Modelling of Fireside Corrosion of Superheaters and ... · Fireside Corrosion of Superheaters and...
Transcript of Modelling of Fireside Corrosion of Superheaters and ... · Fireside Corrosion of Superheaters and...
www.cranfield.ac.uk
Modelling of Fireside Corrosion of Superheaters and Reheaters Following Coal and Biomass Combustion
12th ECCRIA ConferenceCardiff University, Cardiff, UK
5th – 7th September 2018
B. Ekpe, J. Sumner, N.J. Simms
© Cranfield University2
OVERVIEW
Data Collection and Gathering:
• Collate corrosion dataset
• UK Coals
• World Traded Coals
• Include more information (Alloy composition, fuel composition)
• Corrosion Dataset
Data Analysis / Model Development:
• Principal Component Analysis / Principal Component Regression
• Partial Least Squares Regression
Model Validation
Introduction
• Fireside corrosion
• Review of existing corrosion models
Corrosion Dataset
• Fuel: UK coals / biomass
• Fuel: World traded coals (US, EU)
• Materials: Austenitic, ferritic, nickel-based alloys, coatings and claddings
Data Analysis / Model Development
• Principal Component Analysis / Principal Component Regression
• Partial Least Squares
• Model suite
© Cranfield University3
PULVERISED COMBUSTION BOILER
www.tbwes.com/products/subcritical-boilers/
© Cranfield University4
FIRESIDE CORROSION:Deposition on Superheater / Reheater Tubing
Vapour species
Condensation onto solid particles &
aerosolsSolid particles &
aerosols
Water / steam
Corrosion
Heat transfer
SOx HCl
00
Thermophoresisof fine particles
Condensation into deposit Vapours, SOx & HCl
diffuse into porous deposits
Coarse particles stick
Reference: Simms et al., 2007
© Cranfield University5 © Cranfield University5
FIRESIDE CORROSION:High Temperature Corrosion Mechanism
Temperature
Corrosion
damageStable
deposit
Deposit
unstable
Oxidation damage
Temperature
Corrosion
damageStable
deposit
Deposit
unstable
Oxidation damage
CORROSIVE DEPOSITS:
• Sulphate deposits (pyrosulphates, alkali
iron trisulphates, mixed sulphates)
• Chloride deposits
• Sulphates-Chlorides-Carbonates (mixed)
DEPOSIT INSTABILITY:• Vapour condensation dewpoints
• Insufficient SO3 available to stabilise
some phases
• Other phases more stable with
temperature change
Reference: Simms et al., 2007
© Cranfield University6
1. Borio Index: The index value was derived from a nomograph defining factors including
acid-soluble K2O, Na2O, CaO and MgO, and Fe2O3. The coal corrosivity index derived was
of the range 0 - 22 with increasing index denoting increased corrosion.
2. Raask Index: i.e. < 0.5 wt % (Na + K) denotes low coal corrosivity; and > 1.0 wt % (Na +
K) denotes high coal corrosivity.
