Minimills
Optimisation of EAF operating practice
using iCSMelt
An optimisation tool developed by CSM
for the EAF has been implemented at
Duferco La Louvière where in different
configurations a reduction of close to
1% in total specific energy consumption
and of 3.25% in specific electrical power
consumption has been obtained
depending on charge mix and
optimisation strategies used.
By P Frittella*, A Lucarelli*,
B Poizot** & M Legrand**
THE iCSMelt® – ‘Intelligent Care Steel
Melting’ is a tool developed by Centro
Sviluppo Materiali (CSM) of Italy with the
aim of optimising EAF operating practices
(OOP) in terms of selected objective functions (OF), such as power-on time, heat cost
and energy consumption.
The need to reduce energy consumption (and
consequently specific CO2 emissions) as well as
enhancing productivity are the main goals for
today’s electric steelmakers. However the
increase in the price of scrap forces steelmakers
to use lower quality scrap or scrap substitutes to
reduce production cost. Therefore attention
must be put on reliability and flexibility of the
process and on repeatability of performance in
terms of yield, tap to tap time, energy consumption (both electrical and chemical), composition
and temperature of the liquid steel at tapping.
Recently several R&D projects[1–3] have
focused on individual processes by monitoring
and control through combinations of:
– Hardware sensors, commercially available,
for process monitoring (such as off-gas
analyser, liquid bath temperature measurement, slag level, etc)[4–6];
– Software sensors for generating information
not directly available from measurements
(eg a metallurgical model to estimate slag
and steel composition and temperature)[7–9].
However the definition of operating practices for steel production in the EAF is neces*Centro Sviluppo Materiali SpA, Italy, p.frittella@c-s-m.it
**Duferco La Louviere S.A., Belgium
Panel
temperature
sary as target for each control system in use
(yellow blocks in Fig 1).
The iCSMelt tool has been developed by
CSM coupling static mass and energy balance
with an optimisation tool. This article
describes the customisation of iCSMelt for the
DC electric arc furnace at Duferco La
Louvière (DLL) in Belgium and presents
results of its application in the frame of the
project ‘Control and optimisation of scrap charging strategies and melting operations to increase
steel recycling ratio’ (CONOPT-SCRAP) which
has been carried out with the help of a financial grant from the Research Fund for Coal
and Steel of the European Community.
The iCSMelt tool has the aim of obtaining
optimised operating practices (OOP), starting
from standard operating practices (SOP) in
use on the specified EAF, following selected
objective functions (OF). The iCSMelt tool
includes different modules (Fig 2):
– iCSMelt Process Model (iCSMelt-PM), the
process simulator;
– iCSMelt Calibrator (iCSMelt-Cal) the tool,
based on Genetic Algorithm (GAs), for automatic tuning of calibration parameters of the
iCSMelt-PM;
– iCSMelt Optimiser (iCSMelt-Opt), the optimisation tool based on the GAs.
In the specific case of Duferco La Louvière,
the iCSMelt modules have been integrated
with an on-line data acquisition system (DAS)
and a process data analyser (iCSMelt-PDA),
the plant data viewer and inputs generator for
the process model. Presently the relevant data
for each heat are recorded in a dedicated database that can be read by iCSMelt-PDA.
iCSMelt Process Model
The iCSMelt-PM can be classified as a pseudodynamic mass and energy balance. Each heat is
represented by several charge baskets (two for
DLL), each subdivided into several phases
(four and six respectively for DLL), (Fig 3).
The end of each phase is defined by a fixed
level of a particular process parameters, or
their combinations, following specific plant
customisation. Each phase is divided in time
steps (⌬t) for which static mass and energy bal-
Off-gas:
Electrical variables
-Composition
- Mass flow
-Temperature
Melting
sensor
Charge material
mix recipes
Electrode
Burners
Different options are available in
iCSMelt-PM:
By Plant Data: iCSMelt works using SOP
acquired by iCSMelt-DAS and converted by
iCSMelt-PDA in input for iCSMelt-PM.
