Documente Academic
Documente Profesional
Documente Cultură
Accuracy of Temperature
Estimates Using Transfer
Coefficients
In the special circumstances that ambient temperature Te varies linearly between its hourly
values, and Tn = 0 at all times, the heat flows qee and qne found from the transfer coef-
ficients ak , bk and dk will be exact.1 When the wall concerned forms part of a room,
however, Te can be taken as an independent variable, but Tn is neither isothermal nor
adiabatic; its value depends on the thermal characteristics of the room and the way it
is excited – it is a dependent variable. Tn can be calculated but its value will not be
exact. In this chapter the alternative assumption will be made for Tn, namely that it is
an adiabatic node, so that qne = 0 always. A linear excitation θkδ is to be imposed at Te
and the two consequent quantities qee and Tne will be estimated using the wall transfer
coefficients ak , bk, ck and dk , and their values will be compared with the corre- sponding
values given by an exact solution. The errors so found provide an indication
of the accuracy of practical calculations of room response. (If Tn represents the room
index node, there is only a single Tn. If Tn represents a wall inner surface, there will be
a Tn node for each wall and the notation would have to make this clear, but we confine
ourselves to a single wall.) The examination is based on the study reported by Davies
(2001a).
Figure 18.1a shows the elementary r-c model, a single lumped resistance and a single
lumped capacity. In effect, the circuit is adiabatic at T2. The model is too simple to be
given a building interpretation and it will only be discussed analytically.
If T0 = θt from t = 0 onward, T2 is given exactly by (10.9) as
1
It will be recalled that the transfer coefficients reproduce the wall resistance and thermal capacity to a high
degree of accuracy; see equations (11.36) and (11.38)
T0 r T1 T2
T0 r T1
T0 r 1/h T1 r T2
Figure 18.1
where z = rc and is the decay time, or time constant, of the circuit. The expression is the
response of the system to ramp excitation in its simplest form: a linear term in t , a
constant, and transient term in t . We are concerned with temperatures at discrete time
steps of δ, so t = kδ and
Since T2 is taken to be adiabatic, these two flows are equal, so the transfer coefficient
estimate of T2 is
kδ/z + (z/δ)T2,k−1 /θ z
T2,k (TC)/θ z = . (18.4)
(1 + z/δ)
Table 18.1 r-c circuit: Comparison of T2 found from transfer coefficients and its exact
value (z/δ = 1)
Time level, k 1 2 3 4 5
T2,k(TC)/θ z 0.500 1.250 2.125 3.062 4.031
T2,k(exact)/θ z 0.368 1.135 2.050 3.018 4.007
Thus the estimates are larger than the exact values of temperature but they converge to
the exact values. The same is true for other values of z/δ. It can be shown that for the
first time step, the difference
To extend the discussion, we consider a slab of uniformly distributed resistance and capacity
(Figure 18.1b), thickness X, initially at zero temperature, adiabatic at x = X and from t
= 0 onward subjected to a rise of temperature T0 = θt at x = 0. An exact solution for T
(x, t ) is given in Carslaw and Jaeger (1959: 104, equation 4). It can be shown from the
solution that the heat flow into the slab at x = 0 is
J ) l
q00(exact) = θc 1 − (8/π 2) (2j − 1)−2 exp(−(2j − 1)2π 2t/4cr ) (18.6)
(18.7)
where c is the slab thermal capacity ρcp X and r is its resistance X/λ; j is to be summed
from 1 to infinity. Equation (18.7) has the familiar structure: the first right-hand term is
proportional to time, the second is the constant component of the slope response and the
third is the sum of the transient components.
The heat flows at x = X and at time level i as estimated by transfer coefficients are
and since the heat flows are equal and opposite, we have
rN lI
) )N
T1,i = bkT0,i−k − ck T1,i−k c0. (18.9)
k=0 k=1
THE SINGLE SLAB DRIVEN BY A RAMP 426
The transfer coefficient values are in excellent agreement with the exact values, although
they marginally overestimate temperature. Some features may be noted:
• The slab has a first decay time of z1 = 1.27 h (the first solution to e12 = 0, the slab
being isothermal at x = 0 and x = X); this and the subsequent zj values are used to
find the transfer coefficients. In its response, however, being adiabatic at x = X, its
first time constant is four times z1, 5.1 h (the first solution to e11 = 0). Thus we should
expect the transient excitation imposed at t = 0 to have largely died out at around
four times this value, about 20 h. After 24 h the temperature at x = X is increasing at
virtually 1 K/h, the speed of excitation. The transient effect has by now almost died
out and the temperature throughout the slab is increasing at nearly 1 K/h.
• The heat flow into the slab must initially be independent of its thickness and can then
found by a simple exact solution. See Carslaw and Jaeger (1959: 63, equation 4), from
which it follows that
j
q00,t = 2θt λρcp /πt, (18.12)
giving a value of 36.41 W/m2 after 1 h, the value noted in the table.
