【作者】 孙鹏；

【导师】 柴天佑；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 我国铝土矿资源丰富,但占储量80%的铝土矿的铝硅比较低,普遍采用烧结法生产氧化铝。回转窑是烧结法生产氧化铝的核心设备,其主要功能是将生料浆烧结成为容重合格的熟料。回转窑烧结工况综合反映了窑内火焰燃烧状态与物料烧结状况,与回转窑过程熟料产品的质量、产量、能耗以及设备安全等因素密切相关,它的自动识别对于实现回转窑过程的自动控制具有非常重要的意义。回转窑长达百米,由于其结构的特殊性以及烧结法工艺的复杂性导致了氧化铝回转窑过程具有质量指标熟料容重难以在线测量,关键工艺参数烧成带温度检测干扰严重,多变量强耦合、强非线性、大惯性以及不确定性干扰等综合复杂特性。受到回转窑窑体倾斜、旋转以及窑内各类复杂的高温固、液、气三相物理化学反应产生的火焰闪烁、物料运动以及窑内各区域间的对流换热和辐射换热等因素的影响,回转窑过程还存在着窑内烧成带火焰区与物料区难于分辨,过程数据检测可靠性差等问题,难以采用常规仪表与监测技术实现对回转窑烧结工况的连续在线准确监测。工业现场长期依赖“人工看火”方式肉眼观测窑内烧成带火焰燃烧状态与物料烧结状况结合过程数据识别烧结工况后进行回转窑过程的控制。容易造成熟料的欠烧或者过烧,导致熟料容重合格率低、设备运转率差、窑内衬使用寿命短、产量低、能耗高、环境污染严重等问题。本文依托国家863高技术计划重点项目“中国铝业公司综合自动化系统总体设计方案及关键技术攻关”的子课题“大型回转窑过程优化控制技术”,以实现回转窑烧结工况的自动识别为主要目标,进行了基于图像处理的氧化铝回转窑烧结工况识别系统的研究与开发,取得了如下成果：1.将图像处理技术与数据融合技术相结合,在分析人工识别回转窑烧结工况的基础上,提出了由烧结工况图像预处理、图像分割、图像特征提取、数据融合与烧结工况识别模型组成的基于烧结工况图像特征与数据融合的烧结工况自动识别方法。具体研究内容归纳如下：针对氧化铝回转窑烧结工况图像频域噪声干扰的特点以及彩色图像处理算法复杂、实时性差的问题,提出了利用频域滤波技术与灰度变换技术相结合对烧结工况图像进行预处理的算法,实现了烧结工况图像的去噪与灰度变换。针对单纯的基于像素灰度值的图像分割方法难以精确分割火焰区与物料区的难题,分析了氧化铝回转窑烧结工况图像火焰区与物料区在纹理特征方面的差别,提出了利用Gabor小波纹理粗糙度对基于像素灰度值的FCM聚类结果进行去模糊化的烧结工况图像分割算法,实现了图像中火焰区与物料区的分割。根据“人工看火”经验描述了物料高度、闪烁频率、整体平均灰度、火焰颜色与物料颜色五个氧化铝回转窑烧结工况图像特征,提出了从整体图像及分割后的图像中提取上述特征的算法。根据“人工看火”过程的数据融合原理,针对烧结工况图像特征以及由烧成带温度、窑头温度、窑尾温度以及冷却机电流构成的关键过程数据的特点,提出了包括数据滤波、同步序列化与归一化处理的融合算法,得到了融合后的混合特征数据。将混合特征数据作为输入,欠烧结、正烧结和过烧结三种基本烧结工况作为输出,建立了基于准正态二叉树支持向量机的烧结工况识别模型。提出了基于上述模型的烧结工况自动识别算法,并利用专家修正样本进行模型的反馈增量学习,提高了模型适应生产边界条件波动的能力。2.研究并开发了基于上述烧结工况识别算法的工况识别软件,该软件具有烧结工况视频监视功能、烧结工况识别功能、过程数据通讯功能和人机交互功能；研制了由前端图像采集设备、网络视频传输设备、图像采集卡、工业控制计算机、显示与存储设备构成的远程分布式系统硬件平台和计算机操作系统、应用程序接口(API)、组件对象模型(COM)、动态链接库(DLL)、软件二次开发包(SDK)、工况识别软件组成的基于图像处理的氧化铝回转窑烧结工况识别系统。3.将本文提出的基于图像特征与数据融合的烧结工况识别方法与基于烧成带温度测量值模式识别的烧结工况识别方法进行比较实验,实验结果表明,该方法克服了由于单一类型的监测数据不可靠而导致的烧结工况识别率低、识别性能不稳定的问题,实现了烧结工况的准确识别。将本文研制的烧结工况识别系统于山西铝厂3#回转窑熟料烧结过程进行了实验研究,实验结果表明,系统实现了上述设计功能,运行过程稳定可靠,能够实时的识别烧结工况,识别率达到93.5%。更多还原

【Abstract】 China has plenty of alumyte, but 80% of alumyte reserves is of low alumina silica ratio. Rotary kiln is the core equipment in alumina sintering process and its main function is to sinter raw material slurry to produce qualified alumina clinker. Sintering status of rotary kiln synthetically reflects the burning state and material sintering status, and tightly correlates with alumina clinker quality, production, energy consumption as well as equipments status etc. Its automated recognition is important for automation of rotary kiln process.Rotary kiln is about one hundred long, because of its special architecture and complexity of sintering process, the liter weight of clinker in the alumina sintering process is hard to directly measured online, and has many integrated complexity such as serious disturblance in temperature detection of burning zone, multi-variables strong coupling, strong nonlinearity, big inertia as well as uncertain disturbance etc.Sintering status of rotary can not be exactly measured continuously on-line. Because of flame flickering, material moving and convection as well as radiation exchange of heat which produced by inclined body and rolling of rotary kiln, as well as complex solid state, liquid state and gaseous physical chemical reactions with high temperature, flame area and material area in burning zone of rotary kiln is difficult to discriminate and disturbance in process data detection is serious, sintering status can not be exactly recognized by routine instrument. Alumina rotary kiln process control has depended on "man-watch" operation for a long time, which observes flame combustion state and material sintering status by eyes, and combines process data to recognize sintering status. As a result, over-burning or under-burning usually happens, the qualification index of liter weight of clinker is low, kiln liner is easy to wear out, the kiln running rate and yield is low; the energy consumption and labor intensity remains high; and environmental pollution is serious.This work is supported by the program of "Optimizing control techniques for large rotary kiln processes", which is a sub-project of the National Hi-tech 863/CIMS Program named "Overall project design and key technology development of the integrated automation system of China Aluminium Corporation". In order to realize the automated sintering status recognition