See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/46435821

A System Dynamics Simulation Model of a Blast Furnace

Article  in  International Journal of Business and Systems Research · January 1999

DOI: 10.1504/IJBSR.2017.10006188 · Source: RePEc

CITATIONS READS
0 760

2 authors:

Medha Ashtekar Goutam Dutta

Indian Institute of Management Indian Institute of Management, Ahmeadbad
10 PUBLICATIONS   112 CITATIONS    81 PUBLICATIONS   697 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Dynamic Pricing Electricity View project

Literature Review on Dynamic Pricing of Electricity View project

All content following this page was uploaded by Goutam Dutta on 24 November 2016.

The user has requested enhancement of the downloaded file.

Int. J. Business and Systems Research, Vol. X, No. Y, xxxx 1

A system dynamics simulation model of a blast

furnace for project evaluation

Goutam Dutta*
Production and Quantitative Methods Area,
Indian Institute of Management,
Ahmedabad-380015, India
Fax: 91-79-26306896
Email: goutam@iimahd.ernet.in
*Corresponding author

Medha Ashtekar
Emerging Capital Group,
30 Cecil Street,
#19-08 Prudential Tower, 049712, Singapore
Email: ash_medha@yahoo.com
Abstract: The manufacturing of hot metal in a blast furnace is one of the
important parts of an integrated steel plant. In a blast furnace, the quality and
productivity of the output (hot metal) depends on a large number of input and
operating variables. In this paper, we simulate complex interactions of
technological and financial variables. We chose system dynamics principles to
simulate the blast furnace for project evaluation as a useful aid to managerial
decisions. The model would be used to answer a number of questions related to
investment decisions. We simulate the actual production of hot metal,
accumulated hot metal production, net works cost and cost of hot metal per ton
for a period of 36 months with real data.
Keywords: simulation; system dynamics; manufacturing; minerals and metals;
blast furnace; project evaluation.
Reference to this paper should be made as follows: Dutta, G. and Ashtekar, M.
(xxxx) ‘A system dynamics simulation model of a blast furnace for project
evaluation’, Int. J. Business and Systems Research, Vol. X, No. Y, pp.xxx–xxx.
Biographical notes: Goutam Dutta is a Professor of Production and
Quantitative Methods Area at the Indian Institute of Management, Ahmedabad.
He holds a PhD from the Northwestern University, USA. He was a Visiting
Faculty at the University of Illinois at Chicago and London School of
Economics and HEC Paris. He is the winner of the Franz Edelman Award
(1994), IFORS OR for Development Prize (1993) and was a finalist in the
EURO Excellence in Practice Award. His papers have been published in
several scholarly journals. He is on the editorial board of, JORS, Interfaces
(INFORMS) and IJRM and International Journal of Business Analytics.
Medha Astekar is one of the Directors of Emerging Capital Group at
Singapore. She has received her Masters in Statistics and MBA from the
University of Singapore. She has wide range of modelling and systems in
Systems Research Institute, Pune, IIM, Ahmedbad and GE Research Center at
Bangalore.

Copyright © 20XX Inderscience Enterprises Ltd.

2 G. Dutta and M. Ashtekar

1 Introduction

This paper is an attempt to develop a system dynamic (SD) simulation mode of a blast
furnace of an integrated steel plant for investment decision. The blast furnace, the heart of
an integrated steel plant, converts raw materials into liquid iron (also called hot metal).
The focus is on the complexity of the blast furnace where the quality, productivity, and
profitability of the output (hot metal) depend on many interacting variables. Hence, it is
difficult for an analyst to predict the dynamics of changes in all the interacting variables
in a blast furnace. The complexities of the technology and cost interdependencies make it
difficult for a manager to predict the technical and financial outcomes, which in turn
affect various investment decisions.
Because of these complexities, a manager has to rely on intuition when an investment
decision is made with respect to a blast furnace. The intuitive solution may not be the
best course of action for a company that is looking for optimal returns on investment. We
chose SDs principles (Forrester, 1964; Roberts, 1978) to simulate the blast furnace for
project evaluation as a useful aid to managerial decision making.
SDs incorporates cause and effect relationships, nonlinearity, and close loop feedback
control. While modelling a blast furnace to take care of uncertainties, we decided to use
SD simulation and not discrete events simulation. In discrete events simulation, we can
measure the variation of different variables by the actual values and can generate the
same variation by simulation. We believe that SD is better suited to model this type of a
relationship than discrete events simulation. An earlier work (Sinha and Dutta, 1985) was
possibly the first reported attempt of SDs modelling of a blast furnace. This paper is an
extension of the same work with nonlinear relationship.
This paper is organised as follows. Section 2 describes the survey of related literature
technology of a blast furnace. It surveys the literature on application of optimisation and
simulation models in an integrated steel plant. Section 3 describes the development of the
SDs with four sub-models. Section 4 discusses the model development, validation and
implementation. Section 5 describes the results of the model. The paper concludes with
directions for future research in the last section.
The paper is aimed at two audiences. The first is SDs researchers who are looking for
an application area in an integrated steel plant. The second is real world modellers of an
integrated steel plant. An elementary knowledge of a blast furnace will help understand
the process. A detailed discussion of blast furnace technology is beyond the scope of the
paper. Interested readers are requested to go through the publication (Lankford et al.,
1985).
Readers not familiar with the manufacturing or the process of iron making are
advised to go through Glossary A and Glossary B before starting Section 3. The first
discusses all the important parts of a blast furnace. The second discusses most of the
technical terms that are used in the influence diagram.

