Temp (C)
>1609
>1588
>1566
>1545
>1523
>1501
>1480
>1458
>1437
>1415 a b c da b c da b c d
BF2RA Project:
Intelligent Flame Detection Incorporating Burner
Condition Monitoring and On-Line Fuel Tracking
1
Duo Sun, Professor Yong Yan, Dr Gang Lu
Industrial Supervisor: Kelly Walker, E.ON
Outline
Aim and Objectives
Current Progress
System Design & Implementation
System Initial Tests
Test on a 660MWth Boiler
Test on a 9MWth Oil CTF
Intelligent Burner Condition Monitoring
Intelligent On-line Fuel Tracking
Future Work
2
Aim and Objectives
• Aim
to develop a cutting edge flame monitoring technology that
can indicate burner conditions and track the type of fuels
(coal and/or biomass) in a power plant.
• Objectives
− To develop a technology for flame stability measurement, burner
condition monitoring and on-line fuel tracking through digital imaging
and flame signature analysis;
− To evaluate the technology under a range of biomass firing,
coal/biomass co-firing, and oxy-fuel fired conditions on a combustion
test facility and on a full scale multi-burner furnace;
− To make recommendations for improvements of existing furnaces
through the use of the new technology.
3
Power plant
Flames inside a boiler
4
Why Monitor Flame?
• The characteristics of a flame, such as size, shape, brightness, colour,
oscillation frequency and temperature, provide instantaneous information
on the performance of the combustion process.
• Existing flame monitoring instruments can only measure global variables
(input/output of boiler), which provide very limited description about the
flame (inside of boiler).
• System Strategy
• Geometric (Ignition point, size and shape)
• Luminous (brightness, non-uniformity)
• Oscillation frequency , temperature distribution
5
Flame
Beam
splitting
unit
Photo-detectors & Signal-
processing board
for oscillation frequency
Optical
probe/
optical
fibre
Photo
diodes
UV
Visible
RGB
digital
camera
Geometric/luminous
parameters &
temperature
R
G
B
Combustion
monitoring and
diagnosis
Data
presentation
IR
Embedded motherboard
Flame parameters measured
Current Progress - Flame Monitoring System
Embedded
motherboard
Ethernet (to main PC)
• Optical probe based system
6
Current Progress - Flame Monitoring System
Imaging and Data
Processing Unit Optical probe
& cooling jacket
Embedded Photo-
detectors & Signal-
processing Board
• Robust
• Compact
• Fast response
• Acceptable cost
Meet industrial requirements
Water in/out Air in
• Optical fibre based system
Air-cooled jacket with a thermocouple
Imaging fibre bundle and camera system
Air-cooled jacket
Thermocouple
3m long and 50.8mm outer diameter
7
Current Progress - Flame Monitoring System
8
Graphical user interface
• Compute and display on-line flame sensorial information (flame image, 2-D
temperature distribution, flame radiation signals, power spectral densities, etc.)
• Setup system (camera exposure time, gain control, etc.)
• Record data (raw images/signals, processed results)
• System Software
Current Progress - Flame Monitoring System
9
• The oscillation frequency measurement was evaluated using a
standard frequency-varying light source.
• The relative error is no greater than 2% (0 to 500Hz).
0 100 200 300 400 500 0
100
200
300
400
500
Reference frequency (Hz)
Measure
d fre
quency (
Hz)
Comparison between the measured and reference frequencies
Current Progress - System Evaluation
10
• The system was calibrated using a blackbody furnace, and verified
by measuring the true temperature of a standard tungsten lamp.
• The max error of 14.8°C occurs at 1650°C (equivalent to the relative
error of 0.9%.
800 1000 1200 1400 1600 1800 800
1000
1200
1400
1600
1800
Reference temperature (°C)
Me
asure
d te
mp
era
ture
(°C
)
Current Progress - System Evaluation
Comparison between the measured and reference temperatures
11
• The system has been tested on
• a 660MWth coal/biomass-fired boiler at a power station in UK
• a 9MWth heavy-oil-fired CTF at Zhejiang University, Hangzhou, China
Field Trials
9MWth CTF at Zhejiang University
A power station in UK
Boiler cross-section and system installation
Test on a 660MWth Boiler
Burner A (biomass)
Burner C (coal)
Burner B (biomass)
Imaging
system Spectrometer
Imaging
system
Imaging
system
Spectrometer
Spectrometer
12
Images of Biomass & Coal flames
Burner A (100% biomass)
Burner B (100% biomass) Burner C (100% coal)
Note: - Location
of the burner.
