This action might not be possible to undo. Are you sure you want to continue?

261

Let us assume that, as is usual, dp/dx is given as a function of x , and, further, that the velocity u is given as a function of y at some initial value of .r , Then we can determine numerically, from each u , the associated au fax , and with one of the known algorithms we can then proceed. step by step, in the x-direction. A difficulty exists, however, in

various singularities which appear on the fixed boundary. The simplest case of the fiow situations under discussion is that of water streaming along a thin flat plate. Here a reduction in variables is possible; we can write !1 "" f(ylx~). By numerical integration or the resulting differential equation we obtain an expression for the drag D

= 1.1 x bv'?,piu~

(b breadth, I length of the plate, Uo velocity of the undisturbed water relative to the plate}. The velocity profile is shown in [Fig. 8.l}. For practical purposes the most important result of these investigations is that in certain cases, and at a point wholly determined by the external conditions, the flow separates from the wall [Fig. 8.2]. A fluid layer which is set into rotation by friction at the wall thus pushes itself out into the free fluid where, in causing a complete transformation of the motion, it plays the same role as a Helmholtz surface of discontinuity. A change in toe coefficient of viscosity ~ produces a change in the thickness of the vortex layer (this thickness being proportional to V(f.ll/ pU}), but everything else remains unchanged, so that one may, if one so wishes, take the limit u- 0 and still obtain the same [low picture, Separation can only occur'if there is an increase in pressure along the wan in the direction of the stream ...

The amount of insight packed into this part of Prandtl's paper is staggering, and much of the present chapter will be spent filling in the details, particularly with regard to the derivation of the

VI

Fig. &.1. Prandtl's diagram of the velocity profile in the boundary layer on a fiat plate.

y

including the impulsively started motion of a circular cylinder (Fig. boundary layer equations (§8.3). The motion in the boundary layers is regulated by the pressure gradient in the mainstream flow but. . Prandtl emphasizes how the flow of a fluid of small viscosity must be dealt with in two interacting parts. 8. The sketch of boundary-layer separation in Prandtl's 1905 paper. the character of the mainstream flow is. 8.2. Prandtl goes on to discuss some particular cases.3).3. 8. Flow relative to an impulsively moved circular cylinder at two different times (from Prandtl 19(5). on the other hand.2) and their solution in the case of flow past a fiat plate (§8. He finally reports some experiments undertaken in a hand-operated Fig. namely an inviseid flow obeying Helmholtz's vortex theorems and thin boundary layers in which viscous effects are important. Dashed lines indicate layers of strong vorticity.262 Boundary layers Fig. After the passage quoted above. in turn. markedly influenced by any separation that may occur.

These include flow past a wall. In the last case he demonstrates that even a very small amount of suction into a slit on one side of the cylinder is enough to prevent separation of the boundary layer on that side (Fig. 8A).5).4. He notes. Prandtl's hand-operated flow tank. too. flow past a circular arc at zero incidence. A substantial adverse pressure gradient will therefore be impressed on the boundary layer on the corresponding side wall of the tank . Iij 11 II :! 11 'I \I Fig. and flow past a circular cylinder. water tank (Fig. and therefore the pressure must rise'. a most interesting consequence of this.Boundary layers 263 ! I 11. 8. because 'the speed must decrease in the broadening aperture through which the water flows. 8.

of course. Prandtl was not alone. Viscous flow theory effectively began with the pioneering paper of Stokes (1845). As early as 1872 Fronde had conducted experiments on the drag on a thin fiat plate towed through still water. then. that inviscid flow theory had its value. The other pioneering work of the time was by Reynolds. t For all its fundamental insights. Yet a major problem remained: that of accounting for the motion of a fluid of small viscosity past a solid body. idea! flow theory and viscousflow theory had more or less gone their separate ways.2). There was no doubt. On the other hand. On the inviscid side there had been the great papers of Euler (1755) on the fundamental equations. and accordingly thai boundary layer must be expected to separate. 8. namely through its cumulative effect. in accounting for many of the most important properties of water waves and sound waves. let alone fully appreciated. Such separation is indeed observed (Fig.1). who not only laid down the equations of motion but obtained many of the elementary exact solutions that are to be found in Chapter 2. He followed this with another important paper (1851) on what we would now call low Reynolds number flow (§7. He points out. of Helmholtz (1858) on vortex motion and of Kelvin (1869) on the circulation theorem. Prior to 1900. over a long time interval. its subject is the motion of a fluid with very small viscosity.10). quite early in the paper. too. and when Hele-Shaw performed his remarkable experiments with irrotational streamline patterns (1898. There had been success. in a region of closed streamlines (see §5.264 Boundary layers (uppermost in Fig.5). and on this successful note the paper ends. and had attributed that drag to + Prandti's paper is not exclusively concerned with boundary layers. 8. and several years were to pass before Prandtl's work became widely known outside Germany. in addressing the matter. the paper was scarcely an overnight success.5). notably his beautiful experiments in 1883 on transition in fiow down a pipe (s9. . see 97. a wbolly different way in which small viscosity can be significant.7) it was Stokes again who produced the relevant viscous theory. it was well known that uniform flow past a 'bluff' body--such as a circular cylinder-bore little resemblance at the rear of the body to the predictions of inviscid flow theory.

