A unique deep learning-based image recognition technique for the extensible implementation of on-demand defrosting control has been presented in a recent study published in the journal Applied Energy.
To increase recognition accuracy, the researchers utilized a convolutional neural network (CNN) model to extract intricate and difficult features for frosty state detection. Labor-intensive labeling tasks and time-consuming experiments can be reduced by incorporating deep clustering and image augmentation,
Periodical Defrosting Control
Commercial energy systems significantly contribute to global energy consumption and are essential for providing operation power and assuring comfort.
The most common issue with heat exchangers in commercial energy systems is frost formation.
Frost inevitably begins to form and deposit on heat exchanger surfaces once the temperature falls below the dew point of air and freezing point of water.
The buildup of frost layers compromises the operational effectiveness and stability of energy systems, which obstructs airflow and raises the thermal resistance of heat exchangers. To optimize the performance of energy systems and restore heat exchangers to their original capacity, periodic control for implementing defrosting cycles is required.
Potential of Image Recognition-Based Defrosting Control
Periodical defrosting is necessary to return heat exchangers to their original performance and increase the operational efficacy of commercial energy systems.
A defrosting control method utilizing image recognition technology is a low-cost and user-friendly technique for implementing demand-based defrosting cycles.
The image processing technology detects the defrosting state. It is more affordable and user-friendly compared with other direct measurement techniques currently in use.
Image Recognition-Based Defrosting Control Limitations
Developing a high-accuracy image recognition model is time-consuming and costly in light of various operating environments, which results in low extensibility between homotypic and heterotypic devices and severely restricts its practical application in commercial energy systems.
Although there have been attempts to develop defrosting state detecting technology based on image recognition, there are still specific challenges.
Multiple tests are carried out to collect and process the image samples to increase the resilience and accuracy of frost detection under various angles and illumination intensities. This results in high cost and labor usage.
Combined control over numerous devices is generally overlooked in commercial energy systems in favor of concentrating on creating a defrosting control plan for a single unit. These drawbacks result in a lack of extensibility between homotypic or heterotypic energy systems, severely hindering the commercial use of on-demand defrosting control.
Development of Deep-Learning-Based Image Recognition for Periodical Defrosting Control
Chen et al. developed a unique deep learning-based image recognition technique for on-demand defrosting control to reduce energy consumption in industrial energy systems.
To increase recognition accuracy, a unique convolutional neural network (CNN) model was presented to extract intricate and complex information for frosty state recognition.
Exhausting trials and significant labor consumption were considerably decreased by combining deep learning with image augmentation. The learning phase of the CNN model was significantly improved using the hyperparameter optimization method to increase recognition accuracy further.
The on-demand strategy and security regulations for the safety and dependability of multiple device control in commercial energy systems were merged in the suggested defrosting control approach. The purpose of the field experiment was to assess the effectiveness of the proposed method for defrosting and its potential for energy savings.
Chen et al. created a novel demand-based defrosting control mechanism for commercial energy systems by utilizing deep learning-based image recognition. Based on deep learning and image processing, the CNN model achieved high recognition accuracy of 95.92%.
The deep learning-based CNN model's recognition accuracy was around 5.50% higher than the traditional CNN model. The network framework and training parameters' hyperparameter optimization improved the frosty level's recognition accuracy, raising it to 97.57% by 1.65%.
A defrosting control approach was put forward for the on-demand defrosting control by the real-time frosty level identification based on the trained CNN model.
The defrosting frequency, total time, and energy consumption were reduced by 31.68%, 65.83%, and 42.92%, respectively, when the proposed demand-based defrosting control system was used.
The deep learning-based defrosting control approach ensured the safe operation of commercial energy systems by preventing frequent mode switching and the occurrence of an over-temperature process.
Deep Learning-Based Imaging Paves Way for Greener Commercial Energy Systems
A deep learning-based image processing method for defrosting control was shown to be an energy-saving and environmentally friendly way to apply demand-based defrosting control in commercial energy systems compared with previous image recognition methods and defrosting control methods.
A sizable amount of the world's energy consumption is currently used by refrigeration systems. Global adoption of the suggested control mechanism will significantly cut energy use and increase the transition of commercial energy systems to net-zero carbon emissions.
The suggested image recognition technique also holds great promise for use in other energy-saving scenarios, such as the defect inspection of photovoltaic panels and the design optimization of heat exchanges. This demonstrates excellent potential for artificial intelligence technology in energy conservation and emission reduction.
Chen, S., Chen, K., Zhu, X., Jin, X., & Du, Z. (2022). Deep learning-based image recognition method for on-demand defrosting control to save energy in commercial energy systems. Applied Energy, 324, 119702. https://www.sciencedirect.com/science/article/pii/S0306261922009965