straightforward, takes time and has its cost, the theory of experimental
design (Box et al., 1978) may help in providing a maximum of information
through a minimum of measurements.
Minimum number of measurements
A map of any scalar variable, v, in a three-dimensional room is in principle
obtained by measuring the variable at each node of a network and interpo-
lating between these nodes. Such measurements are, however, very expensive
and may be unfeasible. If only five values are taken on each axis, at least 125
measurements are re quired, meaning analysis of the air every few minutes at
125 locations over a couple of hours. Therefore, it makes sense to apply a
method that needs a minimum number of measurements points. This
minimum number depends on the objective of the mapping experiment, or
more precisely on the required mapping details. Since the interpolation
between meas urement points needs a model, the mapping network
indeed depends on the empirical model chosen to represent the map of the
variable, v.
Any infinitely derivable function (as v is assumed to be) can be developed
in a Taylor series around a given point. This gives a polynomial, which can
be approximated by its k þ 1 first terms, k being the degree of the poly-
nomial. In the following, models of degree 1 and 2 are considered. If a linear
model is ado pted (degree 1), such as:
v ¼ a þ
X
i
b
i
x
i
ð3:29Þ
where x
i
are the three coordinates of the measured point, only four measure-
ments are neede d to obtain a set of coefficients ( a; b
i
). If more meas urements
are made, the coefficients may be obtained by a least square fit procedure
provided there is no (or negligible) uncertainty on the coordinates. If their
coordinates differ for the other points, these supplementary measurement
points give information on the validity of the model used.
If the linear model does not appear to be valid, higher degree models may
be used. For example, a quadratic model:
v ¼ a þ
X
i
b
i
x
i
þ
X
i 6¼j
b
ij
x
i
x
j
þ
X
i
b
ii
x
2
i
ð3:30Þ
that contains ten coefficients, can be chosen. Such a model may already fit
many practical situations and present minimal and maximal value(s). To
determine its coefficients, measurements taken at ten locations are the
minimum necessary.
An intermediate model is the interactions model:
v ¼ a þ
X
i
b
i
x
i
þ
X
i 6¼j
b
ij
x
i
x
j
ð3:31Þ
50 Ventilation and Airflow in Buildings
Commenti su questo manuale