Intelligent Dynamic Modeling for Online Estimation of.pdf

  • Upload
    qssr123

  • View
    214

  • Download
    0

Embed Size (px)

Citation preview

  • 8/17/2019 Intelligent Dynamic Modeling for Online Estimation of.pdf

    1/5

    Intelligent Dynamic Modeling for Online Estimation of

    Burning Zone Temperature in Cement Rotary Kiln* Ping Zhou and Meng Yuan

    State Key Laboratory of Synthetical Automation for Process Industries

     Northeastern University

    Shenyang, Liaoning Province, China

    [email protected]

    * This work is partially supported by NSF Grant #61104084, #61290323, #61333007, IAPI Fundamental Research Funds Grant #2013ZCX02-09, and theFundamental Research Funds for the Central Universities Grant #N130508002 to P. Zhou and M. Yuan. 

     Abstract   - Cement rotary kiln is a complex multivariable,large-disturbances and nonlinear system which is full of mass

    transfer, heat transfer, and physical and chemical reactions. The

    burning zone temperature (BZT) in cement rotary kiln is a very

    important production index and has a significant role on the

    quality of the clinker. However, the BZT is generally difficult to

    be measured online using the conventional instruments. Although

    the BZT can be detected by using the expensive infraredpyrometer which located at the kiln head hood, it generally loses

    veracity due to the complex dynamics of the cement rotary kiln.

    Obviously, such an inaccurate measurement may guide the

    operator to do some improper operations in practice. To attack

    such a practical engineering problem, an intelligence-based

    dynamic soft-sensor modeling approach is proposed to online

    estimate the BZT in cement rotary kiln in this paper. The

    proposed approach mainly includes two digital filters which are

    used to pre-process the original measurement data, and an

    intelligent CBR soft-sensor system which is adapted to online

    predict the BZT in time, according to the measured secondary

    variables. At last, industrial tests have been performed to

    demonstrate the good estimation performance of the proposed

    method for a real cement rotary kiln process.

     Index Terms - Burning zone temperature, Cement rotary kiln,

     Dynamic soft-sensor, Case-based reasoning, Digital filter.

    I.  I NTRODUCTION 

    Rotary kiln is a kind of large scale sintering device widely

    used in various process industries, such as metallurgical,

    cement, refractory materials, chemical and environment

     protection [1-3]. In the cement production industry, the

    cement rotary kiln decomposition is the most important unit,

    and its operating status serious affects the output, quality,

    energy consumption, and environment pollution. The

    automation problem of such complicated processes remains

    unsolved because of the following inherent complexities. It isa multivariable nonlinear system with strong coupling. The

    complicated working mechanism includes physical change

    and chemical reaction of material, procedure of combustion,

    thermal transmission among gaseous fluid, solid material fluid

    and the liner. Moreover, the key controlled variable of burning

    zone temperature is difficult to be measured. In fact, most of

    rotary kilns are still under manual control with human

    operator observing the burning status; this is especially true in

    China. As a result, the product quality is hared to be kept

    consistent and energy consumption remains high. Although

    several advanced control strategies including fuzzy control,

    artificial neural network based control and predictive control

    have been introduced into process control of rotary kiln, all

    these researches focused on trying to achieve complete

    automatic control without human operators [1, 3-6]. As amatter of fact, the boundary conditions of a cement rotary kiln

    change heavily. For example, the material load, water content

    and components of the raw material slurry vary frequently and

    severely. Moreover, the offline analysis data of components of

    raw material slurry reach the operator with large time delay.

    Thus complete automatic control without human operation for

    such a complex process is unpractical.

    For the cement rotary kiln process, the most difficult

    control problem is that the key technical index, burning zone

    temperature (BZT), is difficult to be measured online using

    the conventional instruments. Even if some factory use

    expensive infrared pyrometer located at kiln head hood to

    measure the burning zone temperature directly. However, due

    to the complex dynamics of the cement rotary kiln, such a

    measurement generally loses veracity, which will misadvise

    the operator to do some improper operations to the running

     process. To attack such a practical problem, this paper

    develops an intelligent soft-sensor modeling approach for

     burning zone temperature using case-based reasoning (CBR)

    [7-10] estimation technique. Industrial test and results have

    show the effectiveness and validity of the proposed method.

