Professional Documents
Culture Documents
07-Multivariate Analysis of Fouling in Kraft Recovery Boilers
07-Multivariate Analysis of Fouling in Kraft Recovery Boilers
Highlights: Operating data from three recovery boilers was analyzed using Principal Component Analysis
(PCA) and Partial Least Squares Analysis (PLS). PCA was used to compare operational periods
of high fouling and low fouling and for analyzing individual runs in all boilers. PLS was used to
extract the correlation structures between model input and output variables showing that some
variables were disproportionably associated with fouling. Together, these two techniques can be
used to identify operating variables that may be adjusted to minimize fouling and for developing
an on-line fouling monitoring technology.
Status: Complete
7-1
correlated variables in complex systems. Large
amounts of process data from multiple variables can
be represented as a single data point that moves in
time along multiple axes called principal components
(PC’s). PC’s are new variables built using PCA; they
represent statistically significant processes in the
system and are usually smaller in number than the
Flue Gas original variables being measured [9,10]. The ability
to visualize large amounts of data across a number of
variables on a simple plot makes this technique
powerful.
7-2
variance in the data, explains the variability due to
increases in the loading of the boiler.
Figure 3. Variation in the t1/t2 plot with respect to black liquor flow rate for Recovery Boiler A.
7-3
Figure 4. Variation in the t1/t2 plot with respect to boiler bank outlet temperature for Recovery Boiler A.
Figure 5. Variation in the t1/t2 plot with respect to final steam flowrate for Recovery Boiler A.
7-4
and having high energy transfer efficiency. The
lower right quadrant (III) shows data points for
operating days when the boiler was at a high load,
but had low energy transfer efficiency due
presumably to fouling. Similar results were found
for Recovery Boilers B, with the first PC explaining
the variability due to boiler loading, and the second
PC explaining the variability due to changes in
energy transfer efficiency. Due to a major process
change in the dataset for Recovery Boiler C, the
statistically significant processes representing boiler
Figure 6. Key processes in the t1/t2 plot of the PCA loading and energy transfer were found at the
model for Recovery Boiler A. higher 2nd and 4th PC’s, respectively.
7-5
Figure 10. “Low-fouling” and “high-fouling”
Figure 9. “Low-fouling” and “high-fouling”
periods for Recovery Boiler B
periods for Recovery Boiler A.
In Figure 10, grey circles again represent days
In Figure 9, grey circles represent days
corresponding to the “low-fouling” period, and
corresponding to the “low-fouling” period, and
black circles represent days corresponding to the
black circles represent days corresponding to the
“high-fouling” period. Data points corresponding to
“high-fouling” period. The data points clearly
the “high-fouling” period were primarily situated at
cluster into two distinct regions. Data points
the bottom of the plot while “low-fouling” data
corresponding to the “high-fouling” period are
points were situated almost entirely in the upper
primarily situated at the bottom of the t1/t2 plot
part of the plot. The two regions are once again
while “low-fouling” data points were situated
distinct in the y-axis corresponding to the second
almost entirely in the upper part of the t1/t2 plot. The
principal component. The data from these two
overall results agree well with the interpretation of
regions was selected and their averages were
the principal components. There is little overlap
compared directly as shown in Table 7 in the
between the two periods in the second principal
Appendix. The results are interesting as several
component suggesting that they were very different
variables stand out. Black liquor flow was actually
in terms of energy transfer efficiency. The averages
higher on average, during the “low-fouling” period
of the points in the two clusters can be compared
as was sootblowing steam flow. White liquor
directly, the results of which are shown in Table 6
sulfidity was also elevated during the “high-
in the Appendix.
fouling” period.
Since Recovery Boiler B has experienced consistent
For the high resolution data available for Recovery
fouling problems, it was not practical to break up
Boiler C, individual runs were compared based on
and compare sections of the dataset in terms of the
run length. A 54 day “high-fouling” period was
number of waterwashes. Instead, “low-fouling” and
chosen from 7/21/2005 to 9/12/2005 which began
“high-fouling” periods were selected by comparing
with a boiler cleaning and ended with a pluggage.
fouling indicator variables such as the boiler bank
This was compared to a 76 day, “low-fouling”
outlet temperature. There are clear periods when the
period from 7/19/2006 to 10/3/2006. The data
boiler bank outlet temperature rose sharply,
points were displayed over Figure 9 with all other
indicating that fouling happened very quickly, and
data points removed, and the result is shown in
there are periods where this is not as much of a
Figure 11.
problem. From this, two regions were then selected
that corresponded to a “low-fouling” period, from
10/3/2003 to 3/14/2004, and a “high-fouling”
period combining 3/5/2005 to 5/4/20005 and
7/13/2005 to 9/5/2005. The data points were
displayed over Figure 7 with all other data points
removed. The result is shown in Figure 10.
