Professional Documents
Culture Documents
Boyd French
Ford Motor Company
Copyright 2003 SAE International
ABSTRACT
This paper discusses a method of measuring the
instantaneous
exhaust
gas
temperature
by
thermocouples. Measuring the exhaust gas temperature
is useful for a better understanding of engine processes.
Thermocouples do not measure the instantaneous
exhaust gas temperature because of their limited
dynamic response. A thermocouple compensation
technique has been developed to estimate the time
constant in situ. This method has been commissioned in
a simulation study and a controlled experiment with a
reference temperature. The studies have shown that the
signal bandwidth has to be restricted, since noise will be
amplified in the temperature reconstruction. The
technique has been successfully applied to some engine
exhaust measurements. A comparison between two
independent pairs of thermocouples has shown that
temperature variations at frequencies up to 80Hz can be
recovered. The medium load results agree with a
previous study, which used fast response thermometers
with a bandwidth of about 50 Hz. However, the results at
low load and two different speeds have highlighted the
need to do some 1-D unsteady flow simulations, in order
to gain more insight into the exhaust process.
INTRODUCTION
With the introduction of even more stringent emission
limits (EURO IV, SULEV), engine developers are striving
to better understand engine processes. Exhaust gas
temperature (EGT) can be very useful in this respect. For
instance, the heat transfer across the exhaust port can
be estimated with EGT measurements, and this would
enable a better heat transfer model to be developed. In
addition, EGT forms the basic calibration data for
computational fluid dynamic models (CFD), which are
commonly used in engine design.
Knowing the exhaust gas temperature can immediately
help many other applications as well. Following the
engine cold start, the catalytic converter has not yet
reached its operating temperature, so most of the
DYNAMIC RESPONSES
By performing an energy balance, the dynamic response
of a thermocouple can be described by Eq. 1.
Tg Tw =
w c w d Tw
4h
k w d 2Tw w 4
4
(1)
+
Tw Tsurr
4h x 2
h
Tg Tw =
where,
= w c w d 4h .
d Tw
dt
(2)
~
~ ~
* d Tw
Tg Tw =
dt
(3)
TEMPERATURE RECONSTRUCTION
RESPONSE EQUATION
In exhaust gas measurements, what is of interest is the
instantaneous gas temperature (Tg). If * and are
known, Eq. 2 or Eq. 3 can be solved for Tg; may be
calculated from a know radiation and conduction
correction factor [8]. Otherwise the thermocouple is
designed to minimize conduction and radiation errors, so
is close to unity. The Conduction and Radiation section
will show that is close to unity, so the main task here is
to find the time constant.
PREVIOUS WORK
It is very difficult to find the time constant analytically for
the following reasons:
d Tw1
= Tg
dt
dT
Tg2 = Tw2 + 2 w2 = Tg
dt
Tg1 = Tw1 + 1
(4)
d
= 2 = 2
1 d1
2 m
(5)
1 =
(T1 T2 )
d T1 /d t d T2 /d t
(6)
1 =
(T1 T2 ) + enum
d T1 /d t d T2 /d t + eden
(7)
Tg1 = Tg 2 =
( 1)T1T2
T2 T1
(8)
1 N (i )
Tg1 Tg2(i )
N i =1
(9)
th
= 0;
1
=0
2
sample of the
(10)
Constant
No
Yes
Constant
Yes
Yes
No
Yes
No
No
No
No
No
Yes
No
KALMAN FILTER
The Kalman filter is a recursive procedure for estimating
the latent variables, , in the following linear stochastic
equation [17].
(i ) = G (i ) (i 1) + w (i )
y (i ) = F ( i ) (i ) + v ( i )
(11)
where:
G is a flow matrix
w is the state noise, normally distributed with zero mean
and covariance W.
y are the multivariate observations
F is a transformation matrix
v is the observation noise, normally distributed with zero
mean and covariance matrix V.