2 Literature survey

The first reported work (Fabian, 1958) of modelling of an integrated steel plant is a linear
programming model that minimises overall cost of production. It is an integration of four
sub-models: iron making sub-model, steel making sub-model, primary mill sub-model,
and finishing mill sub-model. The iron making sub-model is an application of linear
 A system dynamics simulation model of a blast furnace for project evaluation 3

programming in the blast furnace. Another work (Fabian, 1967) also discusses the
application of linear programming in a blast furnace. A later publication (Fukui et al.,
1983) describes a selection system for finding the optimal operating condition of a blast
furnace. Another work (Sweeny and Roppol, 1983) discusses a burden calculator, a quick
computational method of calculating raw materials required for the blast furnace at a
particular operation condition. A study by Oaknock et al. (1983) uses material balance
and energy balance equations to improve the operating conditions of a blast furnace.
Another publication (Perez and Hanikar, 1983; Savas, 1961) describes linear
programming applications to a blast furnace. The application of expert system is
described in a study (Inkala et al., 1995) in a Finnish steel plant, Rautarukki. The
application of artificial intelligence in a blast furnace burden distribution control is
documented in a publication (Iida and Yuichi, 1995). Angstenberger (1996) describes the
application of fuzzy clustering and neural networks to blast furnace modelling and
control in a German steel plant. A technical report (Valenti and Deitz, 1997) describes a
system developed by Voist Alpine Sthal, an Austrian firm for controlling blast furnace
operations. This system is a series of computer models for phases of blast furnace
operation. It includes a burden optimisation model for determining the optimal mix of
inputs.
There are several artificial intelligence application papers that deal with blast furnace
modelling and control. The report (Zuo et al., 1998) outlines the design of models based
on neural networks. This model discusses a neural network for recognition of top gas
distribution, distribution of heat fluxes, and prediction of slips in a blast furnace. Paul and
Chaney (1998) attempted to apply genetic algorithms to the problem of optimising an
existing simulation model One of the recent applications is a work (Lasdon et al., 1998)
that discusses the application of nonlinear programming techniques where the model and
solver act as the controller in an on-line blast furnace control system in the
USA. Comprehensive survey of application of mathematical programming models is
discussed in Dutta and Fourer (2001). Chen (2001) presents a predictive system for blast
furnaces by integrating a neural network with qualitative analysis. Brännbacka et al.
(2005) describes the simple simulation of a blast furnace hearth. Pettersson et al. (2007)
describe in detail the development of genetic algorithms-based multi-objective
optimisation techniques, wherein the training process of a feed forward neural network
was used with noisy data from an industrial iron blast furnace. Hooey et al. (2010)
discusses the spreadsheet-based model of blast furnace. In a recent paper, Giri et al.
(2013) describe the development of a new bi-objective genetic programming (BioGP)
technique was developed. It show that the initially attempts to minimise training error
through a single objective procedure and subsequently switches to bi-objective evolution
to work out a Pareto-trade-off between model complexity and accuracy.
These studies are based on linear programming, discrete event simulations or process
control models and do not discuss application of SDs, in either an integrated steel plant or
the blast furnace.
A publication (Narchal, 1988) deals with the problem of building a SDs simulation
model for helping corporate planning in an integrated steel plant. The management of the
steel company used this model for simulating the impact of strategic policies on corporate
objectives. The model also helped management in designing a long-term investment
policy related to expansion and modernisation and project a smooth flow of production.
4 G. Dutta and M. Ashtekar

We have described most of the models used in the blast furnace. The blast furnace is
always a matter of interest for simulation model builders. Any process improvement and
technology enhancement needs investment of several million dollars and there is no
standard way to find out if the return is worthy of that investment. An earlier study by
Sinha and Dutta (1985) describes SDs techniques to find out the returns on the
investment a steel company made in modernisation in the early eighties. This is the first
use of SDs techniques, where hot metal production is based on material balance
equations. In the current paper, we have considered hot metal production as a nonlinear
function of several technological variables like productivity of the blast furnace, working
volume, and available hours in a blast furnace.

3 SDs model of blast furnace

The blast furnace model developed is based on many feed back loops, some of which are
described in the next section. The model has four sub-systems: production, workforce,
financial, and information. We describe each sub-system in detail.