7-6
Figure 12. Points corresponding to operation from
Figure 11. “Low-fouling” and “High-fouling” July 28-August 24 for Recovery Boiler A
periods of hourly average data for Recovery Boiler
C.
7-7
TABLE 3. INPUTS AND OUTPUTS FOR PLS liquor droplets both of which tend to decrease
Inputs Outputs carryover. Similarly, white liquor sulfidity, when
Liquor Firing Load Steam Flowrate increased, showed corresponding decreases in
Liquor Solids Content Steam Temperature energy transfer related variables and increases in
Liquor Temperature Steam Pressure boiler bank outlet temperature and ID fan speed
Liquor Pressure Steam Enthalpy which indicates decreased energy transfer
Liquor Composition Feedwater Flowrate efficiency. Also interesting were the sootblowing
Air Flowrates Feedwater Temperature steam flow variables. The north sootblowing steam
Air Temperatures Boiler Bank Gas Pressure flow had positive correlations with both final steam
Sootblowing Steam Flow Economizer Gas Pressure flow rate and the steam enthalpy, while the south
Boiler Bank Gas sootblowing steam flow variable had no consistent
Excess O2 correlation with any of the outputs. This supports an
Temperatures
observation by mill personnel that deposits may be
Economizer
TRS Concentration forming faster on the north side of the boiler.
Temperatures
Consistent correlations were not found for black
SO2 Concentration ID Fan Speed
liquor chloride or sodium contents. This does not
necessarily mean that these were not important
For Recovery Boiler A, seven PC’s were calculated
variables, but rather, this may be due to limited
through cross-validation accounting for 70% of the
liquor composition data available for the analysis.
variability in the data. Final steam flow rate, steam
enthalpy, boiler bank outlet temperature, and ID fan
The correlation structure from the PLS analysis on
speed were well modeled output variables, where
Recovery Boiler B is shown in Table 5. The
more than half of their variability (Q2>0.5) could
correlations between many of the operational
be predicted based on the input variables. The
variables behave as expected. An increase in white
resulting correlation structure derived from the
liquor sulfidity corresponded to an increase in
model coefficients is shown in Table 4.
boiler bank outlet temperature, differential draft
pressure, and ID fan speed indicating that the boiler
In Table 4, the number of indicators (plus or minus
had a tendency to operate during these periods with
signs) infers the strength of the correlation. Absent
lower energy transfer efficiency. Likewise, an
indicators imply that the model was unable to
increase in sootblowing steam flow corresponded
extract a consistent correlation between the
with an increase in final steam flow rate and the
variables. Shaded regions represent areas of
steam enthalpy. White liquor sulfidity and
particular interest. Most of the variables in the
sootblowing steam flow were also the variables that
correlation structure in Table 4 support expected
were high and low respectively in the “Low
operational conditions. Black liquor flow rate, for
Fouling” period in the corresponding PCA analysis.
example, had the strongest positive correlation with
all the outputs indicating that when black liquor
The correlation structure from the PLS analysis on
flow increased, the output variables, final steam
Recovery Boiler C is shown in Table 6. The results
flow rate, steam enthalpy, boiler bank outlet
from Table 6 shows that black liquor flow rate
temperature, and ID fan speed, also increased. Air
showed strong positive correlations with both final
flow properties and excess O2 also showed
steam flow rate and steam enthalpy, but only a
operational adjustments reflecting increased or
weak positive correlation with the boiler bank outlet
decreased black liquor flow.
temperature. Black liquor flow increases the heat
available to the boiler and therefore should raise the
Black liquor solids, however, showed strong
boiler bank outlet temperature. One would then
positive correlations with both final steam flow rate
expect, however, a much stronger correlation
and steam enthalpy, but a negative correlation with
between black liquor flow rate and boiler bank
the boiler bank outlet temperature. This indicates
outlet temperatures if boiler loading was the main
that during periods of high black liquor solids, the
problem behind deposit buildup.