    II. PROCESS DESCRIPTION OF CEMENT R OTARY K ILN

    Rotary kiln is one of the key equipments in a cement

    industry used to convert calcareous raw meal to cement

    clinkers. The kiln, as shown in Fig. 1, is a long and complextunnel, with a cylindrical shape. The cement rotary kiln

     process under study can be described as follows:

    Raw material slurry is sprayed into the rotary kiln from

    upper end (the kiln head), the coal powders from the coal

    injector and the primary air from the air blower are mixed into

     bi-phase fuel flow, which is sprayed into the kiln head hood

    and combusts with the secondary air, which comes from the

    cooler. The heated gas was brought to the kiln tail by the

    induced draft fan, while the material moves to the kiln head

    978-1-4799-5825-2/14/$31.00 ©2014 IEEE

    Proceeding of the 11th World Congress on Intelligent Control and AutomationShenyang, China, June 29 - July 4 2014

    6167

  • 8/17/2019 Intelligent Dynamic Modeling for Online Estimation of.pdf

    2/5

    Coal

    Powder 

    Air 

    Clinker 

    Cooling

    zone

    Bunring

    zone

    Reaction

    exothermic

    zone

    Decomposite

    zone

    Drying

    zone

    Pre-

    heating

    zone

    Raw

    material

    Head-

    end

    of kiln

    Back -end

    of the

    kiln

    Cement Rotary Kiln

     Fig. 1 Schematic diagram of cement rotary kiln

    ?

     Input

    VariableCement Rotary Kiln Process

    Data Pre-

     processing

    CBR 

    Soft-sensor 

    Model

    Learning Algorithm

    Burning zone temperature

    (Sampled value)

    Estimated value

    Fig.2 The proposed intelligent soft-sensor modeling strategy for burning zone

    temperature

    1 z 

    1 z 

    1 z 

    1 z 

    1( ) x t 

    2( ) x t 

    3( ) x t 

    3( 1) x t  

    2( 1) x t  

    1( 1) x t  

    4 ( 1) x t  

    ( ) y t 

    4( ) x t 

    CBR System for

    Soft-Sensor of 

    BurningZone

    Temperature

    1 z ( 1) y t  

     Fig.3 Dynamic ANN soft-sensor model

    via the rotation of the kiln and its self weight, in counter

    direction with the gas. Raw material is carried along the kiln a

    very low speed. Near the middle of the kin is the firing zone,

    where gas burners are placed to impose a given temperature

     profile. In a kiln, the back-end is responsible for the

    calcification of meal before the main baking, so if the

    temperature of back-end is more than the acceptable range, the

     baking will be done before entering the burning zone, and vice

    versa is to happen for lower the temperature to be. At the

     burning zone, the high temperature melts the classificated

    meal. Then the main chemical reactions between silicates andoxygen of the air occur. A part of the combustion gases is the

    Co gas produced here. Finally, the cement crystals are made

    and go out from the kiln as the clinker [1-3]..

    The burning zone temperature (BZT  zt  B ) in cement

    rotary kiln is a very important production index and has a

    significant role on the quality of the clinker. According to the

     physical phenomena taking place in the rotary kiln, the main

     process variables (measurable and adjustable) that affect the

    BZT are coal (fuel) feed rate  F C  (t/h), exhaust air feed rate

     F  A (m/s), raw marital feed rate  F  R (t/h), and kiln rotation

    speedS  K  (r/m).

     

    Increasing  F C   will enhance the reaction in the rotary

    kiln, causing the lame to burn well and the BZT to rise.

      Increasing  F  R  will increase the reactant in the rotary kiln

    and cause its temperature to rise; however, when  F  R  

    increases to a certain extent, it will cause the BZT to

    decrease.

      Increasing  F  A  will speed up the reaction in the kiln and

    increase the exhaust emission, causing the kiln

    temperature to rise. However, when  F  A   increases to a

    certain extent, there will be insufficient air for

    combustion, and incomplete combustion will generate

    CO, causing the sintering temperature to decrease.

      S  K   can be adjusted by the volume of the input materials

    to rotary kiln, feed rate of raw mix. In the case of more

    input materials, S  K    should be adjusted to complete the

     burning process. On the other hand, S  K   can itself

    interfere in the kiln temperature.

    III. DYNAMIC SOFT-SENSOR MODELING FOR BZT

    The key problem of closed-loop control of the BZT is that

    it is cannot be measured online with conventional methods.