3.1 Production sub-system

Production of hot metal in a blast furnace is computed from the productivity of the blast
furnace, available hours per blast furnace, and working volume of the blast furnace.
Productivity of the blast furnace is the amount of hot metal produced per day per working
volume of the blast furnace. We discuss some of the fundamental equations used in this
model.
Hot metal produced = Available hours per month * Productivity
(1)
*Working volume

Available hours = Calendar hours − Downtime (2)

Downtime = Major shut down + Minor shut down + Unplanned delays (3)

Calendar hours = No. of shifts * Hours per shift * Number of days / Month (4)

Sinter charged in the blast furnace = Sinter produced in sintering plant

(5)
*Transfer loss of sinter
Coke charged in the blast furnace = Coke produced in coke ovens
(6)
*Transfer loss of coke
Ore charged in the blast furnace = Ore produced in mines
(7)
*Transfer loss of ore
The steel plants we have chosen have a number of blast furnaces and total hot metal
production is the sum of production of all the furnaces. The calendar hour is a function of
the number of shifts per blast furnace, hours per shift, and number of days per month per
blast furnace. The downtime consists of planned delay (preventive shutdown), unplanned
delay (breakdown maintenance), and capital and mid-term repair. Capital repair hours,
 A system dynamics simulation model of a blast furnace for project evaluation 5

mid-term repair hours, and breakdown hours are defined in the form of table functions of
actual values of the parameters. The number of days per month is a table function.
Important inputs of raw materials are sinter, ore, and coke. The blast furnace is the
customer for a sintering plant, coke ovens, and mines. The modelling of material balances
of coke and iron-ore is similar. We describe one of the material balance equations in the
model and its SD model implementation.

3.2 Workforce subsystem

The workforce in the blast furnace consists of three components: operating crew,
maintenance crew, and contract labour. Operating and maintenance crews are regular
employees of the company and contract labourers can be hired as and when required
(mostly in maintenance). Operation and maintenance persons are modelled as a level
variable, whereas turnovers (number of persons leaving) are modelled as a rate variable.
Other factors considered for modelling the workforce subsystem are average working life
of an employee, number of weekly-off and leave of an employee, and absentees in every
time period (generally observed for contract labour).
The life of a blast furnace is about five years after which it undergoes major repair
(known as relining). The period of a major reline varies from 60 to 75 days. In a five-year
life span, a blast furnace can undergo one or two minor repairs lasting 15–30 days. Since
the blast furnace is the heart of a steel plant, any delay (in the form of breakdown) can
affect production. Therefore, the total amount of steel produced in an integrated steel
plant is dependent on the amount of hot metal produced in the blast furnace. At the time
of relining, the maintenance department needs additional workforce. This is an important
part in modelling as the production is stopped during this time and the cost of manpower
is higher than that during normal operations.

3.3 Accounting and financial sub-system

In the financial model, we have considered the price and cost structure. The cost of hot
metal produced is the sum of net metal cost and conversion cost (activity cost). The major
components of material expenses are iron ore, coke, sinter, and fluxes. As the slag and
blast furnace gas are by-products of blast furnace, scrap credit is the sum of slag credit
and gas credit. We add the conversion cost to the net material cost. The conversion cost is
the sum of stores fuel, transfer labour refractory, and GAGW cost (general administrative
and general works). The major components of the transfers are the shops, steam, water,
power, air, and it may be known as general administration and general works (GAGW).
The GAGW is the overhead part of the cost.
The total labour cost has three components: operating labour cost, maintenance labour
cost, and contract labour cost; each of these three components has a normal and an
overtime element. For each of the three components of labour cost, we have considered
different wage rates for different types of labour and their inflation rate. The number of
maintenance personnel is dependent on the downtime for maintenance.
Cost of raw materials will depend on the rate of input material, buying price of input
material, and the inflation rate. For example, coke cost at time (t + n) will be coke cost at
time t, multiplied by the inflation rate for the year n. We now write some of the basic
equations of this sub-system.
6 G. Dutta and M. Ashtekar

Cost of hot metal = Net metals cost + Activity cost (8)

Net metals cost = Gross metal cost − Scrap credit (9)

Scrap credit = Slag credit + Gas credit (10)

Gross metal cost = Sum of all raw materials cost (11)

Sum of all raw materials cost = Iron ore cost + Sinter cost + Coke cost
(12)
+ Pellet cost + Dolomite cost + Limestone cost
Coke cost at time (t + n) = Coke cost at time t (1 + inflation rate of coke) n (13)

Activity cost = Stores cost + Fuel cost + Labour cost

(14)
+ Refractory cost + Transfer cost + GAGW cost

Labour cost = Operating labour cost + Maintenance labour cost

(15)
+Contract labour cost
Operating labour cost = Normal operating labour cost
(16)
+Overtime operating labour cost

Maintenance labour cost = Normal maintenance labour cost

(17)
+ Overtime maintenance labour cost
Contract labour cost = Normal contract labour
(18)
+Overtime contract labour cost

3.4 Management of information sub-systems

This sub-system computes various actual and possible rates like required hot metal
production, maximum hot metal production, possible hot metal production, actual
production losses owing to breakdown, mid-term repair and capital repair, maximum
capacity in hours, available capacity in hours, utilised capacity, surplus man-hours, and
shortage man-hours. Comparison of actual and possible rates can aid in policy (is
something missing here?).

