Sharp JX-9400 Technické informace stáhnout pdf (Strana 75)

As mentioned above, the experimental domain is about 20 per cent

smaller than the measured space, samples of air being taken at least 0.1 times

the characteristic enclosure dimension from the walls.

Factorial designs

A factorial design for k dimensions and l levels is obtained by dividing the

experimental domain (for example, the interval [1; 1]) on each axis into l

equidistant levels. The complete factorial design contains all the points

obtained by the l

combination of the l possible values of the k coordinates.

The number of points in a full factorial design is hence l

.Ifl and k are

greater than 2, the full factorial designs often have many more points than

the minimum required, and are therefore seldom used. However, partial

factorial designs can be obtaine d by selecting the necessary number of

measurement points from the full design. Some examples are given below.

The 2-D, two-level full factorial design (see Table 3.3) is optimal for a

linear model, providing the coeﬃcien ts of that model with the best accuracy.

If, for economical reasons, one point is omitted, the conﬁdence intervals of

the coeﬃcients are twice that based on four measurement points.

It is very important to note that the frequently used design consisting of

changing one variable at a time (see Table 3.4) is less accurate than the former.

Adding a ﬁfth point at the centre (0,0) of the 2-D, two-level full factorial

design allows assessing the coeﬃcient of the interaction term b

, without loss

of accuracy. The following two points:

No x y

6 10

710

Table 3.3 2-D, two-level full factorial design

No x y

1 1 1

211

3 11

411

Table 3.4 2-D design changing one variable at a time

No x y

110

201

3 10

401

54 Ventilation and Airﬂow in Buildings

1 2 ... 70 71 72 73 74 75 76 77 78 79 80 ... 210 211

Žádné komentáře

Sharp JX-9400 Technické informace Strana 75