    The most effective method of overcoming it is that employ

    6168

  • 8/17/2019 Intelligent Dynamic Modeling for Online Estimation of.pdf

    3/5

    soft-sensor technique to online estimate the BZT. Due to case-

     based reasoning (CBR) has good ability to identify and

    control complex nonlinear systems [6-10], the CBR-based

    soft-sensor dynamic modeling approach is therefore employed

    to develop a BZT soft-sensor in this paper.

    The CBR utilizes the specific case information availableas historical precedence for proposing solutions to current

     problem. The most important aspects of the existing cases are

    first stored and indexed. New problem situations are then

     presented and similar, existing cases are identified from the

    knowledge base. Finally, the previous problem solutions are

    adapted and the revised solutions are proposed for the current

    situation [7-10].

    The proposed CBR based soft-sensor modeling approach

    for dynamical estimating the BZT is shown in Fig.2. It mainly

    consists of a process data pre-processing module and a

    dynamic CBR soft-sensor system.

     A. Data Pre-ProcessingIf the original secondary variables O { , , F F C A  

    , } F S  R K  are used for soft-sensor modelling and calculating

    directly, it will cause some adverse influences on estimation

     precision. Therefore, digital filtering technique is employed to

     pre-process these original data.

    a) Noise peak filtering algorithm [11]:  It is used to

    eliminate the noise peak jump.

    O E

    E O

    E O

    E E

    O E

    E E

    If ( ) ( 1)

    Then ( ) ( ),

    If ( 1) ( )

    Then ( ) ( 1)

    If ( ) ( 1)

    Then ( ) ( 1)

    t t 

    t t 

    t t 

    t t 

    t t 

    t t 

     

    where t   denotes sampling time, E ( )t   denotes the pre-

     proceed data by the noise peak filter,   is the maximal

    allowed variety value of O ( )t   at successive sampling time.

    b) Average moving filtering algorithm: It is used to

    eliminate the lower and high frequency noise fluctuation.

    E EF F

    ( ) ( )( ) ( 1)

    t t N t t 

     N 

     

    where  N   is the length of average moving filtering.

     B. CBR-Based Dynamic Soft-sensor Algorithm

    Fig.3 illustrates the CBR-based BZT soft-sensor model,

    where 1 2 3 4, , , x x x x  present the secondary variables of

    , , , F F F S C A R K  ,  y  stands for the main variables, i.e. the

     burning zone temperature (  zt  B ). This means that the CBR

    soft-sensor model consists of 9 inputs and 1 outputs, therefore

    the dynamic input-output relation of the soft-sensor can be

    represented as follows.

    1 2 3 4 1

    2 3 4

    ( ) [ ( ), ( ), ( ), ( ), ( 1),

      ( 1), ( 1), ( 1), ( 1)]

     X t x t x t x t x t x t 

     x t x t x t y t 

     

    It is noted that to capture the system dynamics, the time

    series and time delays of the input and output variables have

     been taken into account in the proposed dynamic CBR model.

    The reasoning flow of CBR-based soft-sensor mainly

    includes case representation, case retrieval and case matching,

    case reuse, and case revision.

    a) Case representation. As shown in Table I, the case

    representation consists two parts, case descriptors and

    solutions of cases. The case descriptors include the coal (fuel)

    feed rate ( ) F C t  , the exhaust air feed rate ( ) F  A t  , the raw

    marital feed rate ( ) F  R t  , the kiln rotation speed ( )S  K t  , .the

     past value of ( 1), ( 1), ( 1) F F F C t A t R t     , ( 1)S  K t   , and

    ( 1) zt  B t   , which defined as 1 2 9, , , f f f  , respectively. The

    case solution is ( ) zt  B t   which is needed to be estimated.

    b) Case retrieval and case matching. Let the description

    characteristics of the current cement rotary kiln process isT T T T

    1 2 9( , , , ) F f f f    . Define the case similarity between the

    current rotary kiln system and the k th (1 )k N  case of the

    case base :{ }k k k C F J   as SIMk  , which is given by 

    9 9T

    ,

    1 1

    T

    ,

    ,

    ,

    SIM ( , ) sim ( , )

    sim ( , ) 1max( , )

    k k i i i i k i

    i i

    i i k T 

    i i i k   T 

    i i k 

     F F f f 

     f f  f f 

     f f 

     

