59
I EXECUTIVE SUMMARY INTRODUCTION/BACKGROUND Steel industries face two main problems: steel quality and energy consumption. Both problems are linked to the reheating furnace. Many scholars have proposed several mathematical models for the furnace and heating process. To study these problems, construct a mathematical model for furnace and steel slab temperature is necessary. This paper applies the control algorithm on proposed model and evaluates the performance in MATLAB/SIMULINK which is a cheap way to study before put into the actual process. The control strategies include PID feedback control and expert experience feed forward control. AIMS AND OBJECTIVES Project Aim: 1. To find an appropriate heating curve for slab. 2. To look at what performance can be achieved with PID approaches and feed forward control based on simulating model Project Objectives: 1. To obtain an appropriate model of walking-beam furnace and the slab heating model in MATLAB. 2. Research PID controller and feed forward control. 3. Implement PID and feed forward control on the model. 4. Evaluate potential approaches (strengths and weaknesses).

EXECUTIVE SUMMARY - · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

Embed Size (px)

Citation preview

Page 1: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

I

EXECUTIVE SUMMARY

INTRODUCTION/BACKGROUND

Steel industries face two main problems: steel quality and energy consumption.

Both problems are linked to the reheating furnace. Many scholars have proposed

several mathematical models for the furnace and heating process. To study these

problems, construct a mathematical model for furnace and steel slab temperature is

necessary. This paper applies the control algorithm on proposed model and evaluates

the performance in MATLAB/SIMULINK which is a cheap way to study before put

into the actual process. The control strategies include PID feedback control and expert

experience feed forward control.

AIMS AND OBJECTIVES

Project Aim:

1. To find an appropriate heating curve for slab.

2. To look at what performance can be achieved with PID approaches and feed

forward control based on simulating model

Project Objectives:

1. To obtain an appropriate model of walking-beam furnace and the slab heating

model in MATLAB.

2. Research PID controller and feed forward control.

3. Implement PID and feed forward control on the model.

4. Evaluate potential approaches (strengths and weaknesses).

Page 2: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

I

ACHIEVEMENTS

This paper proposes furnace and slab temperature model and applies PID

feedback control and expert experience feed forward control for that. Finally, simulate

that in the MATLAB/SIMULINK.

CONCLUSIONS / RECOMMENDATIONS

This paper first constructs furnace and slab temperature model and then simuliate

in MATLAB/SIMULINK and then proposes dynamic optimization control strategy of

the reheating furnace set value based on PID feedback control and expert experience

feed forward compensation. Finally, compare these two control strategies result and

prove the latter one is better.

Although this paper proposed the mathematical model of the furnace and the

furnace temperature dynamic optimization settings for a certain research, the

following areas for further work will be done due to the complexity of the heating

process.

Page 3: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

I

ABSTRACT

To improve slab quality and decrease the energy consumption, it is necessary to

optimize the heating curve of slab. This paper presents a study of slab and furnace

temperature modeling and corresponding control strategies. Considering the cost and

the difficulties in applying control strategies in real production line, modeling in

MATLAB/SIMULINK first is the best choice. When the model is constructed, PID

with the feed forward compensation is applied for the model which performance is

satisfactory.

Keyword:

Reheating furnace model, Slab temperature model, PID, Feed forward

compensation.

Page 4: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

TABLE OF CONTENTS

ABSTRACT .............................................................................................................................. 3

Chapter 1- Introduction ...................................................................................................... 1

1.1. Background and Motivation ...................................................................................... 1

1.1.1 The development of furnace .............................................................................. 1

1.1.2 Furnace temperature control .............................................................................. 2

1.2 Literature review .............................................................................................................. 3

1.3 Problem ............................................................................................................................ 6

1.4 Aims and Objectives ........................................................................................................ 6

1.5 Project Management ........................................................................................................ 7

Chapter 2- Model of the walking-beam reheating furnace .............................................. 9

2.1 Introduction of walking-beam reheating furnace ............................................................. 9

2.2 The model of slab temperature....................................................................................... 10

2.2.1 The optimization of slab temperature curve............................................................ 10

2.2.2 Slab temperature tracking ....................................................................................... 12

2.3 The model of furnace temperature ................................................................................. 15

2.4 Optimal setting for furnace temperature ........................................................................ 18

2.5 Simulation ...................................................................................................................... 20

2.5.1 The entire model ..................................................................................................... 20

2.5.2 Set point of furnace temperature ............................................................................. 21

2.5.3 Slab temperature model .............................................................................................. 22

2.5.4 Furnace temperature ................................................................................................ 25

2.5.5 Disturbance ............................................................................................................. 29

2.5.6 Feedback control ..................................................................................................... 30

2.6 Conclusion and new problem ......................................................................................... 32

Chapter 3- The optimization of furnace temperature set point based on feed forward

compensation

Page 5: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

3.1 Introduction of feed forward compensation ................................................................... 33

3.2 Theory of feed forward compensation ........................................................................... 33

3.3 Simulation ...................................................................................................................... 36

3.3.2 Result ...................................................................................................................... 41

3.4 Validation ....................................................................................................................... 43

3.5 Conclusion ..................................................................................................................... 44

Chapter 4- Summary ......................................................................................................... 45

Reference ................................................................................................................................ 47

Page 6: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

1

Chapter 1- Introduction

1.1. Background and Motivation

In steel industries, reheating furnace brings a huge amount of energy

consumption. Many scholars worked on energy saving control problem of reheating

furnace and come out with variety of optimal control strategies [1-11]. However, due

to the characteristics of steel industry, there always be unpredictable parameters

during the heating furnace design, or advanced control algorithms cannot be tested

directly in the actual process. Hence, it is necessary to develop a mathematical model

of reheating furnace which can be used not only to determine some undetectable

parameters according to the real process but also to apply some advanced control

algorithm in offline simulation providing foundation for online control.

1.1.1 The development of furnace

Reheating furnace is the main equipment used in steel industry rolling slab by

heating to a certain temperature distribution. According to the different ways of

heating it can be divided into two types: cycle and continuous. Continuous heating

furnace is most widely used in the current production. Cycle furnace is heating slab in

a fixed position; it does not apply to the case of mass production. A continuous

heating furnace is heating slab during the slab moves from the furnace entrance has

been moved to its outlet, and in this process.

With the increase in international demand for steel, mill toward high efficiency

and large capacity development, which corresponds to increase furnace load in the

limited space increased heating capacity. Therefore, continuous furnace is developed

towards the multi-stage, and thus appeared the walking-beam reheating furnace.

Page 7: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

2

1.1.2 Furnace temperature control

With the rapid development of computer technology, scholars from various

countries made use of computer technology to optimize the metallurgical furnace

control, such as computer-controlled mathematical furnace model and the optimal

combustion control, and achieved some economic benefits. However, due to the

complexity of the actual production system and complex slab heating process affected

by various factors, the model is non-linear distributed parameter system. In general,

advanced computer-controlled furnace strategy is not mature and there is rarely

success application in furnace control.

Furnace mathematical models generally can be classified into empirical ones and

theoretical ones: empirical model of the furnace is to obtain main factors may reflect

the furnace based on analyzing a large number of field experiments and statistical data,

which is relatively simple with narrow applications and inability to meet modern

multi steel production; while theoretical one is to formulate mechanism model of slab

heating process through the finite element analysis or discrete the slab by finite

difference, then the unknown parameters in the equation can be determined according

to the experiment results. However, there are great difficulties to establish an accurate

mathematical model due to the complexity of the heating process and modeling

methods shortcomings.

The model of furnace is actually a mathematical description of thermal processes,

which can reveal the basic law of thermal processes occurring in the furnace to

determine the quantitative relationship between the parameters of heating process. It

can be used to study the thermal theory, furnace design and thermal process

optimization by computer. In steel rolling production, in order to meet the slab heating

quality and yield requirements, it is necessary to establish the reliable automatic

control system to monitor the temperature of the slab accurate and directly. Since the

Page 8: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

3

temperature distribution inside the slab still cannot be achieved on real-time online

testing, the steel temperature is estimated by slab heating process mathematical

models so that optimal control of slab heating temperature can be achieved.

