Recognition of Sintering State in Rotary Kiln Using a
Robust Extreme Learning Machine
Hua Chen,Jing Zhang and Hongping Hu
School of Information Science and Engineering
Hunan University
ChangSha,China
anneychen@126.com
Xiaogang Zhang*
College of Electrical and Information Engineering
Hunan University
ChangSha,China
zhangxg@hnu.edu.cn
Abstract—Sintering is a key process for the industrial clinker
production. The sintering state estimation in clinker is an
essential factor for its process control. In this paper, a feature
extraction method from flame image and a robust extreme
learning machine (RB-ELM) classifier are provided to recognize
sintering process in rotary kiln. After a preprocessing of image
denoising and illumination compensation, material region of
flame image is segmented by region growing algorithm and a 5-D
statistic feature vector is extracted from it for the following
classifier. In order to reduce the influence of outliers in training
data caused by blurring image and to achieve a real-time
application on site, a robust extreme learning machine, which
adopted iterative weight least square (IWLS) method based on
M-estimator, is used for fast classification of sintering state.
Experimental results show that the proposed method can
recognize sintering state accurately, quickly and robustly.
Keywords—rotary kiln; vision detection; extreme learning
machine; robust estimation; flame image
I. INTRODUCTION
Rotary kilns are widely used in nonferrous metallurgy,
cement and other industry. The sintering state estimation of
clinker in rotary kiln is very essential in production control
process, which can directly influence the quality of products,
energy consumption and pollutant emission level.
Traditionally, the sintering state of materials in kiln can be
divided into three categories: 1)normal sintering,
2)oversintering, and3) undersintering. Normal sintering
materials are qualified products and others are unqualified.
The kiln typically consists of a refractory steel cylinder of
diameter 4-5m with a ratio of length to diameter greater than
20m. It is inclined along its length at an angle to the horizontal
of a few degrees and rotate about its axis slowly at a rotation
speed
ω
about 1 rev per 60-80 seconds [1]. The raw material
and coal powder are fed into the kiln from the feed port of the
cylinder. Sintering procedure is finished in kiln with the
revolution of cylinder, and the clinker flows continuously to
the discharge port through the inclination and rotation of the
cylinder. The kiln is filled with material generally less than
30% by volume. The motion of granular in the kiln flow
through the cylinder is shown in Fig.1 [1].
Fig. 1. Motion of granular in the kiln.
For a long time the sintering state of material is observed
by operator in a manual mode. Recently, combining image
analysis and machine learning algorithms to detect burning
condition and control coal-feeding of rotary kiln have
received more and more significant attentions. The camera
locates in the discharge port of kiln and the image captured by
it is shown in Fig.2. Lin [2] et al. calculated the 1-4 order
statistic HSI data of rotary kiln flame image to establish the
identification model for combustion condition using
multivariate regression method. By Gabor wavelet based
texture coarseness and Fuzzy C-Means cluster algorithm Sun
[3] et al. proposed an improved segmentation method for flame
image of rotary kiln burning region. He [4] et al. calculated the
Grey-Level Co-occurrence Matrix (GLCM) texture features of
clinker region and classified the sintered clinkers to different
qualities. Zhang [5] et al. realized a content-based retrieval
system for kiln sintering region flame image by the texture and
fire features of it. Li [6] et al. used a set of heterogeneous
features and fusion techniques to construct a flame-based
sintering state recognition system. The flame images captured
by the digital camera were always partitioned to several
regions such as coal region, bright region and material region
et al (Fig.2). The above mentioned methods mostly extract
various features such as HSI, texture feature and so on from the
segmented regions. Fig. 3. is the typical flame image sequence
captured from the rotary kiln of a large alumina plant. Because
of illumination and fog in kiln, the image is so blurring that it is
difficult to segment material zone, and extracting texture
feature from it is even more impossible. So the sintering state
detection methods mentioned above can not work well in
blurring flame image. On the other hand, the above methods
* Corresponding Author.
This work was supported by National Natural Science Foundation of
China ( No. 61203016, No.61174050, No. 61174140 ).
2014 International Joint Conference on Neural Networks (IJCNN)
July 6-11, 2014, Beijing, China
978-1-4799-1484-5/14/$31.00 ©2014 IEEE