of rotary kiln, this dissertation has made detail work on the research and development of burning zone status recognition system of alumina rotary kiln, The detail works are summarized as follows:1. Bases on "man-watch" recognition experiences, image processing technique has been combined with data fusion technique and a sintering status recognition method ground on features of sintering status image and process data fusion has been proposed which include image pre-processing, image segmentation, image features extraction, data fusion and sintering status recognition model.Consideing frequency characteristic of noise of sintering status image of alumina rotary kiln and the complexity of color image processing algorithm which is a time-cosuming algorithm, an integrated pre-processing algorithm is proposed for sintering status image with frequency filter technique and gray transform technique. Frequency noise can be removed from sintering status image by this algorithm. As a result, a filterd gray sintering status image is abtained.Because of difficulty of segmentation between flame area and material area of sintering status image based on different gray levels of the pixels, the texture difference between flame area and material area of sintering status image of alumina rotary kiln has been synthesized and a segmentation algorithm which uses texture coarseness that is depicted by Gabor wavelets is proposed to improve the fuzzy clustering result of gray levels, as a result a exact segmentation of sintering status image is abtained.Depending on "man-watch" experiences, five features of sintering status image of alumina rotary kiln is depicted, such as height of material, flickering frequency, average gray of whole image, color of flame as well as color of material, and the feature exaction algorithms for the above features from the whole image as well as segmented image areas are proposed.Refering to "man-watch" recognition process, a data fusion algorithm which is made up of data filtering, synchronization and standardization is proposed with the characteristics of features of sintering status image and key process data such as temperature of burning zone, temperature of head of rotary kiln, temperature of end of rotary kiln to as well as current of cooler,as a result, a hybrid data features is abtained.A recognition model of sintering status based on quasi-symmetrical binary tree svm(support vector machine)constructed with hybrid data features as input and over-sintering, just-sintering and under sintering status as output. An automated algorithm for sintering status recognition is proposed based on the above model. In order to adapt to the change of production boundary conditions, professional modified samples is used to realize feedback increment learning for recognition model. 2. A status recognition software based on the above algorithms has been and developed, which realizes four functions, such as video surveillance of sintering status, recognition of sintering status, process data communication and human-machine interaction. a remote distributed system hardware platform has been constructed which is made up of fore-end image capture equipment, network video transformation equipment, image capture card, industrial control computer, display and storage. The above hardware, operation system of computer, API (Application Programming Interface), COM (Components Object Model), DLL (Dynamic Link Library), SDK (Software Development Kit) and recognition software of sintering status constitutes the recognition system of sintering status of alumina rotary kiln.3. The experiment which compares recognition method proposed by this dissertation with the status recognition method based on pattern recognition of temperature of burning zone shows that this method which based on features of sintering status of alumina rotary kiln and process data fusion can recognize the sintering status correctly. Carried out system experiment research on 3# rotary kiln in Shanxi alumina factory, the result shows that the recognition system can realize the designed function, run reliable and steady and recognize sintering status in real-time, accurate recognition rate is up to 93.5%.更多还原