挥发窑鼓风管的关键点识别及其摆放位置监测

Keypoints identification and position monitoring of the blower pipe for volatilization kiln

中文关键词  火焰视频  鼓风管位置  关键点检测  卷积神经网络  聚类分析

英文关键词  flame video  the position of blower pipe  keypoints detection  convolutional neural network  cluster analysis

中文摘要

      窑头鼓风管的摆放位置是影响氧化锌挥发窑燃烧状态的重要操作参数之一, 现仍由人工看火来手动调节, 同时现场没有为挥发窑的运行优化记录完善的鼓风状态数据, 也难以及时发现鼓风管被窑内渣块击中等安全隐患. 针对上述问题, 本文提出一种基于关键点识别的鼓风管位置监测方法. 首先, 文章对从窑头看火口采集的火焰视频 数据集设计一种邻域关键点辅助的数据扩充方法, 并构建级联金字塔网络(CPN)来预测鼓风管管口中心点的位置; 然后, 本文提出一种基于多帧图像的聚类分析算法来消除因烟尘遮挡所产生的异常点, 并采用一种量化指标来实现 对挥发窑鼓风管摆放位置的实时感知与记录; 最后, 本文基于现场采集的火焰视频数据进行了对比实验, 结果表明 所提出的关键点检测模型精度高、鲁棒性强, 且鼓风管位置的量化准确率高达92.3%.

英文摘要

      The position of blower pipe at the kiln head is one of the important operating parameters that affect the combustion state of the zinc oxide volatilization kiln. Currently, it is still manually adjusted through fire observation. At the same time, there is no blower state data recorded for the operation optimization of the volatilization kiln. It is also difficult to find the hidden safety hazards in time, such as the blower pipe being hit by the slag in the kiln. To tackle the above problems, a method for monitoring the position of blower pipe based on the keypoints identification is proposed. Firstly, for the flame video dataset collected from the kiln head, a data augmentation method assisted by neighborhood keypoints is designed, and the cascaded pyramid network (CPN) is constructed to predict the center position of blower pipe nozzle. Then, a cluster analysis algorithm based on the multi-frame images is proposed to eliminate the outliers caused by smoke and dust occlusion, and a quantitative index is used to realize the real-time perception and recording of blower pipe position for the volatilization kiln. Finally, some comparative experiments are carried out based on the flame video data. Experimental results show that the keypoints detection model has high accuracy and strong robustness, and the quantification accuracy proposed is as high as 92.3%.