3. PE-8 model: 𝐶𝑜𝑟𝑟𝑜𝑠𝑖𝑜𝑛 𝑟𝑎𝑡𝑒𝑛𝑚
ℎ𝑟= 𝐴𝐵( 𝑇𝑔
𝐺)𝑚[
𝑇𝑚−𝐶
𝑀]𝑛 %𝐶𝑙 − 𝐷
4. Modified PE-8 model:
𝐶𝑜𝑟𝑟𝑜𝑠𝑖𝑜𝑛 𝑟𝑎𝑡𝑒 𝑛𝑚 ℎ𝑟 = 𝑒𝑥𝑝[( 𝑟 𝑇𝑚) + 𝑠 %𝐶𝑟𝑎𝑙𝑙𝑜𝑦 + 𝑡 𝑇𝑔 − 𝑇𝑚 + 𝑢 + 𝑣 %𝐶𝑙𝑓𝑢𝑒𝑙 ]
5. Laboratory Test Equation:
𝐶𝑜𝑟𝑟𝑜𝑠𝑖𝑜𝑛 𝑟𝑎𝑡𝑒 𝑛𝑚 ℎ𝑟
= exp[( 𝑎 𝑇𝑚) + 𝑏 + 𝑐 𝐻𝐶𝑙 + 𝑑 𝑆𝑂𝑥 + 𝑒(𝑁𝑎 + 𝐾)𝑑𝑒𝑝%+ 𝑓 %𝑆𝑑𝑒𝑝 + 𝑔 %𝐶𝑙𝑑𝑒𝑝 ]
EXISTING CORROSION MODELS
References: Borio et al., 1968[1]; Clapp, 1991[2]; James et al., 2007[3,4, 5]; Lant et al., 2010[3,4]; Wright and
Shingledecker, 2015[1,2,3]
© Cranfield University7 © Cranfield University7
OVERVIEW
Data Collection and Gathering:
• Collate corrosion dataset
• UK Coals
• World Traded Coals
• Include more information (Alloy composition, fuel composition)
• Corrosion Dataset
Data Analysis / Model Development:
• Principal Component Analysis / Principal Component Regression
• Partial Least Squares Regression
Model Validation
Introduction
• Fireside corrosion
• Review of existing corrosion models
Corrosion Dataset
• Fuel: UK coals / biomass
• Fuel: World traded coals (US, EU)
• Materials: Austenitic, ferritic, nickel-based alloys, coatings and claddings
Data Analysis / Model Development
• Principal Component Analysis / Principal Component Regression
• Partial Least Squares
• Model suite
8
ALLOYS Austenitic, Ferritic, Coatings/Claddings
FUEL Coal (UK, US, EU); Biomass
OPERATING CONDITIONS Metal Temperature, Gas Temperature
TUBE POSITION Leading (facing gas flow), Non-leading
CORROSION DATASET Input Parameters from Plant Data
© Cranfield University9 © Cranfield University9
OVERVIEW
Data Collection and Gathering:
• Collate corrosion dataset
• UK Coals
• World Traded Coals
• Include more information (Alloy composition, fuel composition)
• Corrosion Dataset
Data Analysis / Model Development:
• Principal Component Analysis / Principal Component Regression
• Partial Least Squares Regression
Model Validation
Introduction
• Fireside corrosion
• Review of existing corrosion models
Corrosion Dataset
• Fuel: UK coals / biomass
• Fuel: World traded coals (US, EU)
• Materials: Austenitic, ferritic, nickel-based alloys, coatings and claddings
Data Analysis / Model Development
• Principal Component Analysis / Principal Component Regression
• Partial Least Squares
• Model suite
© Cranfield University10 © Cranfield University10
MODEL DEVELOPMENT: Principal Component Analysis / Partial Least Squares
Reference: PCA illustration - Transformation of
input data using PCA (University of Illinois, 2000)
\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\PCA is employed to
reduce the
dimensionality of a
set of variables while
retaining maximum
variance
PLS extracts linear functions
of the predictor dataset that
has maximum covariance with
the dependent variable
(corrosion rate)
© Cranfield University11
MODEL STRUCTURE
Arrhenius Equation:
Taking the natural logarithm on both sides gives: 𝑰𝒏 𝑲 = 𝑰𝒏 𝑨 −𝑬𝒂
𝑹𝑻
Kelvin Temperature
Pre-exponential Factor
Activation Energy
Rate Constant
Gas Constant
𝐾 = 𝐴𝑒−𝐸𝑎𝑅𝑇