• When the transient excitation has died out, the constant heat flow is given as
The theory of response factors and transfer coefficients as used here presupposes that heat
flows and temperatures result from a ramp increase in temperature. This is appropriate in
finding the response of a building to varying ambient temperature and solar gains since
they vary steadily in time and the response is based on a linear variation between given
hourly values. However an internal source, in particular an electrical heater, when switched
on, rapidly attains its steady-state value. It constitutes a step excitation and cannot be
modelled exactly by the transfer coefficients. A 1 kW source switched on
at say 1000 and off again at 2000 has to be handled as having values of 0, 1 2 and
1 kW at 0900, 1000 and 1100 respectively, remaining at this value until 1900, then decreasing
again to zero at 2100. It is implicit that the source increases linearly from 0900 to 1100
so that the total heat input is correct, but in handling it in this way, its effect is less
abrupt. It is useful to test consequences of this procedure against an exact solution.
Carslaw and Jaeger (1959: 112, equation 3) provides an expression for the situation
where a slab, initially at zero and insulated on its rear surface, is subjected to a constant
flux F at its front surface from t = 0 onward. The temperature at the excited surface is
J ) l
T00,t (exact) = (F /c) t + rc/3 − (2rc/π 2 ) [(1/j 2 ) exp(−j 2 π 2 t/rc)] , (18.14)
where Fi has the values 0, 1 F 2 and F for i < 0, i = 0 and i > 0, respectively. Solution
of the equations gives the front and back temperatures (Table 18.5).
The TC estimate of the front temperature at t = 0 is in error, as it must be, and subsequent
values are a little high. Rear surface temperatures are somewhat overesti- mated. In the
steady-slope state, the hourly increase of either temperature is Ff".t/c = 100 ×
3600/(0.17905 × 2300 × 1000) = 0.87 K, which is the difference between the 23 h and 24
h values. Furthermore, the steady-slope difference between front and back tem- peratures
is (F /c)(rc/3 − (−rc/6)) = F r/2 = 100 × 0.17905/(1.63 × 2) = 5.49 K. The 24 h values
show this.
THE SINGLE SLAB DRIVEN SINUSOIDALLY 429
Table 18.5 Front and back temperatures (K) in a slab, adiabatic at x = X, driven by a
heat flux of 100 W/m2 at x = 0
Time (h) Flux temperature at x = 0 temperature at x = X
Exact From TCs Exact From TCs
−1 0 – 0.00 – 0.00
0 50 ∗ 1.37 0.00 0.01
1 100 3.50 3.55 0.03 0.16
2 100 4.95 4.97 0.38 0.57
3 100 6.07 6.12 1.00 1.19
6 100 8.89 8.98 3.43 3.57
23 100 23.77 23.88 18.27 18.39
24 100 24.64 24.76 19.15 19.26
∗
Converges slowly to zero.
These are complex quantities with magnitudes W/m2 K and the phase leads in y0 and y1 and
the phase lag in u are conveniently expressed in hours. Parameters for the condition where
instead a surface is maintained adiabatic can be similarly defined:
Values of the standard parameters were found directly from the transfer coefficients as in
Section 17.7.2 and in their exact versions as in Section 15.3. For the slab in question,
Since the slab is symmetrical, y1 = y0. The exact values are found assuming strict sinusoidal
variation in the driving temperature whereas the TC values are based on a 24-segment
approximation to the sinusoid; strict agreement is not expected.
Transfer coefficients versions of y01 and r1 were evaluated from a time sequence using
(18.8) and (18.10), convergence being comparatively slow, and the exact versions from
Davies (1994). The values obtained are:
Certain heat inputs to an enclosure act directly on interior room surfaces. These include
solar radiation transmitted as short wave through a window and the flux from a heated
floor or cooled ceiling. Long-wave radiation from an internal heat source need not be
handled in this way. If a flux acts at a surface node, the transfer coefficients for the
wall must be based on the construction from the exterior up to that surface, but not including
the inner film. The wall will respond to this flux and also to variation (in part concomitant)
in the room temperature to which it is linked by its film. In this section, therefore, we
examine the response of the above slab when driven by a ramp increase in
FILM AND SLAB DRIVEN BY A RAMP 431
As before, T2 will be taken to be adiabatic so that q22 and q21 are equal and opposite at all
times. The continuity equation at x = 0 is
Heat is driven into a wall exterior by ambient temperature Te acting through an external
film h and by solar gain I , of which α1 I is absorbed at the exterior surface. These are
independent but their action can be combined as the sol-air temperature Tsa = Te + α 1 I/h
whose value is supposed known. In this case the wall may be taken to consist of a film h
(8 W/m2K) and the slab as before. We now return to the former interpretation of
Figure 18.1c: T0 denotes sol-air temperature, T1 is no longer of concern, and the transfer
coefficients refer to nodes 0 and 2. Their values ),
are given
), in Table
), 18.6.