1.2 Literature review

Continuous furnace mathematical model have been utilized for online control in

a few developed countries in the late 1960s. In the mid-1980s, research on

mathematical models began to be more advanced; the study focused more on the

automatic control strategy. However, continuous furnace mathematical model is still

the basis to achieve optimal control of the furnace. [1, 2]

Slab heating process involves the gas flow, fuel combustion, heat transfer,

thermal conductivity, and billet internal oxidation, decarbonization and other complex

physical and chemical phenomena, which depends on the furnace structure,

production operations and many other relevant factors. Such a complex thermal

process is difficult to be described by simple mathematical equations and also more

difficult to solve with internal and external random factors interfering and large inertia,

pure hysteresis nonlinearity distribution. Therefore, no matter for online controlling or

offline design or calculation, the model should be simplified according to the purpose.

For different subjects and research purposes, many scholars have proposed

various forms of furnace mathematical models. Misaka. J.& Takahashi.R.[3] made use

of the total heat absorption rate method to establish a mathematical model for the

prediction of the slab temperature and achieved some energy savings. Pike, H. E.&

Citron, S. J.[4] utilized distributed parameter theory for modeling and applied

approximate lumped parameter model to study the static and dynamic optimization of

furnace model. Wick, H. J.[5] took use of Kalman filter technology to estimate slab

temperature distribution, but the surface temperature of the slab can be obtained by

Page 9: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

4

this method which limits its further use.

A.Kusters [6] proposed a parameter estimation using multivariate methods and

established a multisession walking-beam furnace ARX model. This method claims

that the furnace is divided into six zones based on the structural characteristics of the

furnace, the furnace temperature model of each zone is established taking into account

the mutual coupling of each zone. Finally, least square method is applied to obtain

various parameters.

Yoshitani, N.& Ueyama,T.& Usui, M. [7] developed an optimal furnace control

system consist of the best heating state model, slab temperature model when unloaded

and optimal set point to solve the problem of slab variety specification large changes

or the precise control of temperature and temperature uniformity. They proposed two

method to reduce the energy consumption and improve the quality of the slab heating:

first one is to modify the temperature of the slab heating curve in real time by the

online simulator based on a nonlinear mathematical and distributed parameter model;

Second one is adopting some means of accelerating optimal process to make the

control effect more obvious. Such model is widely used.

Yongyao Yang, YongZai Lv, etc. [8-9] proposes furnace discrete state space

model and optimal control theory based computer control strategy. First, list PDE and

its associated two-dimensional boundary for slab thermal heat conduction, then

transform a series of subsystems associated with large discrete state equations using

the system decomposition and discretization methods, and the slab temperature

distribution at any position at each time can be calculated in real time according to the

actual parameters furnace and each section slab initial parameters. Furthermore taking

this model as a basis, design optimization strategy based on heuristic continuous

furnace computer control system. Control strategy is divided into two parts: the steady

state optimization calculation and dynamic compensation setting value for furnace

Page 10: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

5

temperature. Steady-state optimization used heuristic search methods, while the

dynamic compensation is based on feed forward - feedback principle in order to

achieve lower power consumption.

With modern rolling to a continuous, large-scale, high-speed, high precision and

multi-species direction, originally simply applied some of the control algorithm DDC

controls have not meet production needs. Because these algorithms are combustion

control value from the perspective of the control design, and the advanced computer

control of the furnace, since the set value is in accordance with certain indicators

calculated optimal performance, and its combustion control with strong servo system

features. For optimal control settings, because the furnace complex conditions of

production, the control system under different conditions may be different quality

requirements, the use of a single setting optimization mode often can not make it in a

variety of conditions have reached the optimal control performance. To this end,

Yangyong Yao, Liang Jun, etc. [10] proposed a multi-mode control scheme furnace

settings. Based on the fundamental theorem of heat transfer and energy balance

principle, the use of time, space discretization technique established for estimating the

temperature distribution in the furnace billet heating furnace nonlinear discrete state

space model, based on the optimization of certain propositions, using heuristic search

strategy path extension to solve it. Meanwhile, in the furnace temperature setpoint

based on optimal control settings increase the fuel flow control, in order to achieve

multi-mode oven setting control.

Dirk. S.& Arend. K. [11] proposed the system which has two components: the

slab temperature calculation model and the controller to achieve the desired

temperature. The error of output of the mathematical model is less than ± 20 degrees

out of approximate 95% of the slab temperature calculation, adjust the actual

temperature of the slab according to the slab condition in the furnace in each zone and

Page 11: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

6

the slab ideal heating curve. System has two control loops: the main loop is the

furnace temperature set point of each zone, which is obtained by calculated by the

actual temperature and the target temperature; auxiliary loop is set value of fuel

calculation. The system achieved better performance in Tata Iron & Steel Company

(India).

Other bachelors [12-15] also studied the furnace temperature set point dynamic

compensation, which is based on feed forward - feedback principle, to compensate

furnace temperature setting point according to the selected quasi-steady-state

conditions and the current differences between the actual working conditions and set

point. The main factors considered are rolling rhythm, the mean temperature

difference of slab in preheating zone and heating zone, the estimated and the optimum

temperature distribution of the slab and the detect signals of the slab surface

temperature. Utilizing feed-forward compensation to correct for furnace billet speed

fluctuations, and using state feedback to amend for other factors caused the slab

temperature distribution and optimum temperature distribution deviation. In order to

decrease the impact of model error and reinforce the adaptability and robustness of

the system, the real-time signal of slab surface temperature when unloading is

collected as a secondary feedback signal further to compensate.

1.3 Problem

As discussed in section 1.2, there are quite a lot of strategies proposed by many

scholars. Here comes the problem: which strategy should be applied leading to better

performance? As the difficulties in obtaining data in real process and applied the

strategy for actual system, simulating in computer seems to be important.

1.4 Aims and Objectives

Project Aim:

Page 12: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

7

1. To find an appropriate heating curve for slab.

2. To look at what performance can be achieved with PID approaches and feed

forward control based on simulating model

Project Objectives:

1. To obtain an appropriate model of walking-beam furnace and the slab heating

model in MATLAB.

2. Research PID controller and feed forward control.

3. Implement PID and feed forward control on the model.

4. Evaluate potential approaches (strengths and weaknesses).

1.5 Project Management

Chapter one introduces the control strategies of furnace, analyzed and

summarized the furnace optimization control research status and significance,

pointing out the problems, and list the structure of whole project.

Chapter two introduces the furnace structure and analyzes the necessity of

optimizing the slab heating curve which is because of the difficulty to measure the

actual temperature distribution directly, so that the slab temperature

prediction/tracking model is introduced. The furnace dynamic model and set point of

the furnace temperature is also proposed. Finally, simulate the basic model in

MATLAB/SIMULINK.

Chapter three claims that there is large time delay causing normal control cannot

work on time, and proposes a feed forward dynamic compensation for furnace

temperature set point. Then simulation results shows better performance than model

proposed in Chapter 2.

Page 13: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

8

Chapter 4 summarizes the above chapter and gives a conclusion and future target

of the project.

Page 14: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

9

Chapter 2- Model of the walking-beam reheating furnace

This chapter will discuss the theory of furnace and slab temperature calculation

for modeling and propose the basic walking-beam reheating furnace temperature

model. The optimal slab heating curve and corresponding furnace temperature

distribution will be discussed as well.

2.1 Introduction of walking-beam reheating furnace

Walking-beam furnace reheating furnace, one of the main equipment in the steel

rolling industry, is used to heat slabs with less energy consumption and more accuracy

in temperature based on controlling the temperature of furnace, air-fuel ratio, air and

fuel pressure and furnace pressure, etc. The slabs are loaded from the furnace head

and driven towards furnace end by the walking beam in a specific speed. In other

words the zones slabs pass through are preheating zone, heating zone I, heating zone

II and soaking zone respectively.

The structure of the walking-beam reheating furnace is shown in Fig 1. There are

several burners in each zone except preheating zone to maintain the furnace

temperature. Hence there is no control in preheating zone which effect by other zone.

In need of analysis, other three zones can be divided into 6 parts, which are upper

heating zone I, lower heating zone I, upper heating zone II, lower heating zone II,

upper soaking zone, lower soaking zone.

Page 15: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

10

Figure 1 The structure of the walking beam reheat furnace [16]

The slab once pushed into preheating zone will be heated by waste gas from

heating zone I. Then heating zone I and II will do the heating job making the slab

reaching the required temperature rapidly. Sent into soaking zone, the surface and

center temperature of slab will reach a balance which leads to improvement of

strength, hardness and toughness of the slab.

