Flame Image Processing and Classification Using a Pre-Trained VGG16 Model in Combustion Diagnosis

Area of Science:

• Combustion Engineering
• Artificial Intelligence in Energy
• Environmental Technology

Background:

• Coal combustion remains a vital energy source despite environmental concerns.
• Advanced combustion techniques like staged combustion and oxy-combustion aim to mitigate negative impacts.
• Real-time process monitoring is crucial for optimizing combustion efficiency, safety, and cost-effectiveness.

Purpose of the Study:

• To develop a method for accurate identification of undesired coal combustion states.
• To leverage flame image processing and deep learning for real-time combustion monitoring.
• To improve the safety and efficiency of coal combustion processes.

Main Methods:

• A novel method combining flame image processing with a deep convolutional neural network (DCNN) was developed.
• Adaptive selection of the gamma correction coefficient (G) was employed for flame segmentation.
• The VGG16 model, pre-trained for image classification, was utilized for identifying combustion states.

Main Results:

• The proposed method achieved high accuracy, ranging from 82% to 98%, in detecting specific combustion states.
• The technique demonstrated a fast processing time, suitable for real-time applications.
• An adaptive gamma correction coefficient based on image intensity improved flame segmentation.

Conclusions:

• The developed DCNN-based flame image analysis method offers a highly accurate and efficient solution for monitoring coal combustion.
• This approach enables real-time identification of suboptimal combustion conditions, facilitating immediate adjustments.
• The method holds significant potential for practical implementation in industrial coal combustion systems to enhance performance and reduce environmental impact.