13
Burner A (feeder off,
burner firing supported by oil)
1200°C
1700°C
1500°C
1400°C
1300°C
1900°C
1800°C
1600°C
Flame Temperature Distribution
Note: - Location
of the burner.
14
Burner A (100% biomass)
Burner B (100% biomass) Burner C (100% coal)
Burner A (feeder off,
burner firing supported by oil)
Temperature histogram Average temperature
Temperature Variation
Burner A
(biomass)
Burner A
(biomass-off)
Burner B
(biomass)
Burner C
(coal)
1200
1300
1400
1500
1600
1700
1800
Avera
ge tem
pera
ture
(°C
)
1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 0
5
10
15
20
25
Temperature (°C)
Norm
aliz
ed f
lam
e a
rea (
%)
Burner A (biomass)
Burner A (biomass-off)
Burner B (biomass)
Burner C (coal)
15
Remarks:
• The biomass flame has a substantial delay in ignition and fluctuates
significantly.
• Increased standard deviations of the flame temperature and luminous region
were found under all the biomass conditions, indicating greater instability of the
biomass flames.
Luminous Region and Brightness
Burner A
(biomass)
Burner A
(biomass-off)
Burner B
(biomass)
Burner C
(coal)
10
15
20
25
30
35
40
Lum
inous r
egio
n (
%)
Burner A
(biomass)
Burner A
(biomass-off)
Burner B
(biomass)
Burner C
(coal)
0
10
20
30
40
50
60
70
80
Brightn
ess (
%)
Luminous region Brightness
16
Remarks:
• The biomass flame has relatively unstable ignition in nature, compared with
coal flame.
• The significant fluctuation in the ignition of the biomass flames may affect the
reliable operation of the flame eye. This is due the fact that the field of view of
the existing flame-eye is very narrow,
100
101
102
103
0
0.05
0.10
0.15
Frequency (Hz)
Norm
aliz
ed a
mplit
ude
Burner A (biomass)
Burner A (biomass-off)
Burner B (biomass)
Burner C (coal)
100
101
102
103
0
0.05
0.10
0.15
Frequency (Hz)
Norm
aliz
ed a
mplit
ude
Burner A (biomass)
Burner A (biomass-off)
Burner B (biomass)
Burner C (coal)
In infrared band In visible band
Frequency Spectrum
17
Remarks:
The biomass flames exhibit very different profiles of frequency spectrum from
that of the coal flame.
Oscillation Frequency
Burner A
(biomass)
Burner A
(biomass-off)
Burner B
(biomass)
Burner C
(coal)
0
5
10
15
Oscill
ation f
requency (
Hz)
Burner A
(biomass)
Burner A
(biomass-off)
Burner B
(biomass)
Burner C
(coal)
0
5
10
15
Oscill
ation f
requency (
Hz)
18
Remarks:
• The amplitude of the low-frequency components for the biomass flame is much
higher than that of the coal flame in both visible and infrared bands, resulting in
much lower oscillation frequencies.
• The different frequency spectral characteristics of the biomass flame may
explain why the exiting flame eye, which is specially designed for coal flames,
fails when it is applied to a biomass flame.
Infrared band Visible band
Spectroscopic Characteristics
19
Biomass flames
Coal flame
Remarks:
Flames from all biomass-fired burners tested have shown very similar
spectroscopic profiles, which are significantly different from that of a pulverised
coal flame. However, the spectroscopic intensity of the biomass flame vary
significantly even under the same condition
20
Test facility
Test conditions
• Variations in the swirl vane angle of tertiary air
• Variations in the swirl vane position of secondary air
• Variations in the ratio of primary air to total air
• Variations in the ration of overfire air to total air and its nozzle position
Tests on the 9MWth Heavy-Oil-Fired CTF
Intelligent Burner Condition Monitoring
21
• Flame images
• Luminous region
• Brightness
• Non-Uniformity
• Flame radiation signals
• Oscillation frequency
• DC, AC, skewness and kurtosis
• Energy distribution
Flame parameters
• Use flame parameters as the “signature” of a particular
combustion condition
Abnormal condition
detection
NOx prediction
Flame state
identification
SVM
KPCA: kernel principal components analysis
SVM: support vector machine (classification/regression)
Algorithms
Intelligent Burner Condition Monitoring
Example - Abnormal Condition Detection
22
Remarks:
• Compared with PCA, KPCA model performed better in illustrating the discrepancy
between the normal and abnormal conditions, and showed no false warnings at all.