(i )
(i )
= F + v (i )
(i )
+w
(12)
d Tw1
dT
+ 2 w2
dt
dt
(i )
2 (i )
y
R (i ) = (i1) + W (i ) ,
( )
2 (i )
y
( ) + ( )
= 2
(i )
2 (i )
( )
(13)
(14)
(i ) = R (i ) K ( i ) F ( i ) R (i )
where:
= F (i ) R (i ) F T
(i )
2 . It
( )
( )
p y (i ) = N y ( i ) , y2
(i )
(16)
( )
2 (i )
q0
( )
= 2
(i )
( )
+ F (i ) (i 1) F T
(i )
(17)
(i ) = (i1) + K (i ) e (i )
( )
2 (i )
th
d Tw(i2) 1(i )
d t 2(i )
d T (i )
Tw( i1) Tw( i2) = w1
dt
Tw1 Tw2 = 1
( )
( )
R (i ) F T
is given by
(i )
(i )
(i )
x Kif x 0
h( x ) =
0Kotherwise
(15)
(18)
(i )
( ) ( )
( )
e 2 (i ) 2
q0
= h
F (i ) F T ( i )
(i )
(19)
Observation noise
( )
2 (i )
= h (e 2 ) F ( i ) R ( i 1) (F T )
(i )
(i )
(20)
3-WAY
VALVE
SHUTTER
EQUIPMENT
THERMOCOUPLE PROBE
ADAPTOR
NOZZLE
3
4
STAINLESS
STEEL TUBE
KEYS:
CHROMEL
ALUMEL
COPPER
LEMO 7-WAY
SOCKET
NOT TO SCALE
REFERENCE
JUNCTION
THERMOCOUPLE
COLD STREAM
HOT AIR
GUN
HOT STREAM
Bore (mm)
Stroke (mm)
Compression ratio
Maximum bmep (bar)
Inlet valve opens (btdc)
Inlet valve closes (abdc)
Exhaust valve opens (bbdc)
Exhaust valve closes (atdc)
75.0
79.0
10.5
11.2 @ 3500 rpm
12
52
52
12
MODEL VALIDATION
To ensure the integrity of the temperature reconstruction,
the Kalman filter was commissioned in a simulation study
and a well-controlled experiment before being applied to
the exhaust temperature measurements.
SIMULATION STUDY
A sinusoidal variation of the gas temperature was
defined. It was then used for solving the two
thermocouple temperatures by Eq. 4 with a common
initial temperature. The time constants, which have mean
values of 30 and 70 ms, were also defined to vary
sinusoidally at 1/5 of the temperature frequency. The two
synthetic thermocouple time series were contaminated
by Gaussian noise with very small variance of 1e-8. The
resulting gas and thermocouple temperature traces are
plotted in Figure 6.
p( y ) ) are shown in
g (t ) = exp( t 2 )
Figure 11 Time constant estimates with noise variance = 0.1. The
transformation matrix is derived from the noiseless data.
y (t ) = x(t ) g (t )
When the simulation is rerun for the same conditions,
except that the transformation matrix (F), (which is
formed by the gradients), is calculated from the noiseless
data, then the result is shown in Figure 11. With accurate
gradients, the Kalman filter is able to estimate the time
constants. So it can be concluded that the Kalman filter
cannot handle noise in F. This is similar to the leastsquares method when it gives a negative time constant.
The Kalman filter works by maximizing the likelihood of
the prediction (y). The prediction is however calculated
through the transformation matrix, F. A noisy F is less
deterministic, so it cannot transform the time constants
to the prediction. Therefore the estimations fail.
Noise and reconstruction
The process of reconstruction is to recover (amplify) the
high frequency component of the thermocouple signals,
which have been attenuated due to the thermal inertia of
the thermocouple junction. In a noiseless signal, no
(21)
(22)
d y (t ) d x(t )
=
g (t )
dt
dt
d g (t )
= x(t )
dt
(23)
Figure 13 The power spectral density of the gradient over its signal
(the transfer function of the gradient). A Hamming window was applied
to the Marr trace.
