Professional Documents
Culture Documents
Stroke Fire Boiler282872873782873
Stroke Fire Boiler282872873782873
net/publication/233542008
CITATIONS READS
2 945
4 authors, including:
Alex Chong
University of Nottingham
36 PUBLICATIONS 241 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by S. M. Thai on 20 March 2017.
non-linear functional mappings and adaptive capabil- a slight suction in the furnace chamber to maintain a
ities.5,6 As a result, ANNs are capable of empirically suitable pressure drop across the boiler. Hot water was
modelling the combustion process in a utility boiler as a generated at a temperature of 120uC and a pressure of
function of parameters, such as the coal feed and airflow 3 bar by the 3-pass shell boiler. Automatic ash removal
rates and to continually learn in an online mode and to and a high efficiency cyclone to arrest grit before the flue
modify set-points and bias adjustments to keep the unit gas was discharged to the atmosphere through the stack
tuned to the desired emission and performance levels.7 were other features of the stokers. Each of these coal
In this case, a non-linear ‘black box’ model is a fired stoker boilers could produce up to 3?7 MW of
promising method to symbolise the expert operator’s thermal capacity and possessed a turndown ratio of 6 : 1.
knowledge based on ANNs. Besides, as it is known that In other words, the lowest operable load was 0?62 MW,
neural networks are very useful for non-linear system which corresponded to 16?67% of the maximum
modelling, the use of neural networks to construct non- continuous boiler output rating (MCR). The overall
linear predictors may be promising for non-linear combustion efficiency of this boiler was claimed to be
process control.8 This is evident from earlier literature 80%, a typical figure for a boiler of this type. The main
which asserted that neural networks have been proven to purpose of this 24 hourly manned boiler house is to
be very useful for non-linear system modelling and supply hot water for domestic usage and space heating
control.9,10 In the case of the stoker boiler, it is necessary for approximately 500 inmates in the prison.
to model the relationship between the carbon monoxide The coal fired stoker boiler was fitted with an
(CO) and oxygen concentration (O2) with the coal feed industrial PID digital control system, Honeywell
and airflow rates, and to develop a controller that UDC3000, which regulated the coal feed and airflow
determines the parameter settings to minimise pollutant rate (so that the desired hot water temperature set point
emissions and maximise efficiency over a wide range of could be achieved) and the whole unit was integrated
operating conditions. Thus, the neural model should be into an overall energy management system. The process
capable of analysing data from various sensors and, sensor available from the plant was a thermocouple
once trained should be able to recognise if the measuring the hot water temperature from the outgoing
combustion process is suboptimal. Backpropagation hot water pipe. However, the existing plant conventional
for the training of the feed forward multilayered PID controller was a single loop process controller with
perceptron (MLP) network is the most widely used the process variable being the pressurised hot water
learning algorithm and the most often considered temperature, which indicated the load demand required.
member of the neural network family, was implemented Individual signals to the various actuators, i.e. rotary
in this work and has been proven through a large valve, grate and damper motor, and variable speed
number of other practical applications.11 drives was ratioed off the PID controller signal.12
The main objective of this work was to control a chain For the purpose of collecting information regarding
grate stoker fired boiler by the development of a neural the behaviour and response of the plant, a stand-alone
network based controller (NNBC) which aimed to data logging system was developed so that the input
increase the overall thermal efficiency and maintain variable (firing rate) and the required process variable
pollutant emissions below future medium term legisla- (hot water temperature) could automatically be
tion during fluctuating load following conditions. At acquired. For firing rate acquisition, the input signal
present, it is usual industrial practise for the efficiency to was gathered and passed through a fifth order
be sacrificed by allowing the stoker to operate at a Butterworth lowpass filter in order to reduce noise
higher excess air to ensure complete burn out of the coal. before being recorded by the logging system. As for the
Therefore, one way to help maximise boiler efficiency hot water temperature measurement, a K-type thermo-
and minimise pollutant emissions is to maintain the couple was used. For automatic logging of these two
amount of excess air at an acceptable level consistent filtered analogue signals, a Pentium 3 550 MHz PC
with satisfactory carbon burn out for the entire firing fitted with the MIO16 data acquisition card with
range. As a result, a series of experiments have been LabVIEW software was used. The data acquisition
conducted to cover a wide range of boiler conditions and sampling period was 30 seconds. Other measurements
the gathered data were used to train the neural network taken from these tests included the pollutant emissions
components of the controller. Data gathered from these (CO and NOx normalised to 6% of oxygen), concentra-
experiments included pollutant emissions (NOx and tion of oxygen in the flue gas on a dry basis and flue gas
CO), the oxygen in the flue gas on a dry basis and the temperature which was measured by a ‘Testo 33’
load condition (coal feed and airflow rate). portable flue gas analyser at a 30 s sampling interval,
fitted at the exhaust duct before the cyclone.
