Abstract—A coating collapse is a situation, where large parts of this coating break away from the refractory in big lumps. If the coating collapse is more pronounced, special attention has to be paid to the cooler, since an excessive amount of material may overfill the cooler and the clinker transport systems overload or lead to an excessive clinker temperature of the cooler outlet. Unstable coating easily leads to problems with the refractory material and causes loss of energy and disturbs stable kiln operation. A changing coating situation in the burning zone strongly influences the kiln torque. Usually the coating collapse is detected primarily through the kiln drive amps trend curve by operator. This paper adopted the application of Empirical Mode Decomposition (EMD) and Fuzzy Inference System (FIS) identify for Kiln Coating Collapse in Cement Process. This paper offers the analysis results that the proposed method is distinct better than the original function.

I. INTRODUCTION

A rotary Kiln is the heart of the cement plant that is a pyro-processing device used to raise materials to a high temperature in a continuous process. The rotary kiln is a cylindrical vessel, inclined abound 3.5 degree to the horizontal and rotated slowly on its axis. The general layout of a rotary kiln is depicted as Fig 1. The material is outlet from the Preheater and fed into the Kiln Inlet. As the kiln rotates, material gradually moves down towards the Kiln Hood undergo a certain amount of stirring and mixing. Hot air pass along the kiln usually in the opposite direction. The hot gases generated by a flame inside the kiln and recovery from the Cooler. The Hot gas outlet of the Kiln is passed to the Preheater for the heating exchange with process material.

Fig. 1. The general layout of a rotary kiln.

Ming-Chin Yang is with the Department of Electrical Engineering,

National Dong Hwa University, Hualien, Taiwan, R.O.C.. (e-mail: d9823005@gms.ndhu.edu.tw)

Jing-Zhong Wang is with the Department of Electrical Engineering, National Dong Hwa University, Hualien, Taiwan, R.O.C.. (e-mail: 810223007@gms.ndhu.edu.tw).

Tsung-Ying Sun is with the Department of Electrical Engineering, National Dong Hwa University, Hualien, Taiwan, R.O.C.. (phone: 886-3-8634061; fax: 886-3-8634060;e-mail: sunty@gms.ndhu.edu.tw).

The essential components of a rotary kiln are the shell and the refractory lining. The kiln shell is made from rolled between 30 and 80 mm thick High-temperature steel plate, welded to form a cylinder which usually around 55-70 m in length and up to 6 m in diameter. The essential function of the refractory lining is to insulate the steel shell from the high temperatures inside the kiln, and to protect it from the corrosive properties of the process material. The refractory selected depends upon the temperature inside the kiln and the chemical nature of the material being processed, the main section divide to Preheating zone, Caleining zone, Sintering zone and Cooling zone. In cement plant, the refractory life can be extended by maintaining a good condition coating of the processed material on the refractory surface. A coating collapse is a situation, where large parts of this coating break away from the refractory in big lumps, due to excessive weight, large temperature changes in the burning zone, fluctuations of the raw material properties, inadequate operation and others.

When coating falls out and the quantity is not too serious, not need to adjustment the control variables. The burning zone temperature maybe drop down slightly for a short period but normally quickly recovers, as the coating was already closed to sintering temperature. If the coating falls out quantify is too serious, the burning zone temperature will be drop down dramatically. A coating collapse results normally in raised free CaO values of the clinker, because it is very difficult to clinkerise the large pieces. If the coating collapse is more pronounced, the operator need to pay special attention to the cooler system, because the excessive amount of material will overfill the cooler and overload the clinker transport systems, finally lead to an excessive clinker temperature of the cooler outlet. The coat in the burning zone play an important role in cement industry and energy keeping, not only it protect the refractory bricks but also affect the type of clinkers produced [1]. If the coating collapse is too serious, the Firebrick Lining will damage and cause the big loss [2].