2.2 The model of slab temperature

This chapter will describe the model of slab temperature including ideal heating

curve and heat transfer function of slab.

2.2.1 The optimization of slab temperature curve

During the heating process, the reheating furnace provides sufficient heat for the

slab to ensure that the slab can be heated to the specified temperature range. To reach

the target and minimize furnace fuel consumption and oxidation loss of the slab, an

optimal temperature curve which depends on the kinds of the slab will be present. The

best performance will be ensured only if the slab following the optimal curve.

The optimal slab temperature curve corresponds to the choice of heating method.

There are three kinds of heating method which are shown in Fig 2.

Page 16: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

11

Figure 2 Illustration of different methods for slab heating [17]

( -the surface temperature of slab, -the mean temperature of slab)

For method (a), the temperature of slab raises slow in the preheating zone

followed by the large temperature gradient in heating zone which will lead to the large

temperature difference of the surface and inner of the slab. This situation is not

acceptable because the uneven heating of the slab will make the rolling process

become more difficult.

For method (c), the temperature of slab soars in the preheating zone causing the

large temperature difference of the surface and inner of the slab. Moreover, most slabs

are still in elastic state under 500-600°C which means large thermal stress caused by

soaring temperature will lead to defect of slab as well as reduction of yield.

For method (b), the heating temperature is moderate in preheating zone resulting

in low thermal stress. Then the slab is in plasticity state after reaching 600°C while

the furnace temperature increases rapid enough making the surface of the slab reach

the required temperature in heating zone. Pushed to the soaking zone, the center

temperature of the slab approaches to the surface gradually. Because of the low

difference of surface and center, such slab is good to be rolled. Oxidation loss of the

slab decreases as well in terms of less time in heating zone.

Page 17: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

12

In a word, the method (b) is the best one for heating slab.

2.2.2 Slab temperature tracking

Thermocouples are installed to detect the furnace temperature surrounding slabs

rather than measuring the slab temperature directly because of the difficulty in direct

measurement. To solve this problem, a mathematical model of the furnace and slab

temperature tracking and distribution are proposed using measurable data to estimate

the slab temperature and its linkages with the preset furnace temperature value.

Based on mechanism knowledge and experiments, the slab temperature

prediction model can be established. For further application in on-line calculation, the

model should not be too complicated. Hence, there are several assumptions to

simplify the heating process:

(1) The furnace temperature is one-dimensional linear distribution along the

direction of furnace of furnace while slab temperature being along the slab

thickness direction.

(2) Assume the furnace temperature corresponding to the position of slab to be

the basis temperature calculation of heat transfer.

(3) Ignore the heat transfer between slab and walking beam, slab and fixed beam.

(4) Since heat radiation in furnace can be absorbed within short distance and

walls are installed between each zone, heat radiation between neighboring

zones can be ignored.

(5) The specific heat capacity of each layer is considered to be equal.

Based on assumptions above, asymmetric one-dimensional heating conduction

equation of slab can be described as below:

Page 18: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

13

{

𝛿𝑇

𝛿𝑡=𝑎

𝛿2𝑇

𝛿𝑡2

𝑇(𝑥,𝑡)| 𝑡=0=𝑇0(𝑥)

𝜆𝛿𝑇

𝛿𝑥| 𝑥=

𝐻2

=𝑞𝑢

𝜆𝛿𝑇

𝛿𝑥| 𝑥=−

𝐻2

=−𝑞𝑑

(1)

Where H is the thickness of slabs, (𝑥, 𝑡) is slab temperature distribution along

the thickness direction, 𝜆 is heat diffusion coefficient, 𝑞𝑢 and 𝑞𝑑 is heat flux of slab

upper and lower surface respectively. One slab is divided into 5 layers along thickness

direction, as figure 3 shows:

Figure 3 Slab layers

𝑞𝑢 and 𝑞𝑑 can be calculated by equation (2):

𝑞𝑢 = εσ [(T𝑓𝑢 + 273)4− (T1 − 273)4]

𝑞𝑑 = εσ[(T𝑓𝑑 + 273)4− (T5 − 273)4]

Where σ is Boltzman constant which equals to 4.88*10-8

, ε represents radiation

coefficient, T1 and T5 is top and bottom surface temperature of slabs.

Using central difference method indicating that difference is expressed by the

mean value of forward difference and backward difference [5], equation (1) and (2)

can be rewrite as follows to calculate the distribution of slab temperature.

(2)

Page 19: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

14

qdTTTTTTTTTTTqdTT

TTTTT

dx

ux

N

N

N

N

N

m

m

m

m

m

m

m

m

m

m

2)(

)(2

)(2

)(2

2)(

000

)(200

0)(20

00)(2

000

0

5

0

4

0

5

0

4

0

3

0

4

0

3

0

2

0

3

0

2

0

1

0

2

0

1

5

4

3

2

1

(3)

where m = 𝐶𝑝 ∗ 𝛾 ∗ 𝑑𝑥2 Δ𝑡⁄ , 𝜆 is the heating transfer coefficient, 𝑑𝑥 is the

thickness of each layer which equals to H/4, 𝐶𝑝 is the specific heat capacity, 𝛾 is the

specific gravity, 𝑖𝑁 represents the slab temperature at current time and 𝑖

0

represents slab temperature at last time, 𝑞𝑢 represents heat flux of upper slab surface,

𝑞𝑑 represents heat flux of lower slab surface, Δ𝑡 is the differential model calculation

step which yields that Δ𝑡 𝐶𝑝 ∗ 𝛾 ∗ 𝑑𝑥2⁄ ≤ 0.25 to ensure no shock on the boundary

value[6]. 𝐶𝑝 𝑎𝑛𝑑 𝜆 are functions of slab temperature, which is shown in equation (4):

𝐶𝑝 = 408.7 + 0.199 + 810.9exp (−𝛼| − 768|)

𝜆 = 55.85 − 31.23/𝑐ℎ[0.003( − 1208 + 273)]

Where, when T < 768°C, ∝= 0.0099; when T ≥ 768°C, ∝= 0.0261.

The mean temperature of slab can be calculated as equation (5) [9] shows:

=𝐶1𝑇1

𝑁+2(𝐶2𝑇2𝑁+𝐶3𝑇3

𝑁+𝐶4𝑇4𝑁)+𝐶5𝑇5

𝑁

𝐶1+2(𝐶2+𝐶3+𝐶4)+𝐶5 (5)

From the relationship of slab (A3) temperature and specific heat capacity as

figure 4 shows, when the temperature difference is small, the specific heat capacity

gradient is small, hence the assumption of 𝐶1 ≈ 𝐶2 ≈ 𝐶3 ≈ 𝐶4 ≈ 𝐶5 at the same time

(4)

Page 20: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

15

of different parts of slab.

Figure 4 Relationship between slab temperature and specific heat capacity

Hence, equation (5) can rewrite as follows,

=𝑇1𝑁+2(𝑇2

𝑁+𝑇3𝑁+𝑇4

𝑁)+𝑇5𝑁

8 (6)

Based on model proposed above and furnace temperature distribution, the slab

heating curve and tapping temperature can be obtained.

2.3 The model of furnace temperature

The furnace temperature is commonly referred to the temperature detected

directly by the thermocouple. The set point of temperature for temperature control

loop regulate valve opening to control gas flow causing the change of furnace

temperature and corresponding slab temperature. Therefore, it is necessary to

establish the dynamic model of the furnace temperature to find the optimal furnace

temperature set point based on slab heating indicator and economic indicator.

Furnace temperature modeling is essentially a time-varying and nonlinear heat

transfer problem, including radiation, conduction and convection. This heat transfer

problem holds lag time and time constant varying from the furnace load (the numbers

of slab). However, the numbers of slab is always the maximum one during the actual

production due to maximizing the profit and efficiency. Hence, the assumption that

the lag time an time constant do not change during the whole process is reasonable.