(b) KPCA
0 10 20 30 40 50 6010
0
101
102
103
Observations
T2-S
tati
stic
False
warning
Abnormal conditionNormal condition
Confidence
limit (95%)
0 10 20 30 40 50 6010
-1
100
101
102
103
ObservationsT
2-S
tati
stic
Confidence
limit (95%)
Abnormal conditionNormal condition
Monitoring charts of combustion process by KPCA
(a) PCA
• An abnormal condition was generated by setting the SA swirl vane
position deviated from its baseline configuration.
Intelligent Burner Condition Monitoring
Example - NOx Predication
23
Remarks:
• For NOx prediction, SVM model exhibits better performance than the tested conventional
NN model, which is due to SVM’s better generalization ability.
• The maximum relative error of the SVM is about 10.22%, much smaller than that of the
NN 23.15%.
Neural Network (NN) Support Vector Machine (SVM)
150 200 250 300 350150
200
250
300
350
NO
x p
red
icte
d b
y N
N (
pp
m)
Measured NOx (ppm)
150 200 250 300 350150
200
250
300
350
NO
x p
red
icte
d b
y S
VM
(p
pm
)
Measured NOx (ppm)
• Comparison between predicted and measured NOx emissions
Intelligent Burner Condition Monitoring
Example - Flame State Identification
24
Remarks:
• For flame state identification, SVM model exhibits not only better but also stable
performance than the tested conventional NN model.
• The increase of success rate with training data size suggests that adequate data should
be collected to represent all the possible patterns of a dynamic process so as to achieve
a more reliable flame state identification.
0 10 20 30 40 50 60 70 80 90 10060
65
70
75
80
85
90
95
100
Training set size (with respect to whole database %)
Su
ccess
rate
of
cla
ssif
icati
on
of
test
data
(%
)
NN
SVM
Variations of the success rate of classification
of testing data with training set size
Intelligent On-line Fuel Tracking
25
• Flame images
• Luminous region
• Brightness
• Non-Uniformity
• Flame radiation signals
• Oscillation frequency
• DC, AC, skewness and kurtosis
• Energy distribution
Flame parameters
• Use flame parameters as the “signature” of a particular fuel
fired flame
Fuel tracking
Pattern Recognition
& Machine Learning algorithms
Challenges in on-line fuel tracking
• Extraction and selection of flame features
• Classifier design and learning for fuel recognition
• Performance evaluation
• It can only be done with field trials
• Neural network
• K-nearest neighbor
• Decision trees
• Bayes classifier
• Hidden Markov model
• …
Coal, biomass, oil
Fuel from different sources
It is hoped the field trials are to be undertaken on
• Ironbridge Power Station (biomass/oil fired),
• Cottam Power Station (biomass/coal fired),
• Wilton Power Station (coal fired), or
• CTF run by Leeds University (oxycoal flames)
• Field Trials
Future Work
26
Concluding Remarks
• An imaging based instrumentation system has been developed for intelligent
flame monitoring, burner condition monitoring and fuel tracking.
• A statistical process control method (KPCA) and pattern recognition method
(SVM) have been applied for intelligent burner condition monitoring.
• The system has been evaluated using a blackbody furnace and a standard
frequency-varying source. -- The relative error of oscillation frequency measurement is no greater than 2% (0-
500Hz).
-- The relative error of temperature measurement is about 0.9% (1000 oC -1650oC).
• The system has been tested on a full-scale coal/biomass fired boiler and an
industrial-scale heavy oil fired combustion test facility.
• The test results have demonstrated the effectiveness and the potential of the
techniques developed for flame characterization and burner condition
monitoring.
Contact Details: School of Engineering and Digital Arts,
University of Kent,
Canterbury,
Kent, CT2 7NT
Telephone: +44 (0)1227 823015
Email: [email protected] or [email protected]
Website: http://www.eda.kent.ac.uk/
The Biomass and Fossil Fuel Research Alliance (BF2RA) for
funding the work. The views expressed are those of the authors
and not necessarily those of the BF2RA.
Acknowledgments