      (1)

    where coefficients  j   denote case feature weights that

    generally attained by expert experience. The cases that satisfy

    the following condition

    T

    z

    T

    1, ,

    T T

    1, , 1, ,

    SIM ( , )

    0.95, max (SIM ( , )) 0.95

    max (SIM ( , )), max (SIM ( , )) 0.95

    k k 

    k k k m

    k k k k  k m k m

     F F y

     F F 

     F F F F 

     

      (2)

    will be retrieved as the matching cases with ranking in

    descending order of SIMk  .

    c) Case reuse. Suppose that the matching cases areM M M:{ }, 1, ,r r r 

    C F J r R   , where  R  is the number of

    matching cases. The case solution TS   of T T T1 2( , , F f f     T

    9, ) f   can be obtained by

    3T M

    1

    3TT

    1

    M

    1

    (SIM ( , ) )

    , 3

    SIM ( , )

    , 3

    r r r 

    r r 

     F F s

    if r S 

     F F 

     s if r 

       

      (3)

    6169

  • 8/17/2019 Intelligent Dynamic Modeling for Online Estimation of.pdf

    4/5

    TABLE I

    CASE REPRESENTATION 

    Case descriptor ( F ) Case solution (S )

    1 s  

    ( ) F C t    ( ) F  A t    ( ) F  R t    ( )S  K t    ( 1) F C t     ( 1) F  A t   ( 1) F  R t   ( 1)S  K t   ( 1) zt  B t   ( ) zt  B t   

    TABLE II

    I NITIAL CASE BASE OF CBR  SOFT-SENSOR SYSTEM 

    Case descriptors ( F ) Case solution (S )

    ( ) F 

    C t    ( ) F 

     A t    ( ) F 

     R t    ( )S 

     K t    ( 1) F 

    C t     ( 1) F  A t     ( 1) F  R t     ( 1)S  K t     ( 1) zt  B t     ( ) zt  B t   

    167.42 28.82 9.02 3.633 166.22 27.852 8.92 3.734 1451.7 1471.1

    166.64 28.23 9.07 3.834 167.34 28.21 8.87 3.721 1463.2 1473.3

    168.76 27.42 8.46 3.656 166.46 27.82 8.25 3.726 1510.2 1500.1

    167.83 27.62 8.71 3.728 168.23 28.22 8.93 3.672 1502.7 1501.3

    … … … … … … … … … …

    0 10 20 30 40 50 60 701475

    1480

    1485

    1490

    1495

    1500

    1505

    1510

    1515

       B  u  r  n   i  n  g  z  o  n  e   t  e  m  p  e  r  a   t  u  r  e ,     ℃

    Sampled data

    Estimated value

    Sampled value

     Fig.4 Testing results of burning zone temperature estimation with the

     proposed method

    10 20 30 40 50 60 700

    1

    2

    3

    4

    5

    6

    7

    8

    9

    10

    Sampled data

       E  s   t   i  m  a   t  e   d  e  r  r  o  r

     Fig.5 Estimating error of with the proposed method

    d) Case revision. The case revision is a very important

    issue in the CBR-based decision system. After the grinding

    system obtains the solution TS   as the estimation of BZT for

    the CBR soft-sensor system, if the actual BZT is better, it

    confirms that the formerly estimation is reasonable. Therefore,

    there is no need to carry out case revision. Otherwise, it needs

    to revise the case and store this revised new case. The detailed

    revision procedures can be consulted in Ref. [3].

    1480 1485 1490 1495 1500 1505 15101480

    1485

    1490

    1495

    1500

    1505

    1510

    Estimated value

       S  a  m  p   l  e   d  v  a   l  u  e

     Fig.6 Scatter diagram of the burning zone temperature estimations with

     proposed method

    IV. I NDUSTRIAL APPLICATIONS 

    In this section, we will use the above proposed CBR

     based dynamic soft-sensor method to model a cement

     production line. In the past, the BZT could not be obtained

    online and closed-loop control for it could not be realized in

    this cement production line. Using the proposed data filter

    method on the sampled data, collect 110 groups of sampled

    data from the industrial process was collected to develop the

    initial case base for the CBR soft-sensor system. A partial

    sequence of the case data of the initial case base is shown in

    Table II.