Figure 22 Cross plot of the true gas temperature against the raw
thermocouple temperature at fT = 2.6 Hz. Black solid line has a unity
slope. Black dashed line is the best fit line for the data.
Figure 23 Cross plot of the true gas temperature against the raw
thermocouple temperature at fT = 10 Hz. Black solid line has a unity
slope. Black dashed line is the best fit line for the data.
Figure 28 Cross plot of the true gas temperature against the raw
thermocouple temperature at fT = 10 Hz. All signals have their
bandwidth restricted to 54 Hz.
1Tg = Tg1
2Tg = Tg 2
(24)
~
~
Tg1 = Tg 2
(25)
Time-averaged
thermocouple
temperatures. Engine was running at 2
bar bmep, 1000 rpm.
Thermocouple
0.002, T2
0.003, T3
0.005, T5
Time-averaged temperature
[C]
582.3
587.5
581.0
METHOD OF ANALYSIS
The thermocouple signals were compared cycle by cycle.
Even though the engine was running in steady state, the
thermocouple temperatures were subject to large cycleby-cycle variations as depicted in Figure 30. For this
reason, each thermocouple signal is ensemble averaged
over 200 cycles to get a representative mean. This is
primarily to ensure fair comparison across different loads
and speeds, but it also improves the signal to noise ratio
of the signal. For fast tracking of the time constants, the
signals are upsampled 10 times. Finally these processed
signals are used for compensating the thermocouple
temperatures in the same way as they were in the air-rig
experiment.
DIFFERENT LOADS
The exhaust measurements at various engine loads
were analyzed. The results are listed in Table 4. It was
found that the 6-bar case was much more difficult to
compensate, and the bandwidth achieved was only 60
Hz. In the Filtering and Reconstructed Temperatures
section, the high frequency features can be revealed by
having different levels of filtering on the gradients. This
method was applied to the 6 bar results as shown in
Figure 34. A bandwidth of 60 Hz is just insufficient. The
initial cooling and the first peak in the 500 Hz traces are
absent in the 60 Hz trace. However, it must be stressed
that the absolute magnitude of these high frequency
features may be subject to large errors as proven in the
air rig experiment.
Table 4
2 bar
80
0.037
0.082
4 bar
80
0.024
0.048
6 bar
60
0.020
0.032
Table 5
2 bar
1.9
2.35
4 bar
4.2
4.71
6 bar
6.1
6.46
-0.56
-0.35
-0.19
29
27.1
22.7
24.3
21.4
18.7
12
DIFFERENT SPEEDS
For 4 bar and 6 bar bmep, what follows the rapid cooling
is an increase in exhaust temperature to the first peak
just before bdc. The gas temperatures decrease from
this peak to a minimum at about 440ca, then increase
again until 480ca. These general changes in exhaust
temperatures agree with Caton & Heywoods study [23],
in which the exhaust temperature was measured by a
resistance wire thermometer [7]. Heywood [22] pointed
out that the local minimum at 440ca is the transition
from blowdown flow to displacement flow. Referring back
to the time constant estimates in Figure 31, the time
constants suddenly increase at this point, whilst the
likelihood of the estimation in Figure 32 indicates there is
a change in the system dynamics. The 2-bar result does
not follow the trend. Perhaps the mass flux becomes so
low at 2 bar bmep that it forms only a small proportion of
the mixed gas (Reason 2), and so delays the
temperature rise.
Table 6
2000 rpm
160
0.017
0.035
CONCLUSIONS
The dynamic response of a thermocouple can be
described by a time constant. Because it is too difficult to
predict analytically, a thermocouple compensation
technique has been developed to estimate the time
constant in situ. By this means, the instantaneous gas
temperature may be found, once the time constant is
known. This technique is one of many 2-thermocouple
techniques, which are based on the assumption that two
thermocouples are exposed to the same temperature
and velocity fields. It is based on a Kalman filter, and has
the following advantages over other methods:
ACKNOWLEDGMENTS
This work was supported by the Ford Motor Company.