By Model Strategy: heat simulation is performed following a SOP defined by process
engineer; two strategies for definition of end
of phases are possible:
– Technological strategy: phases end is granted by technological steps as in Fig 3;
– Operator strategy: phases end is set by specific electrical energy consumption levels.
End of simulation: defines the condition of
termination of the simulation according to:
– End Point Temperature: termination corresponds to desired steel target temperature.
– Until Last Data: heat simulation terminates at the end of defined SOP.
Free board model
- energy model
Dynamic evaluation of:
- Bath T
- Bath & slag comp.
Carbon injectors
Energy
demand
Static energy &
Mass balance
The output of the iCSMelt-PM simulation
are:
– Steel bath at tapping: temperature, composition, tapping weight;
– Slag bath at tapping: temperature, composition, weight, basicity index (IB);
– Off gas: temperature, flow rate, composition
at IV hole and in the duct after Postcombustion;
– Consumptions: electrical energy, methane,
oxygen, carbon;
– Energy Balance: energy inputs (electrical, natural gas, oxidations, post combustion) and
energy outputs (losses by cooling water panels, by furnace bottom and by off-gases); and
– Duration of each phases.
Metallurgical model
- slag & bath composition
Harmonics
Charge mix
composition
ance is calculated (Fig 4). Terms included are:
– Material inputs and their energies (white in
Fig 4): basket charge, EAF additions, previous heat residuals.
– Process outputs (green): steel, slag and
off gas.
– Terms from the SOP managed/exchanged
between iCSMelt-PM and iCSMelt-Opt
(red): electrical voltage, current, total natural
gas flow rate, total oxygen flow rate, total
carbon flow rate.
– Undesired/uncontrolled terms in input and
output from the process (cyan).
Optimisation
tool
Oxygen injection
Post-Combustion injectors
operating patterns for
each individual heat
Charge
monitoring
system
Fluxes additions
Process data
(L1)
Steel target
T and bath composition
(L2)
Expert system
(L3)
Fig 1 Example of the global approach of EAF process management
November/December 2009 – Steel Times International
Fig 2 Architecture of iCSMelt environment
Minimills
Information Flux
Start
Plant
geometry
Charge
mix
Elaboration
SOP
Sett 1
EAF
Total heat
I Basket
Sett 2
Tapping
Boring
Molten metall formation
Sett 3
Main Melting Period
Sett 4
% off I baskett Melting
Sett 5
.. Basket
Output
Phases / Global
EAF process
model
Boring
Sett 6
Molten metall formation
Sett 7
Main Melting Period
Sett 8
Slag Foaming
Sett 9
Totall Melting
Sett 10
Tapping Aim
e and
d Composition off tapped steell and
Weight Temperature
d
slag - Global Consumptions – Global Losses - Time
Input
Fig 5 Scheme of interaction between iCSMelt-Opt and iCSMelt-PM
Fig 3 Flux of information in iCSMelt and evolution of calculations
iCSMeltt Customization
n Site
On
iCSMelt®
iCSMelt®
Calibratorr
installation
Process Data
Analyzer
Definition
n off
n parameters
calibration
iCSMelt®
iCSMelt®
Optimizer
iCSMelt®
iCSMelt®
s Modell
Process
Heatt Data
a
Manual - Improved
Operating Practice
Electrical/Chemicall
Parameters
Automatic Optimized
Operating Practice
y Losses
Energy
Data
a
n
Acquisition
System
OOP/IOP
P
Implementation
EAF
Technology
Engineers
Fig 6 General configuration of iCSMelt
0.8
k_O_Inj
%FeO
slag Weight
K_O_Inj
- Average = 0.76
- Std Dev = 0.043
K_O_Inj
35
30
0.7
25
0.6
0.5
20
0.4
15
0.3
10
0.2
5
0.1
0
50589
40
% FeO in slag, Slag
Weight (ton)
1
0.9
50590
50591
50592
50593
50594
0
50595
Heat
Fig 4 Mass and energy balance for iCSMelt
Fig 7 Variability of K_O_Inj for Charge Mix 70
iCSMelt Optimiser
iCSMelt-Calibrator
iCSMelt-Opt is based on the GAs (Genetic
Algorithms) method. The objective functions
(variables yi ) are:
– power on;
– specific total cost;
– specific electrode consumption;
– specific coal powder consumption;
– specific total energy consumption; and
– specific electrical energy consumption.