    The prediction effect of the developed dynamic CBR-

     based soft-sensor under a wide range of operation conditions

    is shown in Fig. 4 and Fig.5. It can be seen that the developed

    CBR soft-sensor system obtains satisfactory performances. No

    matter what operation conditions are changed in the cement

    rotary kiln process, the output of the developed soft-sensor

    can estimate the actual burning zone temperature very well.

    According to statistical analysis, the average absolute

    estimation errors are small than 3.2.

    The performance of estimation can also be visualized by

     plotting the measured results against the predicted ones, which

    6170

  • 8/17/2019 Intelligent Dynamic Modeling for Online Estimation of.pdf

    5/5

    as shown in Fig.6. The abscissa of this figure is the value of

    actual measurement, and corresponding coordinate is the

    value of estimation with the proposed algorithm. The closer

    distribution of splashes gets to the black diagonal line, the

     better estimation effects are realized. When the estimated

    values match the measured ones, all points would lie on adiagonal. It can be seen from Fig.6 that the proposed modeling

    method gives the best estimation of the burning zone

    temperature. Although some points are relatively far from the

    diagonal line, the prediction of the proposed model is closer to

    the actual value very well. Such results show that this

     predictor can satisfy the requirement of BZT control.

    IV. CONCLUSIONS 

    The burning zone temperature in the cement rotary kiln

     process is a very important technical index, on which the

    sinter quality mainly relies. However, due to the complex

    dynamic charactertics in terms of nonlinearity, large timedelay and time-varying, it is difficult to online measure the

     burning zone temperature using conventional instruments. In

    this paper, an intelligence-based dynamic soft-sensor

    modeling approach for burning zone temperature using CBR

    estimation technique is proposed in this paper. Industrial test

    results show that the developed soft-sensor can online

    estimate the burning zone temperature very well.

    R EFERENCES 

    [1] 

    X. J. Zhou, H. Yue, et al, “Supervisory control for rotary kiln temperature

     based on reinforcement learning,”  Lecture Notes in Control and

     Information Sciences, vol. 344, pp. 428-437, 2006.

    [2] 

     N. Fallahpour, A. Fatehi, et al, “A supervisory fuzzy control of back-endtemperature of rotary cement kilns,”  Proceeding of International

    Conference on Control, Automation and Systems, 2007 in COEX, Seoul,

     Korea, pp. 429-434.

    [3]  Q. B. Huang, X .F. Lin, S. J. Song, “Model of cement rotary kiln based on

    Elman neural network and design of DHP controller,”  Journal of System

    Simulation, vol. 23, no. 3, pp. 583-587, 2011.

    [4]  M. Jarvensivua, K. Saari, S. L. Jamsa-Jounela, “Intelligent control system

    of an industrial lime kiln process,” Control Engineering Practice, vol.9,no. 6, pp.589-606, 2001.

    [5]  M. Jarvensivua, J. Esko, A. Oilli, “Intelligent control a rotary kiln fired

    with producer gas generated from biomass,”  Engineering of Artificial

     Intelligence, vol. 14, no.5, 2011.

    [6]  R. Zanovello, H. Budman, “Model predictive control with soft-constraints

    with application to lime kiln control,” Computers and Chemical

     Engineering, vol.23, no.6, pp.791-806, 1999.

    [7] 

    D. Soumitra, B. Wierenga, A. Dalebout, “Case-based reasoning systems:

    from automation to decision-aiding and stimulation,”  IEEE Trans. on

     Knowledge and Data Engineering , vol.19, no.6, pp.911-922, 1997.

    [8]  J. L. Kolodner, “An introduction to case-based reasoning,”  Artif. Intell.

     Rev. vol.6, no.1, pp.3-34, 1992.

    [9]  A. Aamodt, E. Plaza, “Case-based reasoning: Foundational issues,

    methodological variations and system approaches,”  Artif. Intell. Comm. 

    vol.7, no.1, pp.39-59, 1994.

    [10] 

    S. Wesley Changchien, Ming-Chin Lin, “Design and implementation of acase-based reasoning systemfor marketing plans,”  Expert Systems with

     Applications, vol.28, no. 1, pp. 43-53, 2005

    [11] H. X. Li and S. Guan. “Hybrid intelligent control strategy. Supervising a

    DCS-controlled batch process,” IEEE Control Systems Magazine, vol. 21,

    no. 3, pp. 36-48, 2001

    6171