In Situ Visual Monitoring of Rotary Air Preheater Blockage: Setup and Image Analysis

Abstract

Blockage of air preheaters has become a common problem in coal-fired power plants due to the formation of sticky ammonium bisulfate. The lack of efficient monitoring technique hinders the efforts to prevent the problem. In this paper, we propose a novel in-situ visual monitoring system on the cold end of the air preheater. The system is inexpensive but powerful enough to reveal the temporal evolution and spatial distribution of blockage fractions. By virtual of 3D printing, we built a lab-scale test structure of the contaminated corrugate plate of the air preheater, and took images under simulating dark environments. We revealed that integrating Gaussian filtration for noise elimination and then K-means clustering for image segmentation exhibits the best performance for images from nearly true circumstances. Moreover, the convolutional neural network manifests its ability to learn the blockage fraction and thus its future applicability for image processing with a well-labeled image dataset.