracy, no generalization, and complexity. Here, temperatures at different locations inside the PCM have been calculated by using the shape of melting PCM with the aid of deep learning. An experimental setup was built to melt the PCM under constant wall temperature and a conventional digital camera took temporal photos of the phase change. Four end‐to‐end networks have been developed to use the captured photos as input and report temperatures of the PCM as output. Initially, the networks were built using different convolutional layers and weights for feature extraction, and then the fully connected layers extracted the temperature profiles of the PCM. Comparison of the networks shows that MobileNets based Weights IV – Deep Neural Network (WIV‐ DNN) detects the temperature at different locations of the PCM successfully with an average error of less than 0.9% during the whole melting process in 0.03 s. This temperature measurement method is cost‐effective, independent of thermographic cameras, accurate, fast response, and can be updated for other related applications in industries and scientific studies. All programs and datasets are available on GitHub. |
|---|---|
| ISSN: | 2050-0505 |