|
|
AN INITIAL
SURVEY OF PORTMEIRION'S GLOST KILN DATA
Chris Woolgar
Neusciences
29th June 2001
OVERVIEW OF TASK
The data described below is from the final stage of the pottery-making process
after the glaze has been applied and is to be heated in a tunnel kiln. This
work focuses on the control of the temperature in the kiln. Previously the
process was increased in efficiency from c50% to c85% by simply increasing
the work output of fans in the kiln, giving a higher more constant temperature
in the main heating zone in the kiln. The current problem lies in that the
pottery must first be heated gradually, then heated at roughly a constant
temperature, then gradually cooled. The cooling has several important temperature
levels that must not be crossed too quickly. The apparent temperature profile
varies considerably throughout the process. Several factors may affect the
temperature profile, including heights of pottery elements, density of groups
of pottery elements, heat energy supplied by the burners, work done by the
fans, other minor factors or any combination of these.
OBJECTIVES
The long-term goal is to augment or replace the on-line control system for
the kiln, with a system that would take account of current monitoring values,
to adjust plant parameters in such a way as to ensure that operational and
environmental constraints are upheld and high plant efficiency achieved.
Two problems were identified as appropriate ones for tackling in this preliminary
study. The first was to explore the relative importance of the 32 variables,
particularly in relation to those most relevant for modelling. This would
clearly be of value when setting up a control system, to avoid unnecessary
expense and complexity. The second was to explore the extent to which the
inputs could be used to predict targets using neural networks or other related
techniques. Such a predictive model, if developed further, could be used
as the basis for a control system.
To reiterate the originally specified project deliverables this initial
project will provide:
- A study using data collected at Portmeirion into optimal control of
the glost kiln.
- A report that will indicate the relative importance of input variables,
model performance, likely performance of the finished system, and budgets
and timescales for any further modelling work required. In addition,
budgets and timescales for implementation of the control system based
on that model will be included.
THE DATA
Two data files were provided as Excel spreadsheets. The first of these (LOGRPT03.xls)
consists of online measurements of plant operating characteristics, as per
the following table:
| COLUMN
HEADER |
COMMENTS |
| Time
|
Time
stamp of measurement. |
| Temp
1-3, 10-15 & 18 |
Thermocouple
temperatures at various positions through the kiln. The indices start
at the entry point of the kiln, 1 being nearest the entrance. |
| R.Setpoint
(4-9) |
Setpoint
temperature for zones (1-6) of the kiln. These zones are fitted with
burners to allow control of temperature. Although not numbered in
the original file the indices 4-9 have been applied to "R.Setpoint",
"Measured "and "V.Position". These are in sequence with Temp above.
|
| Measured
(4-9) |
Thermocouple
temperatures for zones (1-6) of the kiln. |
| V.Position
(4-9) |
Position
of the gas supply valve for the burners of zones (1-6) of the kiln.
|
| Pressure
|
Pressure |
| U.Car
|
Under
car temperatures, to be ignored on the advice of Ceram. |
There are 32 different variables in total and c700 rows for each covering
a period from 00:01 to 23:26 on 4th March 2001, each indexed with a date
and time stamp. The sampling frequency is 30 per hour.
A basic schematic of the kiln was provided together with the following thermocouple
position data. Every push the cars move one car width, i.e. 1.16m. To allow
comparison of car weight with temperature, the time when a given car is
in a given position is calculated by the time taken to carry out the number
of pushes needed to move it to that position.
With temperature, the time when a given car is in a given position is calculated
by the time taken to carry out the number of pushes needed to move it to
that position.
| THERMOCOUPLE
|
DISTANCE
FROM ENTRANCE / mm |
DISTANCE
FROM ENTRANCE / CARS |
DESCRIPTION
|
| 1
|
1987.5
|
1.7
|
Recirculation
Duct
|
| 2
|
3975.0
|
3.4
|
Exhaust
- taken from both 2&3
|
| 3
|
7402.5
|
6.4
|
Point
in preheat
|
| 4
|
9557.5
|
8.2
|
Zone
1
|
| 5
|
13425.0
|
11.6
|
Zone
2
|
| 6
|
15407.5
|
13.3
|
Zone
3
|
| 7
|
16837.5
|
14.5
|
Zone
4
|
| 8
|
19462.5
|
16.8
|
Zone
5
|
| 9
|
21362.5
|
18.4
|
Zone
6
|
| 10
|
23257.5
|
20.0
|
Cooling
10
|
| 11
|
25240.0
|
21.8
|
Cooling
11
|
| 12
|
27312.5
|
23.5
|
Cooling
12
|
| 13
|
28975.0
|
25.0
|
Cooling
13
|
| 14
|
31002.5
|
26.7
|
Cooling
14
|
The second spreadsheet provided contained data on height, weight and the time of entry into the kiln of each car, for a total of 53 cars. A kiln car transports a group of eight square refractory bats through the kiln. The bats are arranged in a 2 X 4 format with the four bats positioned across the width of the kiln. The height and weight of each refractory bat was manually estimated before the car entered the kiln.
ISSUES
- There is no automated system to measure the height and weight of each
refractory bat, so this data has been manually estimated. As this would
not be acceptable in operation a method of characterising the car loading
should be determined.