with some investigators convinced of the no-slip condition only in the case of slow flow (see Goldstein 1969. in his Aerodynamics of 1907. that the drag should be proportional to I" lI2u~n. and pp. In an essay on pp. nigh on a hundred years later. 676-680 of Goldstein 1938). it was known that when ideal flow theory predicts a negative value of the absolute pressure at any point in a liquid. the concept of a boundary layer and its separation seem to have been a long time in the making. e.Boundary layers 265 the layers of fluid in intense shear near the plate. independently of Prandtl. 497-506). . Thus when irrotationai flow past sharp corners (e. Fig. 96-99 of his 1845 paper. and affirmed correctly that on a rotating cylinder in a uniform stream separation would be delayed on one side and hastened on the other. It was this quite different explanation. that the drag varied not in proportion to the length { of the plate. there seemed to be a ready explanation: ideal flow theory implies an infinite speed at the corner. and by Bernoulli's theorem this means an infinitely negative pressure. but at a slower rate. will give rise to a different and 'separated' flow (see. He had found. may be expected. too. Batchelor 1967. Lighthill emphasizes how this obscured the possibility that there might be a quite different (viscous) explanation for flow separation. although just how much earlier the work was done is not entirely clear. He published all this and much more. in so doing. but it is quite another to find this uncertainty continuing right up to the turn of the century. 4.g.6(a)) bore little resemblance to the actual flow of real liquids such as water. Lanchester later proposed. the formation of bubbles of vapour. Second. it is worth recalling that there were at least two factors which clouded the issue at the time. that Prandtl was eventually. one which would obtain. known as cavitation. It Is one thing to find Stokes unsure about the matter on pp. pp. indeed. First. He also discussed separation. there was substantial uncertainty about whether the correct boundary condition was one of no slip. If. The onset of cavitation will prevent such a Singularity occurring but.. to squeeze into just a few pages in 1904. along with so much else. for liquids or gases and whether the rigid boundary was sharp-cornered or smooth. 1-5 of Rosenhead (1963).g.

l) (8. p(x) being a function of x alone.i. and let u change by an amount of order Uo over an z-distance of order L. . If (. layer flow must also match with the inviscid mainstream in some appropriate manner. The boundary layer.266 Boundary layers 8. 8.!137"}Y". There are two key ideas involved in boundary layer theory. the coordinate normal to the boundary.YY&'#/"ffi7it x Fig.. The first is that u and v vary much more rapidly with y. the coordinate parallel to the boundary. than they do with x . This is a matter of Some subtlety.'.6. The boundary aty =0. The steady 2-D boundary layer equations We now derive the equations adjacent to a rigid wall y = 0: for a steady 2-D boundary laver au au 1 dD u-+ v-= __.3) -+-=0 au au ax By .------- ! I ~ BOUNDARY LAYER WOYd!.2) conditions at the (8. oy ax I I U(x) --------. say.. The boundary wall are u=v=o if the wall is at rest. Let Uo denote some typical value of u._+ ax 3y p ill '\f- cPu &y2 ' (S. our basic approximation is I I au » ) au I.2. denotes a typical value of the thickness of the boundary layer.. and we postpone it for the time being.

a function of x alone. On viewing the first two equations as expressions for apt ax and op I 8y respectively it then follows that I ? 1« ISP ). and as v is zero at y = 0 it follows that v is of order Uoo I L in the boundary layer. i. This justifies the use of dp/dx.4) au au 1 3p u-+v-=---:. the exact 2-D equations are (8.:-+v ax oy P ClX u ax + v 2y (' cPu av 2v =- p ay + v 1 dP (a v 2 \8x -+-~ . and bears out Prandtl's remark that 'the pressure distribution of the free fluid will be impressed on the transition layer'. estimate of o«L. to a first approximation.e . in eqn (8.Boundary layers 267 and this amounts. notwithstanding the small coefficient of viscosity v.5) -+-=0. Thus v is much smaller than u . to Vol 0 » uol L.4) the term riu/8x2 is negligible compared with the term fiu /2y2. au dX 2v 8y It follows at once from the last of these that av f 2y is of order Vol L. 2 oy 2X2 CPU') .:Pv\ + ay2)' (8.1). rather than 2p/2x. I oy I I ax which means that in the boundary layer p is. We may at once use this consideration . But the most dramatic simpufication of eqn (85) arises on account of the following estimates: (8. and with this major simplification of eqn (8.6) In view of eqn (8. The other key idea involved in boundary layer theory is that the rapid variation of u with y should be just sufficient to prevent the viscous term from being negligible.1). Now.5) We complete our derivation of eqn (8. by making an order of magnitude each term.