Chain grate stoker fired boiler
The three 3?7 MW chain grate stoker fired shell boilers
Control approach
were situated in a boiler house featuring Proctor’s fire Figure 1 shows a schematic diagram of the NNBC.
break unit; pneumatic coal feeding system; fire bed area Fundamentally, the NNBC identified the load condition
of 1?2 m (width of grate)63?5 m (length of fire bed); the and through the setting network, based on rules that
primary combustion air was provided via a variable were constructed using information drawn from the
speed forced draught fan motor in order to achieve commissioning tests, the corresponding optimum set-
better air distribution over the entire firing range. The tings of coal feed and airflow were delivered to match
air distribution was controlled by a damper which was the load required. Once the coal feed and airflow were
positioned half way along the length of the grate inside brought to the new settings, the second part of the
the windbox and spread the air into the two sections on NNBC was implemented to fine tune the primary
the grate. An induced draught fan was fitted to maintain airflow for the purpose of optimising the combustion
type was dependent on the firing rate (Table 1), with the
NNBC being implemented in MATLAB.
1 Schematic of overall NNBC for chain grate stoker fired
shell boiler
Results
Five combustion optimisation tests whilst burning Bean
process on the grate. This AI based ‘oxygen trimming’ Coal were carried out with the boiler controlled by the
altered the primary air in order to ensure that the excess NNBC, where the plant response under the influence of
air level fell within the oxygen target band and the CO the NNBC was compared with and without the neural
emissions were within an acceptable limit. Instead of models. During these tests, a constant firing rate of 40
using the actual concentration of oxygen and CO and 70% of MCR was used for validation purposes and
emission from the flue gas analyser, the predicted future a further test at 60% of MCR was also conducted to test
values produced by the neural models were used. the ability of the neural network to cope with new
The oxygen and CO neural models were constructed conditions. The sampling period and control action was
using a feed forward MLP network architecture with an set to a 3-minute interval, which allowed the NNBC to
error back propagation learning algorithm together with take action one control period ahead, which should
improve the response of the controller. At the beginning
an appropriate set of regressor vectors. The auto
of each test, the airflow was manually adjusted to a
regressive with eXogenous inputs (ARX) regressor
higher or lower excess air (depending on the test profile)
structure was selected. The trained models were inte-
and the NNBC was activated to adjust the airflow
grated into the overall control strategy to optimise the
(Table 1).
industrial stoker boiler.
During the tests, constant loads of 40, 60 and 70% of Results obtained at constant load of 40% of
MCR were considered as input parameters to the setting MCR – Tests 1 and 2
network. The network was then requested to provide the The results obtained by operating the boiler at a
corresponding firing rate with an appropriate air/fuel constant load of 40% of MCR with and without the
ratio, meanwhile the present and past values of coal feed O2 neural model are shown in Fig. 2a and 2b
and airflow together with the current and previous respectively. The NNBC altered the airflow over a range
oxygen and CO readings, were fed as input parameters of 35 to 60% (Fig. 2a) with the O2 neural model as
to the neural models predicting oxygen and CO 3 steps compared to 25% to almost 90% without the neural
into the future. model (Fig. 2b). Consequently, the oxygen concentra-
The estimated oxygen and CO were then fed to the tion was inside the target band for more of Test 1 than
adjusting network for appropriate airflow adjustment to Test 2. The O2 model predicted reasonably well, thus
consistently maintain an optimum air/fuel ratio. The narrowing the range of airflow adjustment (Test 1). The
NNBC was configured so that airflow adjustment would reason for the fluctuations in the oxygen level can be
only be activated if the oxygen level fell outside the partly attributed to the difficulty in controlling stokers
target band with any action being based on the predicted at low fire, normally a PID controller will operate at a
values. The target oxygen band for this particular coal higher excess air to minimise this effect.
a with oxygen neural model (Test 1); b without oxygen neural model (Test 2)
2 Boiler performance with NNBC following combustion optimisation test at constant load of 40% of MCR with and with-
out O2 neural model
Figure 3a and b shows the CO and NOx emissions without the O2 neural model. The small alteration of the
recorded during the combustion optimisation test at 40% airflow taken by the NNBC with the O2 neural model
of MCR with and without the CO neural model. For the resulted in stable combustion on the bed (Fig. 4a). After
test conducted with the CO neural model (Fig. 3a), 20 min, the oxygen level was brought into the target
although the prediction did not coincide with the actual band of 9 to 11% from the preliminary low excess air.