- Although, the ultimate target is to maximise product quality/yield
and minimise scrap, currently the quality of the product is not recorded
in direct relation to a given kiln car. In fact, it may be too onerous
in administrative overhead to implement the traceability required to
gather such data. The ware on each kiln car is separated into product
type and loaded onto pallets. When full the pallet is taken for inspection.
It will be very difficult using the existing system to correlate product
quality with the kiln car and hence the conditions in the kiln. It has
been agreed that the next best thing to ensure consistency of quality,
would be to ensure that a consistent temperature profile is maintained
in the kiln. Thus, each piece of ware (regardless of whether it is in
the middle of a heavily loaded car, or has a whole car to itself) should
be subjected to the same set of temperatures over the same timescales,
as any other piece.
- Comparison of time stamped data has been complicated by the unequal
push times, which vary from c8 minutes to c28 minutes in the data provided.
A method for improving this situation is suggested. See recommendations.
SUMMARY
OF RESULTS TO DATE
Statistics
Statistics on the variation in temperatures are presented in Table 1 and
Figure 1 in the appendix.
The variation in temperature is greatest in: preheat (T3), zones 1 and 2
(M4, M5), and cooling (T10, T11 & T12). See figure 1. Zones 3-6 appear better
handled by the current control strategy. Maximum absolute deviations from
the setpoints observed in the data in the controlled zones (M4-9) are: 75,
47, 14, 27, 9, 18 °C respectively.
TEMPERATURE VARIATION vs CAR HEIGHT AND WEIGHT
In order to explore the relationship between car weight and height, and
temperature a comparison of these factors was made. Normalised weight and
height correlated so well that there was no value in looking at these individually,
therefore weight alone was used. T3 was selected, as it is the zone with
greatest variation, and also, because it is the zone directly preceding
the first controlled zone.
As mentioned above, the unequal push times have made the analysis more difficult
as a simple time offset from entry to T3 cannot be applied. Instead, the
accumulated push times have been used. However, the results will be skewed
by 2 factors:
- A longer time at a particular position will allow for some settling
of the temperature
- The airflow rate, related to the pressure variable supplied, has the
effect to shift the profile along the kiln
Even so, there does appear to be a significant correlation of car weight
in zone T3 and temperature at T3 although there is some offset apparent,
see Fig.2. The trend line in Fig. 3 shows that there is a fair degree of
variance of pressure over the same time span, which would potentially give
rise to variations in the above offset.
CONCLUSIONS AND RECOMMENDATIONS
- Without introducing more controlled sections, the largest potential
for reducing temperature variation would be better (predictive) control
of zones 1 and 2.
- Advice is sought on the possibilities for controlling temperature
in the cooling section, as it is understood that this is an area where
there are critical phase changes occurring, during which the right rate
of cooling is critical.
- Height and weight data has been provided at individual bat level,
however given the relatively low resolution of the zones to which individual
temperatures apply, it is suggested that the 8 individual bat estimates
be replaced with a single estimate per car. Also that weight only is
provided.
- Some method of matching readings to pushes rather than just time would
be desirable.
- To complete this initial phase the objectives need to be reviewed
in the light of the information herein. Suggested activities are as
follows:
- Collect further data; complete runs for about another 100 cars
would be desirable.
- Ideally the data should be related to pushes, say temperature
at start of push and at the midway point. Though this could be calculated
if need be.
- Attempt to build initial model using temperatures T1-T3 and pressure
to predict car weight entering Zone 1 (M4). This would be the precursor
to predicting burner requirement at M4.
Questions
- What is the advice on the apparent temperature variation? Particularly,
indicate which zones would benefit from better control and which are
adequately controlled by the current system.
- What are the possibilities for controlling temperature in the cooling
section, as it is understood that this is an area where there are critical
phase changes occurring, during which the right rate of cooling is critical?
APPENDIX
Table 1
|
|
T1
|
T2
|
T3
|
M4
|
M5
|
M6
|
M7
|
M8
|
M9
|
T10
|
T11
|
T12
|
T13
|
T14
|
T15 |
| Min
|
38
|
207
|
686
|
756
|
1008
|
1111
|
1148
|
1096
|
1046
|
801
|
537
|
511
|
462
|
233
|
173 |
| Max
|
45
|
274
|
800
|
885
|
1075
|
1134
|
1178
|
1114
|
1078
|
881
|
607
|
590
|
519
|
286
|
205 |
| Average
|
42
|
242
|
725
|
815
|
1042
|
1125
|
1168
|
1105
|
1060
|
855
|
583
|
558
|
491
|
263
|
189 |
| Setpoint
|
|
|
|
810
|
1055
|
1125
|
1175
|
1105
|
1060
|
|
|
|
|
|
|
| Variance
|
4
|
149
|
379
|
480
|
112
|
8
|
27
|
10
|
23
|
210
|
216
|
237
|
149
|
134
|
46 |
| Max
absolute deviation from setpoint |
|
|
|
75
|
47
|
14
|
27
|
9
|
18
|
|
|
|
|
|
|
| SD
|
2
|
12
|
19
|
22
|
11
|
3
|
5
|
3
|
5
|
15
|
15
|
15
|
12
|
12
|
7 |
Fig 1
Fig 2 - Comparison of normalised Temperature (T3) with normalised inverted
car weight
Fig. 3 - Pressure over the same time span
 |