y ). Goldstein 1938. at y = 0. Both non-linear terms on the left-hand side of (8. provided that x denotes distance along the boundary and y distance normal to it. To actually determine the pressure distribution p(x).1).u» is large. actual changes in p across the boundary layer are stili much smaller. the whole procedure is then self-consistent. pp.4) is evidently correct if the Reynolds number R::::: u. Equation (8.1) is also valid in the case of a curved boundary. if the fluid were (mistakenly) treated as being entirely inviscid. for although the two pressure. gradients are comparable.6 will be almost equal to U(X) . The velocity at the 'edge' of the boundary layer in Fig. and by Bernoulli's theorem p + ~pU2 will be constant along a streamline at the edge of the boundary layer. save that ap! 2)' is then comparable in magnitude with op J &x. This may be demonstrated by writing the full Navier-Stokes equations in a suitable system of curvilinear coordinates (z . (8. It follows that 1 dp dU ~--=U-' pm dr ' (8. It is still the case that within the boundary layer p is essentially a function of x alone. by a factor O( (5 IL). and may indeed be put on a more formal basis (Exercise 8. simply because the boundary layer is so thin.1) are of order and in order that the viscous term be comparable magnitude we require that uu: i. 119-120). than changes in p along the boundary. for a substantia! pressure gradient in the y-direction is required to balance the centrifugal effect of the flow round the curved surface (Rosenhead 1963.e. The basic hypothesis (8.268 Boundary layers to obtain an order of magnitude estimate of the boundary thickness.8) . the argument is much as before. suppose that U(x) denotes the slip velocity that would arise. pp. 201-203. 8.

The exact solution is easily shown to be (8. u(O) ee 0.7. noty-+oo. An elementary differential equation with a 'boundary layer' Consider the following problem for a function EU"+U' u(y): = 1. Note that the limiting process here is ylv~-+oo. This important distinction may become dearer with the following elementary example. which would correspond to rocketing out of the laboratory and into the heavens. We must finally address the matter of how to ensure a 'match' between the flow velocity in the boundary layer and that in the inviscid mainstream. 9) b denoting a typical measure of the boundary layer thickness. u(l) = 2. impose on the boundary layer flow the condition u=» U(x) (8. denotes a small positive constant. proportional to v~.11) y Fi. In the sections which follow we shall.10) where f. and vice versa. to this end.Boundary layers 269 thus if U(x) increases with x the pressure pC.:} decreases. (8.10) for small e.g. . u The solution to eqn (8.8.

9) in this elementary example. . in two parts. exact solution (8. on this basis. by ignoring that term we lower the order of the system and are unable to satisfy all the boundary conditions. unless y is of order e.10) to Y=y/£. the small parameter e multiplies the highest derivative in the equation. We may variations of u in this boundary layer to be much more rapid those elsewhere by changing the independent variable in (8. in order satisfy the boundary condition there.... we obtain ub= 'outer' solution. the first being obtained by letting e~ 0 at fixed y. Tne solution may therefore be approximated. i. is needed near y = 0. Uo "" Y + c. It is instructive to take the analogy further by returning to eqn (8. O UM1 and this is the equivalent statement to eqn (8.270 Boundary layers Now. If we neglect the term eu" entirely.e.11). lim yIE-OV UBL = lim y . Notably.. and so is «r: for O<y < 1. and the second being obtained by letting E_"'O with y / e fixed. e-11f is extremely small. exploiting the fact that e is small. and on making this satisfy the condition uo(1) "" 2 we obtain an UoCv) = y + 1.10) and proceeding on an approximate basis from the outset. This procedure thus far is comparable with treating a high Reynolds number flow as being entirely inviscid. Here an 'inner' solution. 1. or boundary layer.. by a 'mainstream' and a 'boundary layer' adjacent to y = 0 with thickness of order [:: These two expressions represent particular limits of the full.

. Q Uo determines that A = 1..1). ~_y/~ as E -C>-O for fixed y I €. from the exact solution in keeping with our deductions 8. The boundary layer on a flat plate On inviscid theory a uniform stream approaching a flat plate at zero angle of incidence is unaffected by the presence of the plate.11).. satisfies dUi -+-=(). and the matching condition y~oc lim = lim y . so U(x) is a constant. in our simple example (and cf. (8. On making the inner solution satisfy the boundary condition u(O) = 0 we obtain u.. au au ciu (8. dy2 dY This is the equivalent of the boundary layer equation (8.Boundary With this scaling the regains its importance: previously negligible second layers 271 derivative 1 d2u 1 du E-~+--=l £2dy2 edY r so that to a first approximation d2Ui . ~ e-Y). Thus u = g~ as E -c>- 0 for fixed y.. = A(l u. Exercise 8.13) au aV ax By We seek a similarity solution in which u is some function of the single variable l) = y/g(x)..14) .12) (8. the inner solution u.1).3. The boundary layer equations then reduce to u-+v-=vax ay 8y2) -+-=0. (8.