boiler had a tendency to operate with higher energy
transfer efficiency. Firing high black liquor solids
tends to stabilize the char bed and produce larger
7-8
TABLE 4. PLS CORRELATIONS FROM RECOVERY BOILER A
Final Steam Flow Boiler Bank Outlet
Rate Steam Enthalpy Temperature (Q2 ID Fan
(Q2 = 0.9) (Q2 = 0.91) = 0.65) (Q2 = 0.78)
Black Liquor Flow rate +++++ +++++ +++++ +++++
Black Liquor Solids ++++ +++ --
Black Liquor Temperature
Black Liquor Pressure -- -- ---
Black Liquor Cl Content
Total Air Flow ++++ ++++ +++ ++++
Primary Air Flow -- -- --
Secondary Air Flow +++ ++ ++
Tertiary Air Flow +++++ +++++ ++++ +++++
Primary Air Temp - --
Secondary Air Temp ++++
Tertiary Air Temp +
Sootblowing Steam Flow +++ ++++
Sootblowing Steam Flow
Excess O2 ----- ----- ---- ---
TRS + + ++ +++
SO2 -- -- +++
Green Liquor Flow + + ++ +++
White Liquor Sulfidity -- -- ++++ ++
Saltcake Cl -- --
7-9
TABLE 6. PLS CORRELATIONS FROM RECOVERY BOILER C
Final Steam Boiler Bank Outlet Boiler Bank Outlet
Steam Enthalpy West ID Fan East ID Fan
Flow Temp Press.
Q2=0.77 Q2=0.62 Q2=0.54
Q2=0.73 Q2=0.57 Q2=0.67
Black Liquor Flow Rate +++ +++ + - ++++ ++
Black Liquor Solids + + + + - -
Black Liquor Mass Flow Rate +++ ++++ + - ++++ ++
Black Liquor Temperature -- -- ++++ ++ ++
Black Liquor Pressure ++ ++ --- +++ +
Primary Air Flow +++ +++ ---- --- +++ -
Secondary Air Flow +++++ +++++ ----- +++++ +++++
Tertiary Air Flow +++++ ++++ ++++ +++ ++++ +++++
Primary Air Temp -- +++ +++++ --- ++++
Secondary Air Temp ++ - ---- ++
Sootblowing Steam Flow ++ ++++ + -- +++ ---
Excess O2 ----- ----- --- -- --- ---
TRS -
White Liquor Sulfidity - ++ ----
Opacity ++ + + + --- ---
Attemperator Flow ----- ++ --- -----
7-10
Deposits” Pulp and Paper Canada, Vol. 84, No. 1, Industrial Fuidized-bed Reactor” Control
T7-T12 (1983) Engineering Practice, Vol 8, 893-909 (2000)
[2] BACKMAN R., HUPA M., UPPSTU E. [7] YIN K. K., YANG H., CRAMER F. “On-line
“Fouling and Corrosion Mechanisms in Recovery Monitoring of Papermaking Processes” Chemical
Boiler Superheater Area” TAPPI Journal, Vol 70, Engineering Communications, Vol 189, 1242-1261
No. 6, 123-127 (1987) (2002)
[3] HUPA M., BACKMAN R., SKRIFVARS B.J., [8] KRESTA J.V., MACGREGOR J.F., MARLIN
HYOTY P. “The Influence of Chloride on the T.E. “Multivariate Statistical Monitoring of Process
Fireside Behavior in the Recovery boiler” TAPPI Operating Performance” The Canadian Journal of
Journal Vol 73, No. 6, 153-158 (1990) Chemical Engineering, Vol 69, 35-47 (1991)
[4] BARSIN J. “Recovery Boiler Sootblowers” [9] KOURT T. “Application of Latent Variable
TAPPI kraft recovery operations short course, Methods to Process Control and Multivariate
TAPPI PRESS 219- 227 (1992) Statistical Process Control in Industry”
International Journal of Adaptive Control and
[5] HODOUIN D., MACGREGOR J.F., HOU M., Signal Processing, Vol 19, 213–246 (2005)
FRANKLIN M. “Multivariate Statistical Analysis