), ), ),
Once again it will be found that a k / dk = b k / dk = ck / dk
= (1/h + X/λ)−1, the steady-state transmittance. The value of c0 is the same as in
Table 18.2, reflecting the fact that the heat flow at x = X due to a ramp imposed at
Table 18.6 Exact calculation parameters and transfer coefficients for the film and
slab
ak bk ck dk
0 6.72172 0.03416 36.41342 1.0
1 −6.80820 0.50768 −53.60080 −0.88595
2 1.07038 0.40255 19.46575 0.11461
3 −0.01327 0.02619 −1.31309 −0.00071
4 0.00002 0.00007 0.00537 0.00000
Table 18.7 A slab (V = 1) adiabatic at back surface, driven through a film at front surface
Time Front heat flows (W/m2) Back surface temperatures (K)
Exact TC, separate TC, combined Exact TC, separate TC, combined
0 0.00 0.00 0.00 0.00 0.00 0.00
1 6.72 6.56 6.72 0.00 0.00 0.00
6 32.35 32.16 32.31 0.45 0.48 0.47
12 54.88 54.63 54.77 2.32 2.37 2.35
24 83.08 82.81 82.93 9.15 9.21 9.19
48 105.73 105.58 105.64 29.01 29.03 29.02
167 114.38 114.38 114.38 146.42 146.42 146.42
168 114.38 114.38 114.38 147.42 147.42 147.42
These comparisons have been conducted for simple walls having well-known exact solu-
tions for adiabatic surfaces. The theory of Chapter 17, however, can be readily adapted to
find the exact solution for a wall of arbitrary (one-dimensional) construction. Sup- pose
that the real wall consisted of three layers, e.g. an outer film S1, concrete block S2 and
plaster S3. Its set of transfer coefficients would be found in the normal way. To
compute the exact response when the inside node T3 is adiabatic, we set up a ‘wall’ of
double construction S1, S2, S3, S3, S2, S1(n = 6) and suppose that both T0 and T6 have rise
rates θ imposed upon them. Physically this amounts to symmetrical excitation of a
symmetrical wall: the temperature gradient at midplane is zero, so T3 is an adiabatic node.
The relevant computer code provides sets of slope coefficients AL1 etc., hence values for
qnO,j ∗∗ ; the double asterisk denotes that they are for the double wall. When j is even,
qnO,j ∗∗ is zero. The amplitudes of the transient components of response at midplane (T3) are
then E3j qnO,j ∗∗ and the transient component of T3 is found by summing terms given in
(17.59a). The slope component of temperature is given straightforwardly by the slope matrix
at its n = 3 stage of evaluation.
DISCUSSION 433
18.7 DISCUSSION
In all cases, transfer coefficient (TC) estimates of temperature and flux tend to their exact
values with increasing time after the onset of transient change; this must be the case since the
steady-slope components are common to both approaches and, although the transient
components differ, both sets tend to zero with increasing time. Deviations will be seen in
the earlier stages of change, when t is of order equal to the first decay time of the system.
It has been pointed out that the ‘system’ here is the wall as it acts, one node being
adiabatic, so its first decay time is larger than the first decay time of the wall used in finding
the transfer coefficients.
The deviations, however, are small compared with the changes themselves and this must
be seen in the context that we rarely know reliable values for the thermal constants of the
wall layers. In conducting calculations on room response, the ingredient values must be taken
as nominal and the estimates interpreted accordingly.2 (This argument cannot be used however
to justify approximating values for the a, b, c, d set. It will be recalled that
2
In a study of the effect of rain on the heat gain through a building wall in tropical climates, Jayamaha et al.
(1997) give the uncertainty in heat flux due to the estimated uncertainty in the parameters.
These uncertainties are of course specific to their study. In particular, they remark that the variation in specific heat,
10–20%, is based on values for the dry and fully saturated state, presumably known reliably. The building analyst
may not know reliably the value for the specific heat of some wall component but the table suggests that a 20%
uncertainty in its value might lead to 2% uncertainty in an estimate of the heat flux. The authors’ uncertainty for
the heat flow predictions was computed ‘based on the root-sum-square method using the individual uncertainties
of the variables in the table and . . . was estimated to be 12%. The uncertainty for the heat flow predictions made
under dry conditions was found to be about 10%’.
the values of a, c, and d alternate in sign and may be large in magnitude. Calculations that involve summing
terms which include them must be conducted to sufficient significant figures to avoid a build-up of errors.)
Finally, when Te varies linearly between intervals of δ and Tn = 0, the flux qne is exact but when qn =
0 Tne has errors, as illustrated above. In practice however, when Tn is just one node among several which
describe a room, it is neither isothermal nor
adiabatic and we should expect errors to be somewhat less than these limiting values. Room
internal radiant and convective exchanges involve no storage, so they impose no further
computational errors. While we recognise that calculating room heat transfer is subject to a
great many uncertainties, listed in the next chapter, the information can be handled reliably
using transfer coefficients.