4 Model development and computer implementation

In this section, we describe the methodologies used for data collection, calibration, model
building, and validation.

4.1 Data collection, calibration and validation

We collected data for this model from different sources. The primary sources of data
were different cost sheets and manufacturing details of the company. The cost figures
were computed from the monthly cost data, which indicated normal cost, variable cost,
and the variable portion of the cost above. In addition, we also looked into works
(factory) performance statistics, which indicates different technological coefficients and
 A system dynamics simulation model of a blast furnace for project evaluation 7

productivity and other details of the company. We also had the profitability statement, net
realisation report, yearly cost sheet, and annual report of the company as ready
references. The model building exercise was based on the following:
These types of model development take time and effort and it is important that we
model the experience. To capture the experience of the metallurgists, a meeting of
metallurgists, blast furnace operators, and model builders was conducted and this led to
the development of SDs influence diagram. While variables in the influence diagram. We
also knew that there are established relationships between the different metallurgical
principles and care was taken that these principles are not violated.

4.2 Influence diagram construction

We describe some of the influencing relationships in this diagram. We refer to Figure 10.
We once again request the reader to go through the glossaries.
1 With the increase in hot metal production, there is an increase in the maintenance
work, which in turn increases delay. With increases in delay there is increase in
downtime, which in turn decreases available hours. With decreases in available
hours, possible production decreases, which results in decrease in hot metal
production. This is shown in Figure 1 (A negative feedback loop).

Figure 1 Hot metal production and maintenance work

_ +
+
DOWN DELAYS
AVAILABLE
TIME
HOURS

_
MAINTENANCE
WORK
POSSIBLE
+
PRODUCTION
HOT +
METAL
PRODUCTION
+

2 When hot metal production increases, number of hours required for capital repair
increases, this in turn cause downtime to increase and available hour to decrease.
With a decrease in the available hours, possible production decreases which in turn
results in a decrease in the hot metal production. This is a negative feedback loop as
displayed in Figure 2.
(A negative feedback loop).
8 G. Dutta and M. Ashtekar

Figure 2 Hot metal production and capital repairs

_
DOWN +
TIME

AVAILABLE
HOURS

CAPITAL
REPAIRS

+ HOT
POSSIBLE METAL +
PRODUCTION PRODUCTION

3 When hot metal production increases, number of hours required for mid-term repair
decreases, downtime increases and available hours decrease. With a decrease in
available hours, possible production decreases which in turn results in a decrease in
hot metal production (Figure 3, a negative feedback loop).

Figure 3 Hot metal production and mid term repair

_
DOWN +
TIME
AVAILABLE
HOURS

MIDTERM
+ REPAIR
POSSIBLE
HOT
PRODUCTION METAL
PRODUCTION
+
+
 A system dynamics simulation model of a blast furnace for project evaluation 9

4 With an increase in coke rate, more of flux (limestone and dolomite) is required.
Therefore, with an increase in coke rate, flux rate increases. With an increase in flux
rate, more slag is generated and slag rate increases (Figure 4, a positive feedback
loop).

Figure 4 Coke rate with flux and slag rate

FLUX +
RATE

+
+
COKE
SLAG
RATE
RATE

5 With an increase in hot metal production, the sinter charged increases. With an
increase in sinter charged fraction sinter in the burden (FSRBDN) increases. With an
increase in the fraction sinter in the burden, coke rate decreases, and this decreases
coke charged. This in effect will reduce coke inventory and reduce hot metal
production (Figure 5, a positive feedback loop).

Figure 5 Hot metal production coke rate and sinter charged

+ _
COKE FRACTION
RATE SINTER IN
COKE BURDEN
CHARGED +

_
_
COKE
HOT
INV METAL SINTER
PRODUCTION CHARGED

_
+

6 With an increase in wind rate, the carbon burning rate increases, which increases the
possible wind rate (Figure 6, a positive feedback loop).
10 G. Dutta and M. Ashtekar

Figure 6 Wind rate and carbon burning rate

CARBON
BURNING
RATE +

+
WIND
POSSIBLE +
WIND RATE
RATE

We explain one of the SDs equation used in this model (Figure 7).