of Mineral Processing Plant Data”, CIM [10] SHI R., MACGREGOR J.F. “Modeling of
Bulletin, Vol. 86, No. 975, 23-34 (1993) Dynamic Systems using Latent Variable and
Subspace Methods” Journal of Chemometrics, Vol
[6] SIMOGLOU A., MARTIN E.B., MORRIS A.J. 14, 423-439 (2000)
“Multivariate Statistical Process Control of an
APPENDIX
TABLE 6. COMPARISON OF THE AVERAGE VALUES OF THE “LOW FOULING” AND THE “HIGH
FOULING” PERIODS FOR RECOVERY BOILER A
Average of “Low-Fouling” Period Average of “High-Fouling” Period
BL Flow rate (USGPM) 303 307
BL solids A (%) 70.2 69.1
BL Temperature (F) 254 253
BL Pressure (psi) 23.5 24.4
BL Cl (% d.s.) 0.9 0.9
Air Flow Primary (%) 35 36
Air Flow Secondary (%) 43 43
Air Flow Tertiary (%) 21 21
Air Flow Total (1000 lb/hour) 693 732
Final Steam Flow Rate (1000 lb/hour) 558 520
Final Steam Temperature (F) 844 826
Final Steam Pressure (psi) 870 873
Sootblowing Steam Flow 1 (1000 lb/hr) 22.0 20.2
Sootblowing Steam Flow 2 (1000 lb/hr) 24.2 23.3
Boiler Bank Inlet Temperature 1 (F) 876 944
Boiler Bank Outlet Temperature A (F) 684 712
Excess O2 (%) 3.0 3.0
TRS (ppm) 1.0 1.2
SO2 (ppm) 1.2 5.7
WL sulfidity (% on TA) 25.1 27.4
ID Fan Speed (% of Maximum) 66 65.8
Steam Enthalpy (TJ/day) 21.7 20.1
7-11
TABLE 7. COMPARISON OF THE AVERAGE VALUES OF THE “LOWER FOULING” AND THE “HIGHER
FOULING” PERIODS FOR RECOVERY BOILER B
Average of “Low-Fouling” Period Average of “High-Fouling” Period
BL Flow rate (USGPM) 454 451
BL solids (%) 71.0 71.8
BL Temperature (F) 247 245
BL Pressure (psi) 46.4 46.5
Air Flow Primary (%) 36 36
Air Flow Secondary (%) 42 41
Air Flow Tertiary (%) 22 22
Air Flow Total (1000 lb/hour) 953 961
Final Steam Flow Rate (1000 lb/hour) 698 697
Final Steam Temperature (F) 847 821
Sootblowing Steam Flow (1000 lb/hr) 41.6 38.6
Boiler Bank North Inlet Temperature (F) 1077 1151
Boiler Bank North Outlet Temperature (F) 743 767
Excess O2 (%) 2.7 2.0
TRS (ppm) 0.4 0.4
SO2 (ppm) 1.8 8.9
WL sulfidity (% on TA) 25.1 28.4
North ID Fan (% of Maximum) 69 72.1
Steam Enthalpy (TJ/day) 26.9 26.5
TABLE 8. COMPARISON OF THE AVERAGE VALUES OF THE “LOWER FOULING” AND THE “HIGHER
FOULING” PERIODS FOR RECOVERY BOILER C
Average of “Low-Fouling” Period Average of “High-Fouling” Period
BL Flow rate (USGPM) 361 389
BL solids (%) 69.5 68.6
BL Temperature (F) 244 238
BL Pressure (psi) 16.8 16.3
Air Flow Primary (%) 37.8 41.4
Air Flow Secondary (%) 49.4 62.2
Air Flow Tertiary (%) 50.8 24.6
Final Steam Flow Rate (1000 lb/hour) 578 611
Final Steam Temperature (F) 837 872
Final Steam Pressure (psi) 1,434 1436
Sootblowing Steam Flow (1000 lb/hr) 50.1 50.0
Boiler Bank Outlet Temperature (F) 862 771
Excess O2 (%) 2.6 3.0
TRS (ppm) 0.8 0.8
WL sulfidity (% on AA) 26.4 25.5
West ID Fan Speed (RPM) 516 618
Steam Enthalpy (TJ/day) 22.0 23.7
7-12
“Point B” “Point D”
Figure 14. Comparison between Point B (August 22, 2005-High Energy Transfer Efficiency) and Point D (August 24,
2005-Low Energy Transfer Efficiency) for Recovery Boiler A
7-13