A coating collapse is detected primarily through the kiln drive amps. The kiln drive amps trend curve is depicted as Fig 2. A sharp increase in the average kiln drive amps indicates that suddenly a higher amount of material has to be moved in the kiln. Constant spiking of the kiln drive amps may also indicate an uneven loss of coating in one area of the kiln. The kiln camera is installed at the Kiln outlet to observe visually the fallen coating. The kiln drive amps and camera information help to estimate the order of the coating loss severity and how to deal with the magnitude of counteraction that has to be made.

Usually the coating collapse is observed primarily through the kiln drive amps trend curve by operator. This paper adopted EMD-based preprocessing and Fuzzy Inference

EMD-based Preprocessing and Fuzzy Inference System to Identify Kiln Coating Collapse for Predicting Refractory Failure in Cement

Process

Ming-Chin Yang, Jing-Zhong Wang and Tsung-Ying Sun*, IEEE Member

System (FIS) to identify Kiln Coating Collapse from analyzing the kiln drive trend curve for Predicting Refractory Failure in Cement Process.

The remainder of this paper is organized as follows: Section II describes the analysis method. Section III describes the data analysis results. Section IV contains a brief conclusion.

Fig. 2. The kiln drive amps trend curve.

II. ANALYSIS METHOD

In general, the trend analysis is used the Fast Fourier Transform (FFT). Through the FFT spectrum to observe the trend data of the kiln drive trend curve. Although FFT can supply the trend analysis function, it cannot look for the relationship between the frequency and time. Due to the inadequate mentioned above of the trend analysis, the study utilized Empirical Mode Decomposition (EMD) [3] to preprocess the kiln drive trend curve, and proposed a Fuzzy Inference System (FIS) to identify Kiln Coating Collapse.

A. Empirical Mode Decomposition (EMD) The Empirical Mode Decomposition (EMD) is designated for

the analysis of nonlinear and non-stationary time series. It was first introduced by Norden E. Huang in 1998. Unlike other signal processing or analysis methods in frequency spectrum, EMD is based on instantaneous frequency defined by Hilbert Transform (HT). Through decomposing the signal into several intrinsic-mode functions (IMFs), we can observe the feature of the information hidden in signal. The advantages of EMD are direct and adaptive. The EMD process flowchart is depicted as Fig. 3.

In order to comply with the strict rules of instantaneous frequency defined by HT, the local zero mean of signal should be symmetrical and have the same number of zero-crossing points and extreme points. EMD decompose the signal into several IMFs and a residue function )(tr :

1

( ) ( )N

ii

X t S t r t

(1)

The procedure, signal X(t) extracts IMF by using EMD, called sifting process, and the summary of sifting process is as follow: 1) Suppose that received signal from the kiln drive trend

curve is X(t). Setting the original data X(t) as the first transform function R0(t) and identifying all extremes minima and extremes maxima of R(t).

2) Find the upper envelope eu(t) and lower envelope el(t) using cubic spline interpolation of X(t) using cubic spline interpolation. Determine the mean envelope m(t) of eu(t) and el(t).

2

)()()(

tetetm lu (2)

3) To deduct m(t) from data R(t) to find IMF S(t). )()()( tmtRtS (3)

The S(t) should satisfy the conditions: (a) The number of local maxima and minima must be

equal or differ at most by one. (b) The mean values of envelope defined by local maxima

and minima is zero. If the S(t) is not satisfy the above conditions, setting the result as the new Ri+1(t), and then repeating the step 1) ~ 3).

4) When the R(t) is verified that being a monotonic function, defining R(t) as residue r(t), and then stop the sifting process. [4]

)()()( tStXtR (4)

The data of the kiln drive amps through the EMD decomposing the original signal into several IMFs, compare the IMFs and the original signal we can observe the feature of the information hidden in signal.

Fig. 3. The EMD process flowchart.