Figure 7 Hot metal flow diagram

HM
DEMAND

HOT METAL
INVENTORY

HOT HOT
METAL METAL
INFLOW OUTFLOW
RATE RATE
PRODUC-
TiVITY
NO OF
DAYS

HM
PRODN.
Figure 7 Hot Metal Flow Diagram

• Level equation: Hot metal inventory in time K = Hot metal in time J + Delta time
(Hot metal inflow rate – Hot metal out flow rate)
L HM.K = HM.J + (DT)(HMI.JK − HMO.JK) HM INVENTORY (19)

• Rate equation: Hot metal inflow rate = Production of hot metal

R HMI.KL = PRDHM.KL HM INFLOW (20)

• Rate equation: Production of hot metal = (Productivity of blast furnace) * (Volume

of blast furnace) * (Number of days)
R PRDHM.KL = (PROD.KL) * (VOLBF) * (NOD.KL) HM PRODUCTION (21)
 A system dynamics simulation model of a blast furnace for project evaluation 11

• Rate equation: Hot metal outflow rate = Demand of hot metal.

R HMO.KL = DMHM.K DEMAND OF HOT METAL (22)

Similarly, coke inventory has been explained in Figure 8.

Figure 8 Flow diagram of coke

COKE
RATE
COKE
DEMAND

COKE
INVENTORY

COKE COKE
INFLOW OUTFLOW
RATE RATE

COKE
PRODN.
Figure 8 Flow Diagram of Coke

In addition to the above relationships there are a number of constants. For example, there
are transfer losses, like the transfer loss of sinter, coke and ore. The percentage of oxygen
in the hot blast is also a constant. We have the following relationships.
1 with increase in the transfer loss of sinter, sinter charged to the blast furnace
decreases
2 with increase in the transfer loss of coke, coke charged to the blast furnace decreases
3 with increase in the transfer loss of iron-ore, iron-ore charged to the blast furnace
decreases.

4.3 Model validation and implementation

After quantification of the individual relations based on historical data and perceptions of
blast furnace operators and metallurgists, a computerised model has been developed with
over 600 dynamo equations. The model is based on five sub-models: production,
workforce, cost, MIS, and data. The model consists of 39 level variables, 210 rate and
auxiliary variables, and 33 parameters. After all these interdependencies have been built
into DYNAMO equations, the model was validated with three-year data for the blast
furnaces of a major steel-making plant in Southeast Asia. We have considered 37 points
of data. We used PC-DYNAMO software (developed by Pugh Roberts Associates). After
the output file was obtained in PC/DYNAMO, we transformed the data file to an Excel
spreadsheet and drew the graphs.
For validating the model, the trend and pattern of modelled variables were studied
against actual figures. We took the following steps for validating the model:
12 G. Dutta and M. Ashtekar

1 Dimensional consistency
We checked the dimensions and accuracy of the data. We made sure that all the
rates, levels and auxiliary variables and the table functions were dimensionally
correct. For example, we checked that the material balance equations were in tons.
Capacities of all machines were in hours. Similarly, we checked that the coke rate,
slag rate, sinter rates were within their normally accepted values. For example, the
coke rate values should be between 300 and 700 tons of coke/ton of hot metal.
2 Contact points
To check that the model represents reality, a number of contact points have been
identified. These are the measurable quantities in the real world and at the same time
they are variables in the model.
a monthly production of hot metal
b accumulated production of hot metal (accumulated over a period)
c net works cost
d works cost per ton.

5 Results

In this section, we discuss the results of the SD simulation. We concentrate on the four
contact points. The monthly production is computed from the equations (1) to (11)
discussed in Section 3.1. Similarly, the actual hot metal production data is available
(unclear). The net works cost is also computed from the set of equations described in
equations (8) to (18). We have multiplied all the real and simulated values by a factor to
maintain confidentiality of the cost and financial figures.
In Table 1, the average value of the actual data and the average value of the
simulation model are given. In addition, we show the mean absolute deviations (MAD),
root mean square errors (RMSE), and mean absolute percentage errors (MAPE). The
percentage errors for the variables vary from 11.84% to 16.37%.
Table 1 Model vs. real values of cost and production

Works cost Works cost/ton Production Accumulated production

Unit Rs million Rs/ton (Million tons) Million tons
Real average 110.710 759.960 0.14094 0.899
Model average 100.705 691.310 0.15668 1.014
MAD 13.400 96.0000 0.0220 0.141
RMSE 17.500 1.980 0.02640 0.171
MAPE 11.480 12.340 16.370 15.050

We have conducted various statistical tests to test the efficiency of the simulation. First,
we show the R2 of actual and simulated values. This is shown in Table 2 along with t
statistics and F statistics. The t statistics and F statistics are computed for actual and
modelled values. We assumed the null hypothesis that there was no difference between
 A system dynamics simulation model of a blast furnace for project evaluation 13

actual and predicted values. The alternative hypothesis assumed that there was a
difference between actual and predicted values.
Table 2 Statistical test results model vs. real

Variables R2 t-statistics F statistics

Production 0.843 0.0066 0.155
Acc. production 0.993 0.002 0.656
Works cost 0.853 0.279 0.426
Works cost per ton 0.993 0.254 0.668
t-critical 2.720
F-critical 1.550

Figure 9 indicates yearly simulated cumulative production of hot metal and actual
production.