Page 21: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

16

Basically, this process can be converted into a multi-volume with pure time-delay

process, that is to say bigger time constant becomes dominant time constant of the

piecewise process while smaller ones are combined into one being equivalent to a

pure time delay. As Equation (6) shows:

𝐺𝑝(𝑠) =𝐾𝑝

∏ 𝑇𝑗 +1𝑛𝑗=1

⇒𝐾𝑝𝑒

−𝜏𝑠

𝑇 +1 (7)

Equation (3) is based on piecewise controllability of each zone. Actually, the

model of furnace temperature is complicated and hard to demonstrate by math model.

To simplify the model, what to be controlled is the average temperature of a region

near sensors rather than the entire furnace temperature field.

According to furnace heat transfer characteristics, the furnace can be divided into

4 parts which are mentioned in section 2.1, and there is no control in preheating zone.

Since each zone of the furnace linked by the open loop, there is a strong coupling

between adjacent zones. For example, fuel flux changes in heating zone I will not

only affect the temperature of heating zone I, but also heating zone II and preheating

zone, and so on.

Based on the analysis of heat transmission characteristic in furnace, heat flux is

mainly transmits from unloading side towards loading side. Therefore, assume that

there is only unidirectional coupling between adjacent zones, that is to say a furnace

zone temperature changes that only affect the loading side one. Thus, the thermal

transfer characteristics can be approximated in figure 5.

Page 22: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

17

Figure 5 Illustration of the thermal transmission in furnace

Where zone 1.3.5 represent lower heating zone I, lower heating zone II, lower

soaking zone respectively, while zone 2.4.6 represent upper heating zone I, upper

heating zone II, upper soaking zone respectively. Fi (i=1,...6) is fuel flow for each

zone; Tfi (i=1,...6) is furnace temperature for each zone.

The model shown in Figure 4 divides the furnace into 3 subsystems which can be

modeled separately. The advantage of this modeling approach is to fully consider the

furnace coupling effects between the various zones, so that the model can describe the

characteristics of the furnace more accurately. According to Figure 4, the

multivariable furnace temperature dynamic model is described by the following heat

balance equation:

𝑑𝑇𝑓𝑖(𝑡)

𝑑𝑡= 𝑏𝑖1 ( 𝑓(𝑖+2)(𝑡 − 𝜏𝑖1) − 𝑓𝑖(𝑡)) + 𝑏𝑖2 ( 𝑓𝑗(𝑡 − 𝜏𝑖2) − 𝑓𝑖(𝑡)) + 𝑏𝑖3Δ𝐹𝑖(𝑡 −

𝜏𝑖3)

𝑑 𝑓𝑗(𝑡)

𝑑𝑡= 𝑏𝑗1 ( 𝑓(𝑗+2)(𝑡 − 𝜏𝑗1) − 𝑓𝑗(𝑡)) + 𝑏𝑗2 ( 𝑓𝑖(𝑡 − 𝜏𝑖2) − 𝑓𝑗(𝑡)) + 𝑏𝑗3Δ𝐹𝑗(𝑡

− 𝜏𝑗3)

Where 𝑓𝑖(i=1.3.5), 𝑓𝑗(j=2.4.6) represent the furnace temperature of ith zone,

Δ𝐹𝑖 (i=1.3.5), Δ𝐹𝑗 (j=2.4.6) represent corresponding changes of fuel flux,

𝑏𝑖𝑘, 𝑏𝑗𝑘(i=1.3.5;j=2.4.6;k=1.2.3) is constant, 𝜏𝑖𝑘, 𝜏𝑗𝑘 (i=1.3.5;j=2.4.6;k=1.2.3) is pure

(8)

Page 23: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

18

delay time. Finally, the furnace temperature dynamic model can be identified as long

as getting enough data from actual process.

2.4 Optimal setting for furnace temperature

Section 2.2.1 has discussed the optimal curve of heating slab. To achieve this,

optimal furnace temperature distribution need to be proposed. That is to say, the

optimal temperature curve of the slab corresponds to the optimal distribution curve of

furnace. Therefore, it is necessary to find the optimal furnace temperature distribution

curve. Once the optimal furnace temperature distribution curve is obtained, the

optimum furnace temperature distribution can be achieved through the furnace

combustion control system. The purpose of optimizing the furnace control, in fact, is

to find the best value of the furnace in each zone within the allowable range which is

the furnace temperature set point, in order to minimize heat energy consumption and

to meet the slab requirements. Furthermore, the furnace temperature applied is

changing along the furnace length direction rather than detecting by thermocouples or

preset one. Hence, the furnace temperature distribution needs to be done first.

Zhang.D.H. [8] has proposed a quadratic function along the furnace length

direction with constrains for the furnace temperature, as equation (9)

𝑓(𝑡) = 𝑑 + 𝑒𝑡 + 𝑓𝑡2 (9)

Where 𝑓(𝑡)the furnace temperature, t is the heating time. Equation (9) should

follow constrains below simultaneously:

1. The temperature of lower heating zone should reach a certain value.

2. At the time entering soaking zone (t1), the furnace should meet a required

temperature.

3. At the time unloading (tf), the slab should reach a certain temperature.

That is to say,

Page 24: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

19

1. When t=0, 𝑓1 𝑖𝑛 ≤ 𝑑 ≤ 𝑓1 𝑎𝑥

2. When t= t1, 𝑓2 𝑖𝑛 ≤ 𝑑 + 𝑒𝑡 + 𝑓𝑡2 ≤ 𝑓2 𝑎𝑥

3. When t= tf, 𝑓3 𝑖𝑛 ≤ 𝑑 + 𝑒𝑡 + 𝑓𝑡2 ≤ 𝑓3 𝑎𝑥

𝑓1 𝑖𝑛, 𝑓1 𝑎𝑥 , 𝑓2 𝑖𝑛, 𝑓2 𝑎𝑥, 𝑓3 𝑖𝑛, 𝑓3 𝑎𝑥 are determined by actual process.

Considering the quality of the slab and minimization the energy cost, the

objective function is proposed to meet the requirements of slab quality as less energy

consumption as possible, and this equation can be solved by fmincon function in

MATLAB:

min J = min {1

2𝑃[ (𝑡𝑓) −

∗ (𝑡𝑓)]2+

1

2𝑄[ (𝑡𝑓) − 𝑐(𝑡𝑓)]

2+

1

2𝑅 ∫ 𝑢(𝑡)2

𝑡𝑓𝑡=0

}

Constrains are shown below:

1. T(t + Δt) = F(T(t), 𝑓(t + Δt))

2. (t + Δt) − (t) ≤ Δ 𝑎𝑥

3. (t) − 𝑐(t) ≤ Δ 𝑐 𝑎𝑥

4. | (𝑡𝑓) − ∗ (𝑡𝑓)| ≤ Δ 𝑜𝑢𝑡

5. 𝑖𝑛(𝑡𝑖) ≤ (𝑡𝑖) ≤ 𝑎𝑥(𝑡𝑖)

6. 𝑓 𝑖𝑛(𝑡𝑖) ≤ 𝑓(𝑡𝑖) ≤ 𝑓 𝑎𝑥(𝑡𝑖)

7. 𝑓 𝑡 𝑖𝑛 ≤ 𝑓 𝑡 ≤ 𝑓 𝑡 𝑎𝑥

Where,

t : Heating time, or the corresponding position of slab in the furnace.

t𝑖 : Some key position of furnace, such as head and tail of each zone.

(10)

Page 25: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

20

t𝑓 : The whole heating time, or the length of the furnace.

(t), (t), 𝑐(t): Slab mean temperature, surface temperature ( 1) and center

temperature ( 3) at time t respectively.

Δ 𝑎𝑥, 𝑐 𝑎𝑥 , Δ 𝑜𝑢𝑡 : Maximum permission of slab heating rate, section

temperature difference and slab temperature difference when unloading respectively.

∗ (𝑡𝑓) : Expectation of slab mean temperature.

𝑎𝑥(𝑡𝑖), 𝑖𝑛(𝑡𝑖) : Maximum and minimum of slab mean temperature at t𝑖.

𝑓 𝑎𝑥(𝑡𝑖), 𝑓 𝑖𝑛(𝑡𝑖) : Maximum and minimum of furnace temperature at t𝑖.

𝑓 𝑎𝑥(𝑡𝑖), 𝑓 𝑖𝑛(𝑡𝑖) : Maximum and minimum of furnace temperature set point.

P, Q, R are weighted coefficient.