At each iteration of the optimisation process
iCSMelt-Opt receives as input:
– indications of phases to be optimised (Phase
Optimised Selection);
– range of variability of the SOP parameters
for each phases (SOP Constraints);
– mathematical parameters for the GAs management (GAs Management Parameters);
– value reachable for specific targets of interest (heat aims).
Calibration parameters are available in iCSMeltPM to reach the best fitness with process data at
the end of the heat. Manual calibration is also
possible. However data available in DAS also
allows the automatic re-calibration of the parameters, treating a higher number of calibration
data in a limited time, in case of changes of
process management or the charge mix used.
Terms of the Operating Practice (OP) managed by iCSMelt-Opt are:
– electrical current intensity;
– electric voltage;
– total natural gas flow rate;
– total oxygen flow rate; and
– total injected carbon.
These are depicted in Fig 5.
iCSMelt-Opt evaluates results of proposed
OP through the iCSMelt-PM. If the target is
not reached the sub-module, named ‘GAs
practices generator’, calculates a new OP to
obtain better results with respect to the previous calculated simulation. The formula for
The ‘End of Simulation/last data’ is used in
a calibration procedure when model parameters are set to reach the target tapping temperature, while the ‘End of Simulation/End Point
Temperature’ is used during the optimisation
procedure when it is necessary to evaluate the
duration of the heat to reach the tapping temperature target.
Steel Times International – November/December 2009
Minimills
Equation 1
Equation 2
The constraint regions are implemented as
an extension of the target function evaluation:
if the GAs practices generator builds a new OP
that contains one or more variables that are
outside of constraints, the output of the EAF
model is forced to a very high value
(equation 2).
where:
– V is the set of valid operating practice values;
– uk is each of m specific component of OP.
The optimisation process ends when convergence occurs or when the time limit of the
procedure is reached. Since the aim of the
optimiser is to find a good solution in a reasonable time, the control of genetics algorithms is included in the GAs practices generator. Two parameters, configurable by the user,
allow the management of the GAs practices
generator: 1) size of population 2) number of
generations.
Implementation at Duferco
La Louvière
The main characteristics of the EAF at Duferco
La Louvière (DLL) are given in Table 1.
Five main steps have been taken during the
installation of iCSMelt at DLL. These correspond to the implementation of the main
modules (Fig 6).
Step 1 Acquisition: The iCSMelt tool has
been integrated at DLL for data interchange
with level 1 automation. As per today, EAF
process signals corresponding to relevant heat
data are recorded in a database, DAS, that can
be read by the iCSMelt-PDA and transferred
to iCSMelt.
Bottom; 1
Roof; 9
specific oxygen consumption, without significant variation of specific electrical energy consumption and with a small decreasing of specific total energy consumption (–0.25%).
IOP3 (Fig 10) has been defined with the
scope of achieving a reduction of specific electrical energy consumption to respond at a lower
production scenario. Therefore, the increase of
power-on time is considered acceptable in a
range of plus 5% with respect to DLL standards.
The iCSMelt simulation shows a reduction of
specific electrical energy consumption of -3.1%,
and as expected, an increase in the power on
time of+3.2% as indicated in Table 2.
Step 5 Definition of Optimised Operating
Practices (OOP): In the automatic definition
of the OOP, to optimise the objective function
(OF) selected by the process engineer, the
modules iCSMelt-PM and iCSMelt-Opt are
involved. At DLL, OOP has been generated
for charge Menu 47 (OOP47) and subsequently the procedure has been adapted to the other
two charging mix (OOP30 and OOP70).
The target of OOP47 has been to minimise
the total specific energy consumption.
Simulation of OOP47 has achieved a reduction of 2.4% of specific total energy input
(Fig 11, Table 3). Furthermore iCSMelt-Opt
has been used to define OOP for charging mix
70 for other objectives functions: minimisation
of specific coal consumption and minimisation
of specific total cost.