Figure 9 Model vs. actual cumulative production (see online version for colours)
2500000

2000000
CUMULATIVE PRODN (REAL/MODEL)

1500000

model
REAL

1000000

500000

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

TIME FIGURE 9

We find that simulated yearly cumulative production follows the same trend as actual
yearly cumulative production. We also find that the difference between the two decreases
every year. This indicates that, on an overall basis, simulated cumulative production of
hot metal is a good replication of actual production. It may be possible, after the initial
noises in the system are taken care of, that simulated production (model) comes close to
actual production.
We now concentrate on actual monthly production over a period of 36 months. In
Figure 10, we find that out of 36 points, 22 crests and troughs match with each other for
simulated and actual monthly production. In the first 12 points of data, the difference
between actual and simulated monthly production is significant. We find that model
values are higher than real values. However, this difference gets reduced over the next
two years. We also find that simulated production is much higher than actual production.
There are two possible reasons for this. The model values are based on the working
14 G. Dutta and M. Ashtekar

volume. The actual working volume may be less than that assumed in the model and this
difference is shown in the graph. We now concentrate on the actual and model
production. As the first (zeroth) observation is one The 13, 25, and 37represent months of
March (last month of the financial year). In March, there are 31days, whereas in
February, there are 28/29 days. Therefore, available hours are much higher in March.
Moreover, there is pressure on the operating manager of the blast furnace to over-perform
and break all previous records. Therefore we find that the production in March is higher
than that in February. Similarly, we also find that the average production in the first six
months (April to September) is lower than the average production in the last six months.
In the months of 26 and 27 (May and June of year 3), there is a decline from that in 25.
Some minor repairs were scheduled in those months and the blast furnace did not perform
to its full capacity.

Figure 10 Production of hot metal: model vs. actual (see online version for colours)
200000

180000

160000

140000

120000
PRODN

MODEL
100000
REAL

80000

60000

40000

20000

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
TIME FIGURE 10

In Figure 11, we show the simulated and actual values of net works-cost. In this figure,
the modelled values are a close replication of real values. We find that out of 36,
20 crests and troughs match while two do not match. The net hot metal cost is a function
of different raw material costs, activity cost and inflation figures. The cost figure
increases from Rs 80 million to 140 million. We also find that the average cost for the
first six months in a year is lower than the average cost in the next six months. The
sudden increase from month 11 to month 12 (and also from month 23 to month 24) is
because of increase in maintenance cost at the time of relining of one of the blast
furnaces. We also find that the simulated net works cost is lower than the actual net
works cost.
In Figure 12, we show the simulated (model) and real values of works cost per ton of
hot metal. Owing to inflationary pressures on the prices of raw materials and increasing
labour cost over the 36-month period, the works cost per ton of hot metal has increased
from Rs. 600/ton to Rs. 900/ton. We find that there are increases in the months of 12 and
24 (from their previous months) owing to increase in the maintenance cost of the blast
furnace (scheduled in this month). The simulated value also shows the same trend.
 A system dynamics simulation model of a blast furnace for project evaluation 15

However, the crests and troughs in this figure do not match.

Figure 11 Net hot metal cost: model vs. real (see online version for colours)
1.80E+08

1.60E+08

1.40E+08

1.20E+08
Net Cost at BF

1.00E+08 REAL
MODEL

8.00E+07

6.00E+07

4.00E+07

2.00E+07

0.00E+00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

TIME FIGURE 11

Figure 12 Works cost per ton: model vs. real (see online version for colours)
1200

1000

800
Works cost/ton at BF

REAL
600 . MODEL

400

200

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
TIME FIGURE 12

6 Extension and conclusions

In this paper, we have discussed a mathematical tool for simulating the operation of a
blast furnace for project evaluation. The model developed in this paper is generic and can
be used by other blast furnaces. It can be used to answer the following questions:
16 G. Dutta and M. Ashtekar

1 If we import coal from Australia, coke produced from Australian coal will have more
productivity. However, this increases the cost of coal and hence increases the cost of
the hot metal produced. Once we know the cost of imported coal, we can run the
model in two different scenarios and see the effect on the cost of production.
2 If a coal injection facility is installed with an investment of X million dollars, will
the productivity of the blast furnace improve? If this improvement results in a
reduction of the variable cost of production of hot metal, is the investment justified?
3 If imported coal is used as a raw material in the blast furnace, the quantity of coke
required per ton of hot metal is reduced. Imported coal contains less percentage of
ash than domestic coal. However, imported coal is more expensive than domestic
coal. Should the company be substituting domestic coal for imported coal?
4 What will be the effect on the cost per ton of hot metal, when productivity of the
blast furnace is improved by 20%?
5 The technology of iron-making is undergoing a change. Voest Alpine Industrieanbau
(VAI) has developed a new technology called COREX technology, Austria which
completely eliminates the need for coke in iron-making. In this process, coal can be
used in place of coke. This is a major advantage for two reasons. First, cooking coal
is in short supply in India and the reserves are limited. Second, the elimination of
coke ovens reduces the fixed and variable costs of production of iron. This
technology has been commissioned in four places. With the first SD model of a blast
furnace ready, a natural extension of this work would be making a SD model for the
COREX method of iron making.
6 One of the productivity improvement projects in the blast furnace is to inject coal
along with the air (hot blast entering). This may require an investment of X million
dollars.
7 The model can be run with two different productivity figures. Based on this, we can
find out how much will be the reduction in the works cost/ton (or the variable cost
per ton of hot metal).