17) = Uf'(rr) and v =_ 3'IjJ a. as in Fig.18) gJ(x). so k(x) "'" 0. We here take the similarity method of §2.x = -U(g'f +gf' = -u(grf = U(rIf' irg') at"!) ax.ng'· Here.13) by introducing a stream function 1. as we shall suppose. this is a assumption if.12) we obtain which simplifies to f'" + _. It is then more convenient to write 1/' in the form tjJ whence u = Ug(x)f(l1). so that V. But we want the plate itself to be a streamline. of course.3 a further by not attempting to guess the function g(x) in we show instead how it can be left to emerge in a rational way the calculation proceeds. V=-2tjJ/dX.!f_!L fr "" O. 2. If we write u in the form u = Uh(7J) we may integrate VJ = Ug(x) r to obtain h(s) ds + k(x). X"" 0 to x = 00./J(XJ y) such that u == 21jJ/oy.14. at 'r/ = 0. withf(O) "'"OJ (8. We first satisfy eqn (8.272 Boundary layers This implies that the velocity profile at any distance x from leading edge will be just a 'stretched out' version of the profile at any other distance x . On . f' denotes 1'(1]). l' U ' . = 0. but g' denotes substituting for u and v in eqn (8. the plate is semi-infinite. say. (8.

as we leave the boundary layer (cf._- CiV) ox y~O fl- au II oy Iy "" =() p.t = (8. [.21) has to be solved numerically.18). We dearly expect some such behaviour at the leading edge. of course. and the results are shown in Fig. (8.22) layer thickens the as indicated in Fig.l \-.8. 8.( \~vx U )1reO) (8. to obtain an ordinary differential equation for f as a function of 1]. the mainstream value.21) f(O) = f'(O) = 0.999936 at 11"" 5..14. we have found that y where rJ= (2vx/U)!' (8. conditions (8. if g vanishes value of x . We therefore choose d "" 0 to ensure that any such behaviour occurs at the leading edge. We must therefore choose gg'-which wo~ld otherw. and the third from the fact that u rn ust tend to U.20) and This equation must be supplemented by the boundary f'(DJ) = 1. the boundary layer thickness <5 is such that o=o(. = f.Boundary layers 273 Our aim is. eqn (8. and integrating gg' "'"v I U gives U • 2 =-+d .19) (8. As the boundary horizontal stress on the plate t.U\? . to sum up.97 at rJ = 3 and 0.19). The first of these stems from eqn (8. 2.23) .9»). According to eqn (8.20). The ratio u] U is 0.ise be ~ function ~f x-to ~e a coustan: Clearly the choice of vi U for this constant 1S convenient m that It rids the equation of all parameters of the problem. certain flow quantities such as au/ay = Uf"/g become singular. therefore. if only because on y "'"0 the velocity suddenly changes from U in x < 0 to zero in x> O. the second from eqn (8. The boundary value problem (8. Now.17).. Thus g(x) = (2vx/U)1 and.1 for some VX where d is an arbitrary constant.- (aU oy + -.

Taking into account both the top and the bottom of thg~' plate. dx ~ 2V2f"(O)pU'LR-. (8. 1.(\ is proportional to d..•.0). if the Reynolds number is~ ~%'. however. n= (0. The velocity profile in the boundary layer on a fiat plate .274 Boundary layers 1) I 4L ~--------------J ! I 3'- i --------~------~I ufU Fig. The drag.:: of f"(0) is 0.24 where R = UL/v. and vanishes as v-? 0: The numerical value-~: Ln i '.9). 8... (Here we have used eqns (6.•. .) Application of the theory to It finite fiat plate of length L It is natura! to hope that the above similarity solution will hora[ reasonably well for a finite plate of length L. both in respect of the expression (8. even if behaviour off' a different kind must be expected near and beyond the trailin~~ edge. we obtain for the drag ::~ D~2 r t.:. of course.7) and (6.8.4696. Thus D is proportional to L1. decreases with x.?' The agreement between boundary layer theory and experimenJ~ is very good.24) for the drag~i and in respect of the details of the velocity profile. rather than to because the velocity gradients at the plate decrease with X~& corresponding to the thickening of the boundary layer. Thi~~: agreement does breakdown. with.'.

Sign up to vote on this title

UsefulNot useful