B. Fuzzy Inference System (FIS)

The Fuzzy Inference System (FIS) includes four inputs and one output Fig. 4 presents the Fuzzy Inference System. The membership functions of the four inputs are depicted as Fig. 5. The membership functions of the output illustrated as Fig. 6. Table I listed the rule-base for inferencing output. The scaling factors S1, S2, S3, S4 for the four inputs will be suitable adjusted based on expert knowledge about the process during the simulation.

Fig. 4. The Fuzzy Inference System.

The computing result of the FIS will be used to identify the collapse of Kiln Coating.

Fig. 5. The membership function of the four inputs.

Fig. 6 The membership functions of the output.

III. DATA ANALYSIS RESULTS

This section describes the several data analysis through the kiln process history. The data analysis settings and results are offered.

A. The Data Analysis Settings

The sampling period of the kiln drive amps is 6 sec, the data quantity is 14400 per day from the first day 08:00:00 to next day 07:59:54. The rule-base for computing output is listed in Table I. The initial data of scaling factors the S1, S2, S3, S4 for the four input are 1.0.

B. Data Analysis Results

The first data analysis is using the DataLog trend curve of the rotary kiln process on July.9.2015. The data including the kiln drive amps, burning zone temperature, kiln outlet clinker temperature and kiln hood pressure on is shown as Fig. 7. The coating collapse is occur at 20:29:18. When the kiln coating collapse occur the burning zone temperature beginning drop down. The sampling data of the kiln drive amps is shown as Fig. 8.

The first step is based on EMD to decompose the kiln drive amps signal into several IMFs. The obtained IMFs of EMD process is shown as Fig. 9. The second step is through the FIS to identify the Kiln Coating Collapse. The obtained result of the FIS process is shown as Fig. 10, and it could easily identify the Kiln Coating Collapse.

Fig. 7 The DataLog trend curve of the rotary kiln process on July. 9. 2015.

Fig. 8 The sampling data of the kiln drive amps for Fig. 7.

Fig. 9 The obtained IMFs of EMD-based preprocessing for Fig. 8.

Fig. 10 The obtained result of the FIS identification for Fig. 9.

The second data set collected by using the DataLog trend curve of the rotary kiln process on July. 6. 2017. The data including the kiln drive amps, burning zone temperature, kiln outlet clinker temperature and kiln hood pressure on is shown as Fig. 11. The main coating collapses are occur at 09:56:42, 15:19:36 and 06:05:48 on next day. In addition have some slight coating collapses in other time. When the kiln coating collapse occur the burning zone temperature beginning drop down. The sampling data of the kiln drive amps is shown as Fig. 12.

Fig. 11 The DataLog trend curve of the rotary kiln process on July. 6. 2017.

The first step is to decompose the kiln drive amps signal into several IMFs via EMD. The obtained IMFs of EMD process is shown as Fig. 13.

Fig. 12 The sampling data of the kiln drive amps for Fig. 11.

Fig. 13 The obtained IMFs of EMD-based preprocessing for Fig. 12.

The FIS is to identify the Kiln Coating Collapse. The obtained result of the FIS process is shown as Fig. 14. The result of the FIS not only identify the main Coating Collapse, but also identify the slight Coating Collapse.

Fig. 14 The obtained result of the FIS identification for Fig. 13.

The third data set collected by using the DataLog trend curve of the rotary kiln process on July. 22. 2017. The data including the kiln drive amps, burning zone temperature, kiln outlet clinker temperature and kiln hood pressure on is shown as Fig. 15. The main coating collapses is occur at 13:11:30. In addition have some slight coating collapses in other time. When the kiln coating collapse occur the burning zone temperature beginning drop down. The sampling data of the kiln drive amps is shown as Fig. 16.

Through the EMD, the kiln drive amps signal of Fig. 16 decomposed into several IMFs shown as Fig. 17. The obtained result of the FIS process is shown as Fig. 18, it could easily identify the Kiln Coating Collapse.