References
Angstenberger, J. (1996) ‘Blast furnace analysis with neural networks’, Artificial Neural Networks
– ICANN 96, pp.203–208, Springer-Verlag, Berlin, Germany.
Brännbacka, J., Torrkulla, J. and Saxén, H. (2005) ‘Simple simulation model of blast furnace
hearth’, Ironmaking & Steelmaking, Vol. 32, No. 6, pp.479–486(8).
Chen, J. (2001) ‘A predictive system for blast furnaces by integrating a neural network
with qualitative analysis’, Engineering Applications of Artificial Intelligence, Vol. 14, No. 1,
pp.77–85.
Dutta, G. and Fourer, R. (2001) ‘A survey of mathematical programming applications in integrated
steel plants’, Manufacturing and Service Operations Management, Vol. 3, No. 4, pp.387–400.
Fabian, T. (1958) ‘A linear programming model of integrated iron and steel production’,
Management Science, Vol. 4, No. 4, pp.415–449.
Fabian, T. (1967) ‘Blast furnace production planning – a linear programming example’,
Management Science, Vol. 14, No. 2, pp.B1–B27.
Forrester, J.W. (1964) Industrial Dynamics, MIT Press, Cambridge, Massachusetts.
 A system dynamics simulation model of a blast furnace for project evaluation 17

Fukui, A., Ifaya, H., Yamana, S., Nigo, S. and Imatoni, N. (1983) ‘A selection system of optimal
operating condition of a blast furnace’, Ironmaking Proceedings, pp.405–410.
Giri, B.K., Petterson, F., Saxen, H. and Chakravarty, N. (2013) ‘Genetic programming evolved
through bi-objective genetic algorithms applied to a blast furnace’, Materials and
Manufacturing Process, July, Vol. 28, No. 7, pp.776–782.
Hooey, P.L., Bodén, A., Wang, C., Grip, C. and Jansson, B. (2010) ‘Design and application of a
spreadsheet-based model of the blast furnace factory’, ISIJ International, Vol. 50, No. 7,
pp.924–930.
Iida, O. and Yuichi, U. (1995) ‘Application of AI techniques to blast furnace operations’, Iron and
Steel Engineer, Vol. 72, No. 10, pp.24–28.
Inkala, P., Karppinen, A. and Seppanen, M. (1995) ‘Computer systems for controlling blast furnace
operations at Rautaruukki’, Iron and Steel Engineer, Vol. 72, No. 8, pp.44–48.
Lankford, W.T., Samways, N.L., Craven, R.F. and MacGanon, H.E. (1985) Making, Shaping and
Treating of Steel, pp.539–590, US Steel Corporation, Pittsburgh.
Lasdon, L.S., McCrady, D.L., Hermansderfer, J., Bennet, W.T., Feltmale, S.D. and Warren, A.D.
(1998) Blast Furnace Charging at U.S. Steel, Working Paper, University of Texas at
Austin [online] http://www.utexas.edu/faculty/msis/PigIronFinal_files/pigoronFinaldoc.htm
(accessed 26 March 2004).
Narchal, R.M. (1988) ‘A simulation model for corporate planning in a steel plant’, European
Journal of Operational Research, Vol. 34, No. 3, pp.282–296.
Oaknock, T.W., Pratt, H.R., Thompson, J.E. and White, D.G. (1983) ‘Utilisation of daily material
and energy balance programme to evaluate and improve blast furnace performance’,
Ironmaking Proceedings, pp.427–437.
Paul, J.R. and Chanev, S.T. (1998) ‘Simulation optimisation using a genetic algorithm’, Simulation
Practice and Theory, Vol. 6, pp.601–611.
Perez, L. and Hanikar, G. (1983) ‘Application of computers for blast furnace control’, Ironmaking
Proceedings, pp.437–445.
Pettersson, F., Chakraborti, N. and Saxen, H. (2007) ‘A genetic algorithms based multi-objective
neural net applied to noisy blast furnace data’, Applied Soft Computing, Vol. 7, No. 1,
pp.387–397.
Roberts, E.B. (1978) Managerial Applications of System Dynamics, pp.3–34, MIT Press, USA.
Savas, E.S. (1961) An Approach to Computer Control of Blast Furnace, Control System, Vol. 4,
Technical Report, IBM Corporation.
Sinha, G.P. and Dutta, G. (1985) ‘A system dynamic model of a blast furnace for project
evaluation’, Proceedings of the International Conference of System Dynamic Society,
Nebraska, pp.838–851.
Sweeny, D.J. and Roppol, L.E. (1983) ‘Development of a burden calculator’, Ironmaking
Proceedings, pp.549–553.
Valenti, M. and Deitz, D. (1997) ‘Modelling furnace performance’, Mechanical Engineering, Vol.
119, No. 12, pp.16–17, ISSN 0025-6501.
Zuo, G., Ma, J. and Bjorkman, B. (1998) ‘Some applications of neural networks for the prediction
of blast furnace irregularities’, Steel Research, Vol. 69, No. 2, pp.41–48.