𝑰𝒏 𝒚 = 𝑰𝒏( 𝜷𝒐 ) + 𝜷𝟏𝒙𝟏𝒊 + 𝜷𝟐𝒙𝟐𝒊… .……+ 𝜷𝒌𝒙𝒌𝒊… .+ 𝜺𝒊
1000/Metal temperature , Tm
(1000/K)
Alloy composition (wt %)
Fuel composition (wt %)Constant
Corrosion rate (µm.(khr)-1)
Residual error
© Cranfield University12 © Cranfield University12
PRINCIPAL COMPONENT ANALYSIS
Input dataset mean centered and standardised
Variance (eigenvalue) and weight vector
(eigenvector) computed from
correlation matrix
Principal components with significant
eigenvalues selected via a selection
criterion
Original variables mapped onto the new
coordinated space
Principal component regression using
selected PCs
Variables with high loadings on selected
PCs added to the regression model
Standardised regression coefficients
transformed to original
(unstandardized) form
Prediction model
© Cranfield University13 © Cranfield University13
PARTIAL LEAST SQUARES
PLS components selected via cross
validation
PLS weights extracted to build
the model and used later for
variable selection
PLSR model is fitted to the data
Variable selection is carried out by
introducing a threshold
(VIP Scores)
Prediction model generated
Reference (VIP scores): Mehmood et al (2012); Farres et al (2015)
© Cranfield University14 © Cranfield University14
Model Materials Tube
Position
Fuel
Model 1Austenitic Fe based
Leading Midland and North Eastern
Region UK Coals
Model 2 Non-leading
Model 3Ferritic
Leading Midland and North Eastern
Region UK CoalsModel 4 Non-leading
Model 5 Austenitic Fe based
Position not
specified
Biomass - Clean wood
pellets, waste wood and
forestry waste
Model 6 Ni based alloys
Model 7 Ferritic
Model 8 Coatings and claddings
Model 9 Austenitic Fe based
US Coal – Eastern and
Western Bituminous coalModel 10 Ni based alloys
Model 11 Coatings and claddings
Model 12 TP316, HR3C10-20% Wood/CCP, Daw
Mill Coal
MODEL SUITE
© Cranfield University15 © Cranfield University15
PLS – VARIABLE SELECTION FOR MODEL BUILDING
(MODEL 1, UK COAL DATA)
0.0
0.5
1.0
1.5
2.0
10
00
/Met
al T
em
p (
100
0 /
K)
(Tg
-Tm
) (°
C)
Car
bo
n (
%)
Man
gan
ese
(%)
Ph
osp
ho
rus
(%)
Sulp
hu
r (%
)
Silic
on
(%
)
Ch
rom
ium
(%
)
Nic
kel (
%)
Mo
lyb
den
um
(%
)
Tita
niu
m (
%)
Van
adiu
m (
%)
Na
(%)
K (
%)
Mg
(%)
Ca
(%)
Al (
%)
Fe (
%)
Si (
%)
Cl (
%)
S (%
)
Ti (
%)
VA
RIA
BLE
IMP
OR
TAN
CE
IN P
RO
JEC
TIO
N
© Cranfield University16 © Cranfield University16
PLS – VARIABLE SELECTION FOR MODEL BUILDING
(MODEL 5, UK BIOMASS DATA)
0.0
0.5
1.0
1.5
2.01
00
0/M
etal
Te
mp
(1
000
/ K
)
(Tg-
Tm)
°C
Car
bo
n (
%)
Man
gan
ese
(%)
Ph
osp
ho
rus
(%)
Sulp
hu
r (%
)
Silic
on
(%
)
Ch
rom
ium
(%
)
Nic
kel (
%)
Mo
lyb
den
um
(%
)
Nio
biu
m (
%)
Tita
niu
m (
%)
Van
adiu
m (
%)
Nit
roge
n (
%)
Co
pp
er (
%)
Tun
gste
n (
%)
Bo
ron
(%
)
Alu
min
ium
(%
)
Co
bal
t (%
)
Na
(%)
K (
%)
Mg
(%)
Ca
(%)
Al (
%)
Fe (
%)
Si (
%)
Cl (
%)
S (%
)
Ti (
%)
P (
%)
Ba
(%)
Mn
(w
t%)
VA
RIA
BLE
IMP
OR
TAN
CE
IN P
RO
JEC
TIO
N
© Cranfield University17 © Cranfield University17
PLS – VARIABLE SELECTION FOR MODEL BUILDING
(MODEL 9, US COAL DATA)
-0.4
0.1