For equation (10), 1

2[ (𝑡𝑓) −

∗ (𝑡𝑓)]2

represents the requirement of slab

temperature when unloading, 1

2[ (𝑡𝑓) − 𝑐(𝑡𝑓)]

2 indicates slab temperature

difference between surface and center, 1

2∫ 𝑢(𝑡)2𝑡𝑓𝑡=0

shows the energy cost of furnace.

Hence, the weighted coefficient P, Q, R can be set according to the actual requirement.

The bigger P/Q/R is, the higher corresponding requirement is needed. What needs to

be concerned is P, Q ≫ R.

2.5 Simulation

Last 4 sections have proposed an almost complete static model for furnace and

slab temperature which can be regarded as steady state of the whole system. To verify

the reliability of this model, simulating in MATLAB/SIMULINK is a direct and

convenient approach.

2.5.1 The entire model

The structure of original system is shown in Fig 6. It is consist of three

Page 26: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

21

subsystems: furnace temperature model, furnace temperature distribution model and

slab temperature model. The input of the system is set point of furnace temperature

and output is slab temperature (mean slab temperature, slab temperature of each layer,

cross section difference).

Figure 6 Original model

2.5.2 Set point of furnace temperature

Taking A3 steel slab (200mm *200mm *3000mm) as an example. Assuming slab

heating interval is 90 seconds and walking step is 500mm, which corresponds to this

heated slab heating furnace is about 1.5 hours. Hence, dt and dx are 90s and 50mm

respectively. The density of slab is 7800kg/m3. Radiation coefficient ε equals 0.35.

Specific heat capacity can be calculated by equation (4). The parameters m and lam_s

are calculated by the block shown on the bottom of the model. The expected mean

temperature of the slab when unloading is 1085°C. The whole simulation is cover 1.5

hours (5400 second) which is the time one slab stays in the furnace.

According to requirement of actual process, the set point of furnace temperature

along the furnace length corresponding to the slab heating time is shown in Figure 7.

The furnace specific parameters are listed: effective length of the furnace is 29348mm;

the length of preheating zone, heating zone I, heating zone II, and soaking zone is

Page 27: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

22

13598mm, 5000mm, 6300mm, and 4450mm respectively. Code in Matlab file, the set

point curve along furnace length is plotted according to the cost function (7). The

range of each zone temperature is 1100 ± 5°C, 1150 ± 5°C, 1130 ± 5°C .

Figure 7 Set point of furnace temperature

2.5.3 Slab temperature model

Slab temperature model can be divided into two parts: radiative heat flux ,

temperature of each layer, slab mean temperature calculation and online parameter

calculation as figure 8 to 11 show. Radiative heat flux model is based on equation (2),

while temperature of each layer model and other two is based on equation (3) (5) and

(4) respectively.

0 500 1000 1500 2000 2500 3000 3500600

700

800

900

1000

1100

1200

Furnace temperature

Page 28: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

23

Figure 8 Top and bottom surface radiative heat flux

Figure 9 Temperature of each layer

Figure 10 Slab mean temperature

Page 29: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

24

Figure 11 Parameter calculation

The whole slab temperature tracking model is shown in figure 12.

Figure 12 Slab temperature tracking model

Hence, the optimal slab heating curve is simulated as figure 13 shows. Figure 13

demonstrates that the slab temperature rises as method (b) discussed in section 2.2.1

under such furnace temperature distribution. The controller designed on following

section will be based on the optimal curve.

Page 30: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

25

Figure 13 Optimal heating curve

2.5.4 Furnace temperature

Considering a 6-zone walking-beam furnace, as Figure 5 shows, it is clear that the

furnace heat is transferred from tail to head. The couple between neighboring zones is

not taken into account although they are coupled actually

Utilizing the decoupled furnace model proposed in reference [2]:

upper heating zone I: 𝐺𝑇𝑓𝑝1 =0.0051

𝑠 + 0.0105

lower heating zone I: 𝐺𝑇𝑓𝑝2 =0.0011

𝑠 + 0.0039

upper heating zone II: 𝐺𝑇𝑓𝑝3 =0.0005

𝑠 + 0.0026

lower heating zone II: 𝐺𝑇𝑓𝑝4 =0.0004

𝑠 + 0.0019

upper soaking zone: 𝐺𝑇𝑓𝑝5 =0.0047

𝑠 + 0.0015

0 1000 2000 3000 4000 5000 60000

200

400

600

800

1000

1200

time( s)

tem

pera

ture

°C()

Slab temperature

Section temperature difference

Set point

Page 31: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

26

lower soaking zone ∶ 𝐺𝑇𝑓𝑝3 =0.0027

𝑠 + 0.0056

The control flow of each furnace zone is shown in Figure 14,

Figure 14 Control flow of each zone

The inner loop is fuel flow control loop while the outer loop is zone temperature

control loop, where fsi is the optimal set point of ith zone, GTfci and GFci are furnace

and fuel flow controller, GFVi is a fuel control valve, GTfMi and GFMi are measure and

transfer devices, GFpi and GTfpi are furnace pipes and furnace models. Transfer

function of each zone is shown below:

𝐺𝑇𝑓𝑐𝑖 = 𝑘𝑝1 +1

𝑘𝑖1𝑠+ 𝑘𝑑𝑠

𝐺𝐹𝑐𝑖 = 𝑘𝑝2 +1

𝑘𝑖2𝑠

𝐺𝐹𝑉𝑖 =1

2𝑠 + 1

𝐺𝐹𝑝𝑖 =5

8𝑠 + 1

𝐺𝐹𝑀𝑖 =1

𝑠 + 1

𝐺𝑇𝑓𝑀𝑖 =1

10𝑠 + 1

Considering the maximum flow of the fuel pipe, it is necessary to add a saturation

Page 32: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

27

to limit the fuel flow. Hence, the system will be more reliable. According to actual

process, the fuel flow limitation of each zone is shown in table 2:

Zone Maximum fuel flow

Upper heating zone I 2500

Lower heating zone I 5000

Upper heating zone II 7000

Lower heating zone II 7000

Upper soaking zone 3000

Lower heating zone 3000

Table 2 Maximum fuel flow

Take lower heating zone I as an example, as figure 15 shows:

Figure 15 The Simulink model of lower heating zone I

PI controller parameters are listed in table 3,

Kp Ki

Fuel flow control loop 0.9 0.01

Page 33: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

28

Upper heating zone I 70 0.2

Lower heating zone I 28 0.1

Upper heating zone II 60 0.15

Lower heating zone II 78 0.25

Upper soaking zone 50 0.13

Lower soaking zone 48 0.1

Table 3 Parameters of PI controllers

Large numbers of experiments show that the smaller the gain of integrator and

the larger proportion between the gain of proportion and integrator are, the better

performance is. The furnace temperature and fuel flow of each zone are shown in

figure 16.

Figure 16 The furnace temperature and fuel flow of each zone

0 1000 2000 3000 4000 5000 60000

1000

2000

3000

time( s)

T/F

low

Upper Heating Zone I

0 1000 2000 3000 4000 5000 60000

2000

4000

6000

T/F

low

Lower Heating Zone I

0 1000 2000 3000 4000 5000 60000

2000

4000

6000

8000

time( s)

T/F

low

Upper Heating Zone II

0 1000 2000 3000 4000 5000 60000

2000

4000

6000

8000

time( s)

T/F

low

Lower Heating Zone II

0 1000 2000 3000 4000 5000 60000

1000

2000

3000

time( s)

T/F

low

Upper Soaking Zone

0 1000 2000 3000 4000 5000 60000

1000

2000

3000

time( s)

T/F

low

Lower Soaking Zone

Fuel Flow(Nm3/h)

Furnace temperature(°C)Fuel Flow(Nm3/h)

Furnace temperature(°C)

Fuel Flow(Nm3/h)

Furnace temperature(°C)

Fuel Flow(Nm3/h)

Furnace temperature(°C)

Fuel Flow(Nm3/h)

Furnace temperature(°C)

Fuel Flow(Nm3/h)

Furnace temperature(°C)

Page 34: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

29

As spoken at previous section, there is no control in the preheating zone, hence

the response at the first 13598mm which corresponds to interval from 0 to 2448

second is not taken into consideration. As shown, every time furnace temperature

changes, which can be regarded as step response, the overshoot is less than 15%

which is acceptable.

