【作者】 华彦平；

【导师】 吕震中；

【作者基本信息】 东南大学 ， 热能工程， 2005， 博士

【摘要】 数字图像型火检是目前锅炉燃烧诊断的主要发展方向。其关键是如何对火焰图像进行处理,提取出火焰图像的特征参量。当前图像型火检主要研究方向是寻找新的理论和方法来对火焰图像进行处理,研究如何提取火焰图像的特征参量来表征燃烧特征,研究如何根据燃烧诊断的需要选取、融合火焰图像的特征参量,从而进一步提高诊断系统的准确性、通用性。数学形态学是目前图像处理领域应用比较广泛的一种理论,彩色数学形态学是当前重点研究方向。本论文从图像处理的角度,首次将数学形态学理论引入锅炉燃烧诊断领域。论文具体内容如下:1、阐述数学形态学的基本理论——二值数学形态学和灰度数学形态学。分别从二值数学形态学和灰度数学形态学两个方面,对如何选取合适的结构元素和形态学算子来提取火焰图像的边缘和骨架进行了详尽研究,并得出火焰图像的二值和灰度形态特征参量;2、引入矢量排序思想,建立起新的彩色数学形态学理论——矢量彩色形态学。矢量彩色形态学在对图像进行形态变换时,不会产生颜色的失真。针对火焰图像的彩色边缘和骨架的提取,研究了如何选取结构元素和构造形态学算子,并得出表征火焰图像的彩色形态特征参量;3、介绍了自适应模糊神经网络——ANFIS模糊神经网络的基本思想和原理。研究了火焰图像的灰度形态特征参量和彩色形态特征参量的融合问题,用以指导神经网络输入的构造;4、分别基于火焰图像灰度形态特征参量和彩色形态特征参量,应用模糊神经网络进行了锅炉燃烧诊断研究。研究结果表明,基于灰度形态特征参量的模式识别虽能准确识别熄火工况(ON/OFF),但对稳定和不稳定燃烧的识别不理想;基于彩色形态特征参量的模式识别能够准确识别稳定燃烧、不稳定燃烧和熄火三种模式;5、论文研究结果表明,基于数学形态学的燃烧诊断系统无需事先定义图像处理区域,很好地解决了火焰的漂移,卷吸和偷看等难题;6、对Kolmogorov复杂性测度的计算及实际遇到的问题作了研究分析,指出了复杂性测度与计算频率及信号幅值大小的关系。提出了改进的Kolmogorov复杂性测度算法,并基于改进的Kolmogorov复杂性测度算法对不同工况下的炉膛负压和火焰图像灰度特征进行了复杂性测度的计算,计算结果表明复杂性测度能够在一定程度上对锅炉燃烧进行诊断和预警;7、介绍了实际开发的基于数学形态学和模糊神经网络的锅炉燃烧诊断系统的基本功能和流程。更多还原

【Abstract】 The main direction of boiler combustion diagnosis is by digital image detection and processing. The key point of this method is how to process the digital flame images, how to extract their characteristic parameters. Now the most important thing of boiler combustion diagnosis is to find and use new theories and methods to process the flame image, to study how to extract the characteristic parameters which can present the combustion characters, and to study how to select and merge the plenty of characteristic parameters for the different combustion diagnosis. And all these are try to develop the diagnosis system to be more accurate and more common in use.Mathematical morphology (MM) is a geometric approach to image processing that was developed as a powerful tool for shape analysis in binary and grayscale images, and more and more specialists are now working for the applications of Mathematical Morphology to process color image. It is the first time to apply Mathematical Morphology to do diagnosis of boiler combustion. The main contents of this paper are as follows:1 The basic Mathematical Morphology theories-binary and grayscale Mathematical Morphology are presented firstly. In order to extract the edge and skeleton both of binary and grayscale flame image, it is studied how to build the Structuring Elements and MM operators. Sequentially, characteristic parameters both of binary and grayscale flame images are studied.2 A new Color Mathematical Morphology-Vector Color Morphology based on vector ordering is set up. When Vector Color Morphology is applied to images, there is no loss or corruption of color information of the images. Then color Structuring Elements and MM operators are described, and color edge and color skeleton extractions are studied. Finally, characteristic parameters of color flame images are discussed. 3 The theory and procedure of adaptive fuzzy neural network-ANFIS FNN is discussed. In order to apply ANFIS to flame image processing, the re-build of their characteristic parameters are studied here. 4 The diagnosis of boiler combustion based on ANFIS FNN and the characteristic parameters both of grayscale and color flame images is studied. The results show that the diagnosis based on the characteristic parameters of grayscale flame images can recognize the ON/OFF flame image, but can’t recognize the normal and abnormal flame images very correctly. On the other hand, the diagnosis based on the characteristic parameters of color flame images can recognize all the 3 flame patterns successfully: normal flame, abnormal flame and OFF flame.5 There is no need to fix the processing area to flame image by using MM, no matter if there is any shifting, curling, or glomming to the flame image.6 A detail discussion is presented about the calculation of Kolmogorov complexity index. The discussion points out the Kolmogorov complexity index is responsible to the signal frequency and amplitude. Kolmogorov complexity index to the negative pressure in boiler and the gray value of flame image shows it can forecast the pattern of combustion.7 The detail functions and flows of a actually combustion diagnosis system, which is developed for Power Station projects based on MM and ANFIS FNN, are introduced.更多还原