Step 4 Definition of Improved Operating
Practice (IOP): in this step iCSMelt-PM is
used alone as support to evaluate the effects
of OP modifications. In particular at DLL two
main optimisation scopes have been pursued:
– IOP 2: reduction of specific carbon consumption, at least by 10%;
– IOP 3: reduction of specific electrical energy
consumption.
Industrial test of iCSmelt at Duferco
La Louvière
Fig 9 shows the results of simulations performed at different levels of oxygen flow rate
used to define IOP2 characterised by a reduction of 10% of the carbon flow rate and 2.5%
of oxygen flow rate at each step with respect
to the previous standard practice (SOP).
The IOP2 gives a reduction of 9.97% in
specific carbon consumption and 1.73% of
Test with IOP3
Electrical Losses;
11
Energy Unbalance;
Off Gas; 109
10
Lateral panels; 65
CH4 NmC
Specific EE_Cons kWh/ton
Diff%CinTapSteel/10
Specific Oxygen Nmc/ton
5.00
Electrical Energy;
312
Slag; 28
Burners; 53
Oxidations; 95
Steel; 393
Charge; 62
Post Combustion;
104
The new OPs obtained with iCSMelt for different charge mixes and different objective
functions (minimising specific consumption of
total energy, minimising specific total cost of
the heat, minimising specific total coal consumption) have been tested at DLL. Two
example are reported here.
Starting from the total oxygen, natural gas and
carbon calculated by iCSMelt for IOP3, the
sets for each of the injectors have been
obtained following the DLL indication.
Comparison of data for heats with SOP and
heats performed with IOP3 (Table 4) shows a
3.24% reduction of specific electrical energy
P On Min
Specific Tot En Cons.kWh/ton
Spec C Cons. kg/ton
Definition IOP Target Red Carbon Cons
Reducing 10 % Flow of Carbon
20.00
4.00
15.00
3.00
10.00
2.00
5.00
1.00
0.00
0.00
-1.00
-5.00
-2.00
-10.00
-3.00
-15.00
-4.00
SOP 50594
- 10% C Flow
- 10% C Flow
- 10% O2
- 10% C Flow
- 5% O2 Flow
- 10% C Flow
- 2.5% O2
-20.00
-5.00
Cases
Fig 8 Typical energy balance at DLL evaluated by iCSMelt for
charging mix 70
November/December 2009 – Steel Times International
Fig 9 Results of simulation performed with modification of
standard practice for the test case IOP2
Variation % - Sp C Cons, Sp O2 Cons,
%C in Ste e l/1 0
where:
– T is the target function value;
– yi are main output of heat simulation for
objective functions and steel melt composition;
– yi* are quantities (n) of model output fixed
for objective functions as aims of heat;
– wi are weights of single model output fixed
as objective function.
Step 2 Calibration: The calibration parameters of iCSMelt-PM to obtain the best fitness
with the results of real heats are defined for
the EAF. Since the off-gas measurements are
not available at DLL, typical values for EAF
process are assumed. The charge mixes typically used at DLL are named ‘Menu 30’, ‘47’,
and ‘70’. The definition of calibration parameters has been performed for homogeneous
groups of heats.
Fig 7 shows, as an example, the variability
of the calibration parameter K_O_Inj, related
to the efficiency of oxygen injection on
the steel/slag bath, for charge mix 70.
Moreover different calibration procedures
have been performed to obtain a sensitivity
analysis of the effect of the different calibration parameters to simulations results.
Step 3 Simulation: the simulation of the
heat by iCSMelt-PM. This step requires general inputs such as plant configuration data, different scrap grades in use (to be filled in the
available data base) and a description of the
heat phases in addition to a SOP. Criteria for
phase transition has been customised based on
specific process practices at DLL. In Fig 8, an
example of energy balance for menu 70 is
shown where:
– Electrical energy is 55% of total energy input
while chemical energy is 45%.
– Energy to the steel/slag bath is 68% of total
energy while 32% are energy losses.