Use zone choosing model to combine all 6 zones temperature into final furnace

temperature. Zone choosing model is consist of 3 square waves and 1 step signal to

make the set point of different zone control specific zone temperature according to the

length of each zone mentioned in section 2.5.2, which is shown in figure 17.

Figure 17 Zone choosing model

2.5.5 Disturbance

As known, there are lots of factors will influence the quality of the slab. To

Page 35: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

30

simplify the disturbance, the assumption that all effect can be gathered to a total

changes of slab temperature brought by all disturbance is made. For the furnace model

above, assuming at time t, the heating process is effect by some disturbance causing a

large slab temperature error, model and result are shown in Fig 18.

2.5.6 Feedback control

The main control loops include the heating flux and furnace temperature control

loops of six zones which is proposed in section 2.5.4 and slab temperature control

loop being discussed in this section. The furnace temperature, heating flux and slab

temperature control loops form the cascade control loops. All controllers of these

three loops are PID controller. The structure of slab temperature control loop is shown

in figure 18. The parameter of PI controller is 0.12 and 0.0001 respectively without

overshoot based on energy saving principle.

Page 36: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

31

Figure 18 Feedback control

Figure 19 Slab temperature with disturbance

It is seen that the disturbance appears in the exit of heating zone I while heating

zone II and soaking zone have not been affected yet. The first slab effected moves

slowly in the furnace, when it reaches the exit of the furnace this error will always

0 1000 2000 3000 4000 5000 60000

200

400

600

800

1000

1200

time( s)

Tem

pera

ture

(°C

)

slab with disturbance

furnace temperature

Ideal T Slab

cross section

T Slab with disturbance

Page 37: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

32

exist, by the end this slab cannot meet the requirements.

2.6 Conclusion and new problem

In last section, the basic model of slab and furnace temperature is contrasted

followed by a problem that the system ability of resisting disturbance is weak. The

reason why PID controller is not applied to control the slab temperature directly is

that this model will be used for online controlling leading to the simpler model the

better, while using PID controller for slab temperature will definitely increase the

calculating time which will make the time delay become larger and the accuracy

become lower. One of the most important reasons is that if the furnace temperature is

controlled in good performance, there will be no need to control slab temperature any

more. That is because the furnace temperature and slab temperature are related, once

one of each is done, the other will be done as well.

Due to the long time heating process, feedback control of the whole process

brings large time delay and cannot reflect the change of the temperature on time. That

is to say, when the effect of disturbance is detected, there has been a lot of slab

unqualified. Hence, applying feedback control cannot reach the requirement. What

can be done is introducing the feed forward compensation based on the first slab

effect by disturbance,

In a word, the feed forward compensation is needed for the system to eliminate

the error of following slab temperature. The feed forward compensation will be

discussed at chapter 3.

Page 38: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

33

Chapter 3- The optimization of furnace temperature set point

based on feed forward compensation

This chapter applies feed forward compensation for the furnace temperature set

point and simulate in MATLAB/SIMULINK. Compare the result of PID feedback

control and feed forward compensation at the end of this chapter.

3.1 Introduction of feed forward compensation

Walking-beam reheating furnace is important equipment in the steel industry,

which main function is to heat the slab loaded and make its temperature and

uniformity meet certain requirements. If the temperature is below standard one, there

will be difficulties of rolling and damage to rolling equipment; in the contrary, if the

slab temperature is too high, there will be excessive oxidation of the surface and large

energy waste. Even if the slab temperature meets the requirements, the energy

consumption might not be the smallest. Hence,it is necessary to set the value of the

steady state dynamic calibration furnace.

Based on the theory proposed by Zhongjie Wang[29], a PID controller-based,

supplemented by expert experience compensating controls is proposed. Considering

all aspects of the impact between the furnace zones, feed forward control is applied to

control each furnace zone through dynamic compensation, simulation results in the

end of this chapter will prove it to be a better approach.

3.2 Theory of feed forward compensation

First, set steady furnace temperature value according to the static model built in

chapter 2. Then according to the actual measured slab temperature at exit and section

temperature difference, utilizing expert experience and fuzzy method to make

dynamic compensation for set point of each zone. [29] The structure is shown in Fig

Page 39: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

34

19.

Figure 20 The control structure based on feedback

Assuming EC is section temperature difference, surface temperature difference is

ES. The maximum allowable values of EC and ES are ECT and EST respectively,

which are about 25°C. If EC and ES values are not satisfactory, expert experience is

applied to compensate furnace temperature correction settings.

Based on practical experience and simulation results, the compensation strategy

is formed as follows. Δ 𝑓 1, Δ 𝑓 2, Δ 𝑓 3 are the temperature compensation for

heating zone I, II and soaking zone respectively.[25]

1. If EC − ECT ≤ 0, ES < −EST, then

Δ 𝑓 1 =|𝐸𝐶 − 𝐸𝐶 |

0.3, Δ 𝑓 2 =

|𝐸𝐶 − 𝐸𝐶 |

1.2+ |𝐸𝑆|,

Δ 𝑓 3 = |𝐸𝐶 − 𝐸𝐶 | +|𝐸𝑆|

2,

2. If EC − ECT ≤ 0, ES > EST, then

Δ 𝑓 1 = 0, Δ 𝑓 2 = 1.25(−ES + 3(EC − ECT)),

Δ 𝑓 3 = −𝐸𝑆 − (𝐸𝐶 − 𝐸𝐶 )

3. If EC − ECT ≤ 0, −EST < ES < EST, then

Δ 𝑓 1 = 0, Δ 𝑓 2 = 0, Δ 𝑓 3 = 0,

Page 40: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

35

4. If 0 < EC − ECT ≤ 10, ES < −EST, then

Δ 𝑓 1 = 0, Δ 𝑓 2 = 𝐸𝐶 − 𝐸𝐶 , Δ 𝑓 3 = |𝐸𝑆|,

5. If 0 < EC − ECT ≤ 10, ES > EST, then

Δ 𝑓 1 = 2|𝐸𝐶 − 𝐸𝐶 |, Δ 𝑓 2 = −(𝐸𝐶 − 𝐸𝐶 ), Δ 𝑓 3 = −𝐸𝑆,

6. If 0 < EC − ECT ≤ 10,−EST < ES < EST, then

Δ 𝑓 1 = 𝐸𝐶 − 𝐸𝐶 , Δ 𝑓 2 = 0, Δ 𝑓 3 = 0,

7. If EC − ECT > 10, ES < −EST, then

Δ 𝑓 1 = 𝐸𝐶 − 𝐸𝐶 , Δ 𝑓 2 = 2(𝐸𝐶 − 𝐸𝐶 ), Δ 𝑓 3 = 0,

8. If EC − ECT > 10, ES > EST, then

Δ 𝑓 1 = 2(𝐸𝐶 − 𝐸𝐶 ), Δ 𝑓 2 = 1.25(−2𝐸𝑆 + 2(𝐸𝐶 − 𝐸𝐶 )),

Δ 𝑓 3 = −𝐸𝑆 − (𝐸𝐶 − 𝐸𝐶 ),

9. If EC − ECT > 10,−EST < ES < EST, then

Δ 𝑓 1 = 2(𝐸C − ECT), Δ 𝑓 2 = 0, Δ 𝑓 3 = 0.

As slab heating process is irreversible, the previous compensation based on the

expert experience for slab heating is mainly according to the information to guide the

subsequent slab heating. However, because of a variety of disturbances, for a single

slab heating process, if the slab temperature distribution is far away from the ideal

heating curve, it will result in the irreparable impact, especially large deviations

making the slab into the waste directly.

Based on the rules above, the feedback compensation can be transferred into feed

forward compensation. Because of the slab temperature tracking model, the slab

temperature and corresponding error at the exit of each zone can be calculated.

According to this error, compensate the set value of each zone for the furnace

temperature. The structure is shown in figure 20.

Page 41: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

36

Figure 21 The structure of feed forward control

When the error is greater than a given ΔTmax, the compensate value should

assigned to the several following zones, it is necessary to add a coordinating upper

layer for feed forward compensate to correct the heating curve of the ideal condition

based on the actual furnace so that target value of each zone can be determined. Feed

forward compensation is mainly to calculate ΔTfsi corresponding to ES based on a

series of rules from experience which is listed before, or determine a local

optimization using objective function is as follows:

min 𝐽𝑖 = 𝑚𝑖𝑛 {1

2𝑃 ∗ [ (𝑡𝑓𝑖) −

∗ (𝑡𝑓𝑖)]2 +

1

2𝑅 ∫ 𝑓𝑖(𝑡)

2𝑡𝑓𝑖+1

𝑡=𝑡𝑓𝑖𝑑𝑡} (11)

To simplify the procedure, the experience rule is taken for the feed forward

compensation.

