28.01.2013 Views

Zienkiewicz O.C., Taylor R.L. Vol. 3. The finite - tiera.ru

Zienkiewicz O.C., Taylor R.L. Vol. 3. The finite - tiera.ru

Zienkiewicz O.C., Taylor R.L. Vol. 3. The finite - tiera.ru

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

108 Incompressible laminar ¯ow<br />

This approximation is made to be a least square approximation to the actual<br />

distribution of curvatures, i.e.<br />

…<br />

N T N @2<br />

"<br />

@xi@x j<br />

#<br />

h<br />

@2<br />

ÿ d<br />

@xi@x j<br />

ˆ 0 …4:29†<br />

and integrating by parts to give<br />

@2 @xi@xj ˆ M ÿ1<br />

…<br />

N T 2 h<br />

@<br />

@xi@x j<br />

d ˆÿM ÿ1<br />

… T<br />

@N @N<br />

d<br />

@xi @xj ~ …4:30†<br />

where M is the mass matrix given by<br />

…<br />

M ˆ N T N d …4:31†<br />

which of course can be `lumped'.<br />

4.5.2 Element elongation<br />

Elongated elements are frequently introduced to deal with `one-dimensional'<br />

phenomena such as shocks, boundary layers, etc. <strong>The</strong> ®rst paper dealing with such<br />

elongation was presented as early as 1987 by Peraire et al. 22 and later by many authors<br />

for ¯uid mechanics and other problems. 70ÿ73 But the possible elongation was limited<br />

by practical considerations if a general mesh of triangles was to be used. An<br />

alternative to this is to introduce a locally st<strong>ru</strong>ctured mesh in shocks and boundary<br />

layers which connects to the completely unst<strong>ru</strong>ctured triangles. This idea has been<br />

extensively used by Hassan et al., 39;53;56 <strong>Zienkiewicz</strong> and Wu 50 and Marchant<br />

et al. 65 in the compressible ¯ow context. In both procedures it is necessary to establish<br />

the desired elongation of elements. Obviously in completely parallel ¯ow phenomena<br />

no limit on elongation exists but in a general ®eld the elongation ratio de®ning the<br />

maximum to minimum size of the element can be derived by considering curvatures.<br />

Thus the local error is proportional to the curvature and making h 2 times the<br />

curvature equal to a constant, we immediately derive the ratio h max=h min.<br />

In Fig. 4.13, X 1 and X 2 are the directions of the minimum and maximum principal<br />

values of the curvatures. Thus for an equal distribution of the interpolation error we<br />

can write for each nodey<br />

h 2 min<br />

@ 2<br />

@X 2 2<br />

which gives us the stretching ratio s as<br />

ˆ h 2 max<br />

s ˆ hmax ˆ<br />

hmin v<br />

u<br />

t<br />

@ 2<br />

@X 2 1<br />

@ 2<br />

@X 2 2<br />

@ 2<br />

@X 2 1<br />

ˆ C …4:32†<br />

…4:33†<br />

y Principal curvatures and directions can be found in a manner analogous to that of the determination of<br />

principal stresses and their directions. Procedures are described in standard engineering texts.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!