CO for the entire test, the CO emissions were well This result was encouraging, as the NNBC without the
controlled by the NNBC and settled below the target limit neural model took more time (35 min) to reach the
of 1000 ppm with an average of 128 ppm (normalised to target band. Consequently, for 66% of the total test
6% of oxygen). This was due to the model exhibiting the duration the oxygen concentration lay within the target
same trends as the recorded CO. For the results obtained band (Test 3) as compared to only 48% attained without
without the CO neural model, an overshoot of CO of the neural model (Test 4).
more than 1000 ppm can be observed in Fig. 3b, although Another interesting point in Test 3 is that after 1?5 h
the CO level for the rest of this test was below the limit into the experiment, there was a sudden increase in the
value with an average emission of 351 ppm. actual oxygen concentration (perhaps caused by a poor
NOx emissions (Fig. 3a) show no concern probably coal distribution on the grate) from 10?5 to 12%
being helped by the natural form of air staging along the (Fig. 4a). However, the oxygen concentration predicted
fire bed, with an average emission of 161 ppm being by the neural model indicates that the oxygen level
recorded. Unfortunately, the NOx reading was not increased only slightly above the target band and hence
available for Test 2 (Fig. 3b) due to a failure of the flue only a small amount of airflow reduction was made by
gas analyser. Although there are no emission restrictions the NNBC. This decision proved to be appropriate as
for a boiler of this range, their emission limits will pro- the actual oxygen dropped to be just above the lower
bably have to comply in the future with those of the 20 to band a few minutes later (coincident with the suction
50 MW range of coal fired boilers, i.e. NOx,245 ppm caused by the automatic ash removal system) and
and SO2,700 ppm at 6% oxygen.3 It would appear that continued to stay in the middle of the band until the
the smaller airflow adjustment range, from the use of the end of the test. This situation could not have been dealt
neural model generates a more stable combustion process with by the NNBC without the neural model, as there
on the bed thus producing more satisfactory combustion. would simply be a decrease in the excess air in
proportion to the unexpected rise in the oxygen level.
Results obtained at constant load of 70% of The consequences of a sudden increase or decrease of
MCR – Tests 3 and 4 excess air can result in unstable combustion on the bed
Figure 4a and b shows the results obtained by operating with a great possibility of the controller and the process
the boiler at a constant load of 70% of MCR with and becoming unstable.
a with oxygen neural model (Test 3); b without oxygen neural model (Test 4)
4 Boiler performance with NNBC following combustion optimisation test at constant load of 70% of MCR with and with-
out O2 neural model
The CO and NOx emissions attained by operating the airflow alteration, smaller adjustments were once again
boiler at a constant load of 70% of MCR with and imposed by the NNBC when using the O2 neural model
without the CO neural model are shown in Fig. 5a and b to fine tune the excess air. This leads to the oxygen
respectively. Although no control action was taken by concentration remaining in the target band for 52% of
the NNBC to fine tune the airflow based on the the test. The reason the actual oxygen was not reduced
predicted CO emissions (as they were all well below was because the predictions made by the O2 neural
the target limit), the prediction (Fig. 5a) was highly model were within the target band and therefore, no
comparable to the actual CO recorded by the flue gas action was taken by the NNBC to trim the oxygen.
analyser. The average CO obtained over the test was Figure 6b shows the results obtained following the
353 ppm (Test 3). On the other hand, the CO shown in combustion optimisation test at a constant load of 60%
Fig. 5b (Test 4) was also well below the target limit with of MCR for CO and NOx. Although the prediction in
an average emission of 231 ppm being recorded the first half was not in good agreement with the actual
(normalised to 6% oxygen). results, the second half was very close to the actual
NOx emissions closely followed the firing rate as the readings taken by the flue gas analyser. However, the
emission was mainly dependent on the fuel and the control action imposed by the NNBC integrated with
availability of combustion air. This is evident where the CO neural model maintained the overall CO
the NOx emission was increased to 300 ppm from emission below the target limit of 1000 ppm (normalised
200 ppm as the airflow was increased from 50 to 90% to 6% oxygen). The NOx emissions poses no cause for
of MCR (Fig. 5a). concern with a mean of 266 ppm being recorded.