3.3 Simulation

The model and structure are the same as what is in chapter 2. Add the

compensator, as figure 21 and 22 shows.

Page 42: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

37

Figure 22 Structure of whole system

Page 43: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

38

Figure 23 Feed forward compensator

3.3.1 Feed forward compensation model

According to compensation strategy opposed in section 3.2, feed forward control

model is obtained. Each case is shown below.

Figure 24 Case 1.2.3

Using if and if action model can be achieved easily. Figure 24 shows the

structure of case 1, 2, and 3 which judge the range of ES first and subsystem case 1,

case 2 and case 3 decide by the range of EC-ECT. When case 3 take place, there is no

change in temperature of three zones. Hence the action constant is zero. Case 1 and

case 2 is shown blow.

Page 44: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

39

Fig 24-1 Case 1

Fig 24-2 Case 2

Subsystem case 4, 5, 6 and case 7, 8, 9 are shown in figure 25 and 26. The theory

is the same as case 1, 2, 3. For space reason, there is no need to discuss.

Figure 25 Case 4.5.6

Page 45: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

40

Fig 25-1 Case 4

Fig 25-2 Case 5

Fig 25-3 Case 6

Figure 26 Case 7.8.9

Page 46: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

41

Fig 26-1 Case 7

Fig 26-2 Case 8

Fig 26-3 Case 9

The subsystem in case 1 to 9 is the same as the zone choosing model proposed in

figure 17. EC and ES equal 25 and 30 respectively. Adjust the gain of EC-ECT and

ES to make the slab temperature difference as less as possible.

3.3.2 Result

The result of slab temperature with feed forward compensation is shown in

figure 27.

Page 47: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

42

Figure 27 Feed forward curve

Compared to figure 19, the error of slab temperature of both PID feedback

control and expert experience feed forward control is shown in figure 28. As

mentioned in chapter 2, when the temperature of slab is in plasticity state after

reaching 600°C which corresponds to approximately 3000 second, the requirement for

slab temperature error is lower than when unloading. It is clear that the performance

with feed forward compensation is much better than that with PID feedback control.

The error of slab temperature with feed forward control is almost zero when

unloading. It is easily to be proved that expert experience feed forward compensation

is better than PID feedback control for system with large time delay and various

disturbances.

0 1000 2000 3000 4000 5000 60000

200

400

600

800

1000

1200

1400

time( s)

Tem

pera

ture

(°C

)

slab with disturbance

furnace temperature

Ideal T Slab

cross section

T Slab with disturbance

Page 48: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

43

Figure 28 Comparison of FB and FF

3.4 Validation

The actual slab temperature when unloading is collected by Shengli Oilfield

Highland Company. Using the set point of the actual process and put into the

simulation model, the result of simulated slab temperature when unloading is shown

in figure 29 comparing to the actual slab temperature. Where the solid line is the

actual slab temperature, the dash line represents the simulated slab temperature. The

maximum error is 7.3 °C while the mean error is 2.1°C. From the error between the

actual and simulation, it is easily seen that the model is approximately close to the

actual process.

0 1000 2000 3000 4000 5000 60000

20

40

60

80

100

120

140

time( s)

Tem

pera

ture

(°C

)

slab temperature error

PID feedback

feedforward

Page 49: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

44

Figure 29 Simulation result and Actual temperature

3.5 Conclusion

During the actual heating slab process, disturbances such as rolling rhythm

mutations, the furnace pressure and others will eventually result in the slab heating

temperature distribution curve deviates from the ideal, and there is a big time delay

using feedback compensation based on large system leading to weakness in reflect

changes in working conditions on time. In view of this, a method based on feed

forward control dynamic compensation for furnace temperature set value is propose.

Each furnace zone of the furnace is regarded as a subsystem. Considering the

interaction between the various subsystems, dynamic compensation is applied for

furnace temperature according to the slab temperature deviation at the exit of former

zone. The simulation results show that when the disturbance happens, the method of

the feed forward compensation can compensate on time to reduce the number of

failed slab, while the accuracy is improved to some extent compared to feedback

control.

0 10 20 30 40 50 60 70 80 90 1001065

1070

1075

1080

1085

1090

1095

1100Slab temperature

Time

Tem

pera

ture

Predict temperature

Actual temperature

Page 50: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

45

Chapter 4- Summary

Scholars from various countries have done many researches on furnace to reduce

energy cost since energy crisis in the 1970s. Along with the modern mill toward large,

continuous, high-speed, high precision and multi-species direction, traditional furnace

based combustion control cannot meet the above requirements, which requires

furnace model and optimization of control being in higher quality. Rolling process

requires to be provided suitable temperature slab, while traditional combustion control

or temperature control is unable to complete this process task. Therefore, the

advanced modeling and control techniques should be used for the actual furnace

temperature and the cross section temperature controlling.

However, the furnace is a typical complex industrial control object, including

thermodynamics, chemical and physical processes of all kinds, with a multi-variable,

time-varying, nonlinear, strongly coupled, distributed parameter, large inertia and time

delay and other characteristics, hence production targets cannot be achieved with the

conventional control methods. In particular, the slab temperature distribution in the

furnace cannot be detected directly and continuously measured; it is to achieve this

purpose to control the furnace. This paper is mainly around reheating furnace

temperature control optimization settings and control strategies studies, a feed

forward compensation based on the reheating furnace setting is proposed.

For optimal control of furnace feedback compensation loop with a time delay in

reflecting the real working conditions, this paper proposes dynamic optimization

control strategy of the reheating furnace set value based on feed forward

compensation. Each furnace section of the furnace is regarded as a series of

subsystems considering the interaction between the various subsystems, and then

decoupled these subsystems.

Page 51: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

46

Although this paper proposed the mathematical model of the furnace and the

furnace temperature dynamic optimization settings for a certain research, the

following areas for further work will be done due to the complexity of the heating

process:

1. Establish a precise mathematical model of furnace is a very difficult task, the

current models are constructed based on certain assumptions and need to be

further improved.

2. Study fewer computation and online real-time optimization for furnace

temperature setting method for the actual production to meet the

requirements.

3. Make the furnace optimization setting control technology into actual

products, and further extended to more industrial processes.

Page 52: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

47

Reference

[1] F.Hollander and S.PA.Zuurbier (1982), Design, development and performance of

on-line computer control in a 3-zone reheating furnace, J. Iron and Steel Engineer,

59(1), 44-52.

[2] Y.Wakamiya, et al.(1986), Computer control system for reheating furnace, J.

Process Control in the steel Industry, 455-460.

[3] Misaka, J., Takahashi, R. (1982), Computer control of a reheat furnace at Kashima

steel work's hot strip mill, J. Iron & Steel Engineer, 51-55.

[4] Pike, H. E., Citron, S. J. (1970), Optimization study of a slab reheating furnace.

Automatic, 6(l), 41-50

[5] Wick, H. J. (1981), Estimation of ingot temperature in a soaking pit using an

extented Kalman filter. J. Preprints for IFAC 8th Triennial World Congress, 94-99.

[6] A. Kusters, van Ditzhuijzen, G.A.J.M.(1994), MIMO System identification of a

slab reheating furnace, Proceedings of the Third IEEE Conference on Control

Applications, 1557-1563

[7] Yoshitani, N., Ueyama, T., Usui, M. (1994), Optimal slab heating control with

temperature trajectory optimization, J. 20th International Conference on Industrial

Electronics, Control and Instrumentation, 1567-1572.

[8] Yangyong Yao, Yongzai Lv, (1987), Slab reheating computer-controlled dynamic

mathematical model development. J. ACTA AUTOMATICA SINICA, 13 (4), 257-264

[9] Yangyong Yao, Yongzai Lv, (1987), Furnace dynamic optimization control

strategy development, J.Information and Control, 16 (5), 1-5

Page 53: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

48

[10] Yangyong Yao, Yongzai Lv, (1989), Walking-beam furnace dynamic control

model development, J.Control and Decision, 1988, 3 (3): 51-53.