Variation % - SpEE, CH4, Power On,
Sp Tot En
the target function which results in minimisation is given in Equation 1.
Minimills
2
Charge weight
ton
96/98
Liquid steel capacity
ton
110
Heat size
ton
89/91
Shell diameter
m
6800
Yield
%
93
Electrode diameter
mm
711
Active power
MW
61 (ave.)
Transformer
MVA
99
Power on
min
30/32
Max power
MW
80
Power off (no delays)
min
10/12
Tapping temperature
ºC
1620
Tap to tap (no delays)
min
40/44
Carbon content at tapping
%
0.05
Electrical Energy
kWh/ton
360–380
Number of buckets
U
2
Electrode consumption
Kg/ton
1.1–1.2
Average scrap yield
%
93
Total Oxygen
Nm3/ton
43–45
Total Gas natural
Nm3/ton
5–6
30
Scraps
%
90
Total Carbon
Kg/ton
Pig Iron
%
10
Productivity max.
ton/hour 130/140
SOP
Difference %
ºC
1657
1658
–0.1
Coal Consumption
kg
1240
1318
–5.9
Nmc
558
589
–5.3
min
32
31
3.2
kWh/ton
316
326
–3.1
Total Energy In Input (EE+CH)
kWh
66847
67449
–0.9
Electrical Energy Consumptions
kWh
31131
32165
–3.2
Table 2 Schematic view of comparison between simulation of IOP3
and SOP by iCSMelt
A test of the implementation of OOP47 has
been made by DLL at heat 56293 with some
adaptation for distributing iCSMelt total flow
rates for each lance and having the proper balance of flows in various EAF zones. In particular, modifications were:
– adjust the ratio between oxygen and gas flow
rate in the step 3 of first basket;
– in phases 5 and 6 of basket 2 the carbon flow
rate was left at the same value as in step 4
In detail, simulation of the original OOP47
showed a reduction of 2.9% in total energy
consumption corresponding to 2.4% in terms
of total specific energy consumption. Using
the adjusted OOP47 the reduction of total
specific energy consumption results in a 1.2%
reduction (Table 3). Heat 56293 performed
with an adjusted OOP47 results with a reduction in total specific total energy consumption
of 0.7% while the power-on time remained in
the range of traditional DLL practice (33.5
minutes) and very close to the predicted
34.07 minutes. Acknowledgement
The authors are grateful to Mrs E Malfa and Mr V Volponi, for
140
30
180
160
Gas Present SOP
Gas New SOP
Carbon New SOP
IOP - Target Red EE Sp Cons
Chem Param IOP VS SOP Basket 2
12000
120
10000
100
8000
80
6000
60
4000
40
2000
20
0
0
0
50
100
150
EEsp (kWh/ton)
200
250
Fig 10 Comparison between SOP and IOP3 for the electrical and
chemical parameters
Tension SOP
Current SOP
References
Tension OOP47
Current OOP47
CH4 SOP
Oxi SOP
Carbon SOP
Basket 1
120000
900
500
60000
400
300
40000
Current (A)
80000
Basket 1
200
180
7000
100000
700
600
CH4 OOP47
Oxi OOP47
Carbon OOP47
8000
800
[1] 9th European Electric
Steelmaking Conference,
19-21 May 2008,
Krakow, Poland
[2] P Clerici et others,
Innovation in EAF and
Steelmaking Processes,
Italy, Milan, 27-28 May
2009
[3] B Kleimt et others,
8th European Electric
Steelmaking Conference,
Birmingham 2005
[4] H D Goodfellow,
M Pozzi, J Maiolo,
AISTech Proceeding
2006
[5] BFI – Direct Optical
measurement
[6] J van Wky and
others, Agellis group
[7] A Di Donato et
others, 2nd European
Oxygen Steelmaking
Congress, EOSC ‘97,
Taranto, Italy, 13-14-15
October 1997
[8] P Nyssen et others,
7th EESC, Venice, Italy,
26-29 May 2009
[9] JG Bekker, IK Craig
and PC Pistorius, ISIJ
International, Vol 39,
No 1, January 1999,
pp 23-32
120
200
20000
100
160
6000
140
5000
120
100
4000
80
3000
60
2000
40
1000
0
0
0
1
2
3
4
5
6
Tension OOP47
Current OOP47
0
0
Phase
Tension SOP
Current SOP
20
0
4
Phase
CH4 OOP47
Oxi OOP47
Carbon OOP47
5
6
Basket 2
14000
200
180
100000
12000
700
80000
500
60000
400
40000
300
3
120000
800
600
2
CH4 SOP
Oxi SOP
Carbon SOP
Basket 2
900
1
Oxigen - CH4 (Nmc/h)
Test with OOP47
100
Oxygen - CH4
and a 2.83% increase in power-on time, values
very close to those estimated by iCSMelt in
Table 2 (3.1% and 3.2% respectively). The
decrease in natural gas and carbon consumption is also in line with the iCSMelt evaluation.