Fig. 15 The DataLog trend curve of the rotary kiln process on July. 22. 2017.

Fig. 16 The sampling data of the kiln drive amps for Fig. 15.

Fig. 17 The obtained IMFs of EMD-based preprocessing for Fig. 16.

Fig. 18 The obtained result of the FIS identification for Fig. 17.

The fourth data analysis is using the DataLog trend curve of the rotary kiln process on July. 24. 2017. The data including the kiln drive amps, burning zone temperature, kiln outlet clinker temperature and kiln hood pressure on is shown as Fig. 19. The main coating collapses occurred at 05:39:30 on next day. In addition have some slight coating collapses in other time. When the kiln coating collapse occurred the burning zone temperature beginning drop down. The sampling data of the kiln drive amps is shown as Fig. 20.

Through the EMD decomposing the kiln drive amps signal into several IMFs and the obtained IMFs is shown as Fig. 21. The obtained result of the FIS process is shown as Fig. 22, and it is easy to identify the Kiln Coating Collapse.

Fig. 19 The DataLog trend curve of the rotary kiln process on July. 24. 2017.

Fig. 20 The sampling data of the kiln drive amps for Fig.19.

Fig. 21 The obtained IMFs of EMD-based preprocessing for Fig. 20.

Fig. 22 The obtained result of the FIS identification for Fig. 21.

The fifth data analysis is using the DataLog trend curve of the rotary kiln process on August. 5. 2017. The data including the kiln drive amps, burning zone temperature, kiln outlet clinker temperature and kiln hood pressure on is shown as Fig. 23. The main coating collapses is occur at 05:39:30 on next day. In addition have some slight coating collapses in other time. When the kiln coating collapse occur the burning zone temperature beginning drop down. The sampling data of the kiln drive amps is shown as Fig. 24.

Through the EMD decomposing the kiln drive amps signal into several IMFs and the obtained IMFs is shown as Fig. 25. The obtained result of the FIS process is shown as Fig. 26. Following the result, it is easy to identify the Kiln Coating Collapse.

Fig. 23 The DataLog trend curve of the rotary kiln process on Aug. 5. 2017.

Fig. 24 The sampling data of the kiln drive amps for Fig.23.

Fig. 25 The obtained IMFs of EMD-based preprocessing for Fig. 24.

Fig. 26 The obtained result of the FIS identification for Fig. 25.

Through the EMD-based preprocessing and FIS identifier, we can easily identify the Kiln Coating Collapse. This method provides a pre-warning function for the process operator.

IV. CONCLUSIONS

This paper proposed a EMD-based preprocessing with the FIS-based approach to identify the Kiln Coating Collapse. This method provides a pre-warning function for the process operator. This analysis method not only can be applied in identify the Kiln Coating Collapse but also useful for other practical process.

REFERENCES [1] Abdulkadhum J K Al-Yasiri, Montadher A. Muhammed, “Estimating

the thickness of coating in the burning zone of cement kilns including the aging factor,” The Iraqi Journal For Mechanical And Material Engineering, Vol.12, No.3, 2012

[2] Ono, M., Kozuka, H., “Damage of refractory bricks lined in cement rotary kiln”, Tehran International Conference on Refractories, 515-524, 2004.

[3] Ming-Chin Yang, Tsung-Ying Sun, “Preliminary Study on HHT-based Refractory Failure Prediction for Kiln Shell”, in Proc. 2016 IEEE International Conference on Industrial Technology (ICIT), pp. 968-972, 14-17 Mar., 2016, Taipei, Taiwan.

[4] N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H. Shih, Q. Zheng,N. Yen, C. C. Tung, and H. H. Liu, “The empirical mode decomposition and the Hilbert spectrum for nonlinear and nonstationary time series analysis, ” Proc. R. Soc. Lond., 454 : 903-995, 1998.

TABLE I. The rule-base for computing output.