[11] Dirk, S., Arend, K.(1996) , Online slab temperature calculation and control, J.

International Symposium on Information Storage and Processing Systems American

Society of Mechanical Engineers, Manufacturing Engineering Division, 307-314.

[12] Ashok D. Acharya, Sirshendu Chattopadhyay, (1998), Reheat furnace

temperature control and performance at Essar steel, J. Iron & Steel Engineer, 75(12),

31-36.

[13] Lu Yongzai, (1996), Meeting the challenge of intelligent system technologies in

the iron and steel industry, J. Iron & Steel Engineer, 73(9), 139-149.

[14] Carpenter, D.G., Proctor, C.W. (1987), Temperature control and optimization of

a reheat furnace using a distributed control system. J. Iron & Steel Engineer, 64 (8),

44-49.

[15] Hofscher, R.A., Harding, J. M. (1987), Distributed process control system for

bloom reheating furnaces, J. Iron & steel Engineer, 64 (3),23-24.

[16] Leden B. (1986), A control system for fuel optimization of reheat furnaces. J.

Metallurgy, 16-24

[17] Zhongjie Wang and Qidi Wu, (2004), Optimal-setting control for complicated

industrial processes and its application study, J. Control Engineering Practice, 65-74

[18] Robert J.Schurko, et al (1987), Computer-control of reheat furnaces: A

comparison of strategies and application, J. Iron and Steel Engineer, 37-42.

[19] Timothy A.Veslocki,et al (1986), Automatic slab heating control at Inland's

80-in hot strip mill,J. Iron and Steel Engineer, 47-54.

[20] Staalman, Rirk F J, Kusters Arend.(1996) , On-line slab temperature calculation

Page 54: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

49

and control, 30-314.

[21] John H. Mathews and Kurtis K. Fink (2004), Numerical Methods Using Matlab,

4th

Edition, ISBN: 0-13-065248-2

[22]. K. S. Chapman, S. Ramadhyani, and R Viskanta, Modeling and parametric

studies of Heat transfer in a Direct-fired continuous Furnace, J. Metallurgical

Transaction, 513-521.

[23] Yoshitani, N., Naganuma, Y., Yanai, T., Optimal slab heating control for

reheating furnaces, J. Proc. of the ACC, 3030-3035.

[24] H.Pierrard, et al (1986).Automation of SOLLAC`S new slab reheat furnace, J.

Process Control in the Steel Industry, 307-325.

[25] Bai M. Q. & Gao F. Q. (1994), Analysis of the dynamic characteristic of furnace

temperature, J. Metallurgical Automation, 18(4), 8-10.

Page 55: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

50

Appendix

1. Set point of furnace temperature.

function f=objfun(n)

x=0;

T_heat1=1100 ;

T_heat2=1150 ;

T_soak=1130

f(1)=640

for i=1:3000

f(i+1)=f(i)+0.46-2.1*10^-4*x;

if x<1359.8

f(i+1)=f(i)+0.46-2.1*10^-4*x;

else if x>=1359.8&x<=1859.8 %%heating 1

if f(i+1)<=T_heat1-5

f(i+1)=T_heat1-5;

else if f(i+1)>=T_heat1+5

f(i+1)=T_heat1+5;

end

end

else if x>=1859.8&x<2489.8 %%heating 2

if f(i+1)<=T_heat2-5

f(i+1)=T_heat2-5;

else if f(i+1)>T_heat2+5

f(i+1)=T_heat2+5;

end

end

else %% Soaking

if f(i+1)>T_soak+5

f(i+1)=T_soak+5 ;

else if f(i+1)<T_soak-5

f(i+1)=T_soak-5 ;

end

end

end

end

end

x=x+1;

end

2. Validation

%% Clear up the environment variable

clc

clear

Page 56: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

51

%% Set network parameters

load slab_temperature input output input_test output_test

M=size(input,2); %the number of input nodes

N=size(output,4); %the number of output nodes

nhid=8; % Number of invisible nodes

lp1=0.01; %Learning Probability

lp2=0.001; %Learning Probability

nit=100; %Iteration

%Initialize the weights

Wjk=randn(nhid,M);Wjk_1=Wjk;Wjk_2=Wjk_1;

Wij=randn(N,nhid);Wij_1=Wij;Wij_2=Wij_1;

a=randn(1,nhid);a_1=a;a_2=a_1;

b=randn(1,nhid);b_1=b;b_2=b_1;

%Initialize the node

y=zeros(1,N);

net=zeros(1,nhid);

net_ab=zeros(1,nhid);

%Initialize incremental weight learning

d_Wjk=zeros(nhid,M);

d_Wij=zeros(N,nhid);

d_a=zeros(1,nhid);

d_b=zeros(1,nhid);

%% Normalize input and output data

[inputn,inputps]=mapminmax(input');

[outputn,outputps]=mapminmax(output');

inputn=inputn';

outputn=outputn';

%% Network prediction

for i=1:nit

%Error accumulation

error(i)=0;

% Training

for kk=1:size(input,1)

x=inputn(kk,:);

yqw=outputn(kk,:);

Page 57: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

52

for j=1:nhid

for k=1:M

net(j)=net(j)+Wjk(j,k)*x(k);

net_ab(j)=(net(j)-b(j))/a(j);

end

temp=mymorlet(net_ab(j));

for k=1:N

y=y+Wij(k,j)*temp;

end

end

%Calculate the sum of errors

error(i)=error(i)+sum(abs(yqw-y));

%Weight adjustment

for j=1:nhid

%Calculate d_Wij

temp=mymorlet(net_ab(j));

for k=1:N

d_Wij(k,j)=d_Wij(k,j)-(yqw(k)-y(k))*temp;

end

%Calculate d_Wjk

temp=d_mymorlet(net_ab(j));

for k=1:M

for l=1:N

d_Wjk(j,k)=d_Wjk(j,k)+(yqw(l)-y(l))*Wij(l,j) ;

end

d_Wjk(j,k)=-d_Wjk(j,k)*temp*x(k)/a(j);

end

%Calculate d_b

for k=1:N

d_b(j)=d_b(j)+(yqw(k)-y(k))*Wij(k,j);

end

d_b(j)=d_b(j)*temp/a(j);

%Calculate d_a

for k=1:N

d_a(j)=d_a(j)+(yqw(k)-y(k))*Wij(k,j);

end

d_a(j)=d_a(j)*temp*((net(j)-b(j))/b(j))/a(j);

end

%Update weight parameter

Wij=Wij-lp1*d_Wij;

Page 58: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

53

Wjk=Wjk-lp1*d_Wjk;

b=b-lp2*d_b;

a=a-lp2*d_a;

d_Wjk=zeros(nhid,M);

d_Wij=zeros(N,nhid);

d_a=zeros(1,nhid);

d_b=zeros(1,nhid);

y=zeros(1,N);

net=zeros(1,nhid);

net_ab=zeros(1,nhid);

Wjk_1=Wjk;Wjk_2=Wjk_1;

Wij_1=Wij;Wij_2=Wij_1;

a_1=a;a_2=a_1;

b_1=b;b_2=b_1;

end

end

%% Network prediction

%Normalize predicted input

x=mapminmax('apply',input_test',inputps);

x=x';

%Network prediction

for i=1:100

x_test=x(i,:);

for j=1:1:nhid

for k=1:1:M

net(j)=net(j)+Wjk(j,k)*x_test(k);

net_ab(j)=(net(j)-b(j))/a(j);

end

temp=mymorlet(net_ab(j));

for k=1:N

y(k)=y(k)+Wij(k,j)*temp ;

end

end

yuce(i)=y(k);

y=zeros(1,N);

net=zeros(1,nhid);

net_ab=zeros(1,nhid);

Page 59: EXECUTIVE SUMMARY -  · PDF fileEXECUTIVE SUMMARY INTRODUCTION ... Both problems are linked to the reheating furnace. ... Reheating furnace model, Slab temperature model, PID,

54

end

%Anti-normalization of predicted output

ynn=mapminmax('reverse',yuce,outputps);

%%

figure(1)

plot(ynn,'-.')

hold on

plot(output_test,'r')

title('Slab temperature')

legend('Predict temperature','Actual temperature')

xlabel('Time')

ylabel('Temperature')