0.6
1.1
1.6
2.1
2.61
00
0 /
Met
al T
em
p (
10
00
/ K
)
(Tg-
Tm)
(°C
)
Car
bo
n (
%)
Man
gan
ese
(%)
Ph
osp
ho
rus
(%)
Sulp
hu
r (%
)
Silic
on
(%
)
Ch
rom
ium
(%
)
Nic
kel (
%)
Mo
lyb
den
um
(%
)
Nio
biu
m (
%)
Tita
niu
m (
%)
Co
pp
er (
%)
Van
adiu
m (
%)
Alu
min
ium
(%
)
Nit
roge
n (
%)
Cer
ium
(%
)
Tun
gste
n (
%)
Bo
ron
(%
)
Tan
talu
m (
%)
Na
(%)
K (
%)
Mg
(%)
Ca
(%)
Al (
%)
Fe (
%)
Si (
%)
Cl (
%)
S (%
)
Ti (
%)
P (
%)
Mn
(%
)
VA
RIA
BLE
IMP
OR
TAN
CE
IN P
RO
JEC
TIO
N
© Cranfield University18 © Cranfield University18
PLS – VARIABLE SELECTION FOR MODEL BUILDING
(MODEL 12, UK CO-FIRING DATA)
0.0
0.5
1.0
1.51
00
0/M
etal
Te
mp
(1
000
/(K
))
Car
bo
n (
wt%
)
Silic
on
(w
t%)
Ch
rom
ium
(w
t%)
Nic
kel (
wt%
)
Nio
biu
m (
wt%
)
Nit
roge
n (
wt%
)
Mo
lyb
den
um
(w
t%)
Na
(%)
K (
%)
Al (
%)
Si (
%)
P (
%)
Mg
(%)
Ca
(%)
Ti (
%)
Mn
(%
)
Fe (
%)
Ba
(%)
Cl (
%)
S (%
)
VA
RIA
BLE
IMP
OR
TAN
CE
IN P
RO
JEC
TIO
N
© Cranfield University19 © Cranfield University19
COMPARISON OF PREDICTED AND MEASURED CORROSION RATES USING PLS
Austenitic (Coal_UK)
Austenitic (Biomass_UK)Austenitic (Biomass)
Austenitic (Biomass_UK)
Austenitic (Coal_UK) Austenitic (Biomass_UK)
R² = 0.85
0
1
2
3
4
5
6
0 1 2 3 4 5 6Pre
dic
ted
rat
es (
µm
/kh
r)
Measured rates (µm/khr)
Model 1 (UK Data)
R² = 0.73
0
1
2
3
4
5
6
0 1 2 3 4 5 6Pre
dic
ted
rat
es (
µm
/kh
r)
Measured rates (µm/khr)
Model 2 (Biomass Data)
R² = 0.90
1
2
3
4
5
6
1 2 3 4 5 6 7 8Pre
dic
ted
rat
es (
µm
/kh
r)
Measured rates (µm/khr)
Model 3 (Co-firing Data)
R² = 0.75
1
2
3
4
5
6
7
1 2 3 4 5 6 7Pre
dic
ted
rat
eS (
µm
/kh
r)
Measured rates (µm/khr)
Model 4 (US Data)
Model 1 Model 5
Model 12Model 9
© Cranfield University20 © Cranfield University20
CONCLUSIONS• The analysis suggests a sulphur dominated fireside corrosion dependent on the fuel Cl and
Na levels in UK data; for biomass data (comprising mainly wood pellets), alkali sulphate
interaction with fireside corrosion dependent mainly on K, Ca and Ti; a fireside corrosion
model dependent on Cl, K, Mg, S and P in co-firing data, and the corrosion model from US
data is dependent mainly on the fuel’s sulphur content.
• Variable selection for fireside corrosion models is in line with conclusions from previous
investigations – for instance; chloride influence in UK coals.
• PLS on average produced better predictive models than PCA, and with fewer latent
variables compared to PCA. On PCA, the more latent variables (or PCs) selected, the
better the model performance; this is not the case for PLS so there is less chance of
overfitting the data.
• PCA also includes more predictor variables with negligible effects on the corrosion rate
compared to PLS
© Cranfield University21
• Biomass and Fossil Fuel Research Alliance (BF2RA)
• Niger Delta Development Commission (NDDC)
• Unpublished data from Doosan, E-ON/Uniper, EPRI
• Supervisors – Professor Nigel Simms
– Dr. Joy Sumner
ACKNOWLEDGEMENTS
© Cranfield University22 © Cranfield University22