80
Current (A)
Special Electrical Energy
60
Current (A)
Power On
40
Tension (V)
Methane Consumption
20
Specific electrical energy (kWh/ton)
Tension (V)
UM
50
40
Oxygen Present SOP
Oxy New SOP
Carbon Present
Oxygen, Methane
(Nmc/h)
IOP3
Tapping Temperature
70
60
Tension (V)
Parameter
90
80
0
Table 1 Main characteristics of EAF at Duferco La Louvière
Comparison between heat simulation
with IOP3 and SOP by iCSMelt
110
100
200
20000
Oxigen - CH4 (Nmc/h)
Charges
1000
900
800
700
600
500
400
300
200
100
0
Carbon Flow Rate
(kg/min)
No
90
Definition IOP - Target Red. Sp EE Cons
Electrical Param. IOP VS SOP - Basket 1
Carbon
(kg/min)
Carbon (kg/min)
Scraps buckets
ton
kA_New_SOP
Carbon
(kg/min)
Carbon (kg(min)
EBT–DC
Tapped steel
V_New_SOP
kA_Present_SOP
Oxigen (Nmc/h)
- CH4 (Nmc/h)
Furnace type
V_Present_SOP
Current (kA)
Operation Data
Tension (V)
General Data
160
10000
140
120
8000
100
6000
80
60
4000
40
2000
100
0
0
0
1
2
3
4
5
6
Phase
20
0
0
0
1
2
3
Phase
4
5
6
Fig 11 Comparison between SOP used and OOP47 defined
Power on
min
Electrical
Energy
kWh
Average of heat values
for heats with SOP
0:31:09
33963
3935
591
1248
Average of heat values
for heats with IOP3
0:32:02
32856
3868
591
1055
Differences of heats values
with IOP3 vs SOP
0:00:53
–3.26%
–1.69%
–0.05%
–15.48%
Test of Implementation
of “OOP47” to DLL EAF
Specific differences of heat values
with IOP3 vs SOP (kWh/ton)
Oxygen
Nm3
Methane
Nm3
Carbon
kg
–3.24%
Table 4 Comparison between heat data with actual standard
operation (SOP) and that predicted by IOP3
their useful collaboration in preparing this paper.
Test of Implementation of
“OOP47” to DLL EAF
Power on
min
Electrical
Energy
kWh
Methane
Oxygen
Carbon
Nm3
Nm3
kg
Sp. Total
Energy in Input
kWh/ton
“SOP”
iCSMelt simulation
32.0
31223
567
4292
1231
628
“OOP47 modified”
Simulation results
43.0
34987
356
3625
1085
621
33.5
35300
337
3663
1085
simulated by iCSMelt
Difference VS SOP
Heat results after
application of
“OOP47 modified”
Heat values
Difference VS SOP
–1.2%
624
–0.7%
Contact
Centro Sviluppo Materiali SpA,
Via di Castel Romano, 100-00128
Rome - Italy - P.IVA 00903541001
Tel. +39 06 50551
Fax. +39 06 5050250
Web. www.c-s-m.it
Table 3 Comparison between ‘SOP’, ‘OOP47’, ‘OOP47 adjusted’ simulated and heat results
Steel Times International – November/December 2009