The first author (KK) would also like to acknowledge the
financial support of Foundation for Research, Science
and Technology of New Zealand.
13.
REFERENCES
14.
1. Koehlen, C., Holder, E., and Vent, G., Investigation
of Post Oxidation and Its Dependency on Engine
Combustion and Exhaust Manifold Design. SAE
Paper 2002-01-0744, 2002.
2. Kee, R., O'Reilly, P.G., Fleck, R., and McEntee, P.T.,
Measurement of Exhaust Gas Temperatures in a
High Performance Two-Stroke Engine. SAE Paper
983072, 1998.
3. Bauer, W., Tam, C., Heywood, J.B., and Ziegler, C.,
Fast Gas Temperature Measurement by Velocity of
Sound for IC Engine Applications. SAE Paper
972826, 1997.
4. Andresen, P., Meijer, G., Schlter, H., Voges, A.,
and Hentschel, W., Fluorescence Imaging Inside an
Internal Combustion Engine using Tunable Excimer
Lasers. Appl. Opt., 1990. 29: p. 2392-404.
5. Koch, A., Voges, H., Andresen, P., Schlter, H.,
Wolff, D., Hentschel, W., Oppermann, W., and
Rother, E., Planar Imaging of a Laboratory Flame
and of Internal Combustion using UV Rayleigh and
Fluorence Light. Appl. Phys. B, 1993. 56: p. 177-84.
6. Benson, R.S. and Brundrett, G.W., Development of a
resistance wire thermometer for measuring transient
temperatures in exhaust systems of internal
combustion engines, in Temperature - Its
Measurement and Control in Science and Industry.
1962, Reinhold: New York. p. 631-653.
7. Benson, R.S., Measurement of transient exhaust
temperatures in I.C. engines. The Engineer, 1964.
217: p. 376-383.
8. Scadron, M.D. and Warshawsky, I., Experimental
Determination of Time Constants and Nusselt
Numbers for Bare-Wire Thermocouples in HighVelocity Air Streams and Analytic Approximation of
Conduction and Radiation Errors. NACA TN 2599,
1952. p. 5-6, 20-22, 40-45.
9. Cambray, P., Measuring Thermocouple Time
Constants: A New Method. Combustion Science and
Technology, 1986. 45: p. 221-224.
10. Forney, L.J. and Fralick, G.C., Two wire
thermocouple: Frequency response in constant flow.
Review of Scientific Instruments, 1994. 65(10): p.
3252-3256.
11. O'Reilly, P.G., Kee, R.J., Fleck, R., and McEntee,
P.T., Two-wire thermocouples: A nonlinear state
estimation approach to temperature reconstruction.
Review of Scientific Instruments, 2001. 72(8): p.
3449-3457.
12. Tagawa, M. and Ohta, Y., Two-Thermocouple Probe
for Fluctuating Temperature Measurement in
Combustion- Rational Estimation of Mean and
15.
16.
17.
18.
19.
20.
21.
22.
23.
CONTACT
Any comments and questions should be directed to
Kenneth Kar (kenneth.kar@eng.ox.ac.uk) or Richard
Stone (richard.stone@eng.ox.ac.uk). The corresponding
address is: Department of Engineering Science,
University of Oxford, Parks Road, Oxford OX1 3PJ
Temperature, K/ C
AC
Time, s
Observation noise
bdcc
bmep
Static noise
btdc
DC
EGT
EVC
Multivariate observations
EVO
FTP-75
imep
Emissivity
PSD
Density, kg/m
Transformation matrix
Frequency, Hz
Flow matrix
Time constant, s
g(t)
Gaussian kernel
h(x)
Ramp function
Superscript
Identity matrix
(overbar)
Estimated value
Transpose of a matrix
Reconstructed temperature, C
th
Subscript
1
Thermocouple 1
23
25
Thermocouple
with
diameter of 0.003
nominal
Thermocouple
with
diameter of 0.005
nominal
den
Denominator
Gas
num
Numerator
q0
surr
Surroundings
Temperature
Wire
Prediction
Latent variables