Glossary A

Technical terms in blast furnace technology

• IronOre: the most important raw material input is iron ore. It is either hematite
(Fe2O3) or magnetite (Fe3O4) and the iron content ranges from 50% to 70%. This
18 G. Dutta and M. Ashtekar

iron rich ore can be charged directly into a blast furnace without any further
processing.
• Pellets: iron ore that contains a lower iron content must be processed to increase its
iron content. Pellets are produced from this lower iron content ore. This ore is
crushed and ground into a powder so that waste material called gangue can be
removed. The remaining iron-rich powder is rolled into balls and fired in a furnace to
produce strong, marble-sized pellets that contain 60% to 65% iron.
• Sinter: produced from fine raw ore, small coke, sand-sized limestone and numerous
other steel plants waste materials that contain some iron.
• Coke: coke is produced from a mixture of coals in coke ovens. The cooked coal,
called coke, is removed from the oven after 18 to 24 hours of reaction time. Coke is
cooled and screened into pieces ranging from one inch to four inches. Coke contains
90% to 93% carbon, some ash and sulphur, but compared to raw coal is very strong.
• Fluxes: limestone and dolomite are fluxes. These are also called reducing agents.
• Burden: Raw materials (iron ore, sinter, pellet, coke, limestone, dolomite)
descending in the blast furnace are called burden.

Glossary B

Technical terms used in the influence diagram

(The term in bracket at the end indicates the unit of technical term).
• Coke rate: tons of coke charged per tons of hot metal produced. The lower the coke
rate, higher the productivity of a blast furnace (tons).
• Carbon burning rate: tons of carbon burnt per ton hot metal produced. it is the
product of coke rate and percentage of carbon in coke (tons).
• Slag rate: tons of slag produced per ton of hot metal produced (tons).
• Wind rate: wind going into the furnace (at normal temperature and pressure) per ton
of hot metal produced (normal metre cube)
• Possible wind rate: It is the theoretical amount of wind (at normal temperature and
pressure) that is required per ton of hot metal produced (normal metre cube).
• Average burden size: The size of raw materials in the furnace has a statistical
distribution. The expected (mean) value of this distribution is the average burden size
(millimetre)
• Sinter rate: tons of sinter charged per ton hot metal produced (tons).
• Hot metal per hour: rate at which hot metal is coming out of the blast furnace. If six
furnaces together produce 7,200 tons in a day, hot metal per hour is 300 (generally
tons).
• Possible production: maximum possible production in a blast furnace without any
delays (tons).
 A system dynamics simulation model of a blast furnace for project evaluation 19

• Downtime: downtime consists of planned delay (preventive shutdown), unplanned

delay (breakdown maintenance), and capital and mid-term repair (hours).
• Available hours: The difference between calendar hours and downtime (hours).
Calendar hour is a function of number of shifts per blast furnace, hours per shift and
number of days per month per blast furnace (hours).
• Mid-term repair: Every year (or second year) the furnace is shut down for repair and
maintenance lasting 15–20 days (hours).
• Capital repair: The life of a blast furnace is about 5–7 years after which it undergoes
capital repair for long-term maintenance lasting 50–70 days. This is also called
relining of the blast furnace.
• Fraction sinter in burden: it is the percentage of sinter in the descending iron bearing
materials, sinter plus iron ore (percentage).
• Sinter charged: quantity of sinter charged in the blast furnace in a month (tons).
• Sinter production: quantity of sinter produced in the sintering plant in a month (tons).
• Ore charged: quantity of ore charged to the blast furnace in a month (tons).
• Ore inventory: inventory of ore at the beginning of the month (tons).
• Coke production: quantity of coke produced in a month in coke ovens (tons).
• Coke inventory: inventory of cokes (tons).
• Coke charged: quantity of coke charged to the blast furnace (tons).
• Hot metal inventory: inventory of hot metal (tons).
• Working volume: the inner volume of the blast furnace. Other things remaining
constant, a higher volume will mean higher production (cubic metre).
• Transfer loss in sinter: percentage of sinter lost in transit from sintering plant to blast
furnace (percentage).
• Transfer loss in coke: percentage of coke lost in transit from coke ovens to blast
furnace (percentage).
• Productivity: The productivity of a blast furnace, defined as the amount of hot metal
produced per unit volume of the blast furnace per unit time. It depends, to a large
extent, on the rate and temperature of air blast. The volume of the blast furnace is the
volume at normal temperature and pressure (tons/m3/day).

View publication stats