Vous êtes sur la page 1sur 30

Comparative solutions to the integralapproximate thermal boundary layer equations for a at plate

J. D. Lewins Engineering Department, University of Cambridge, Cambridge CB2 1PZ, UK E-mail:jdl@eng.cam.ac.uk


Abstract The at plate in uid ow is the paradigm of heat transfer problems. The polynomial approximation of Pohlhausen to the Blasius solution for the ow leads to messy expressions in the integral formulation for the associated heat transfer between plate and ow, together with a continuation of the paradoxical results that the termination in the approximate solution implies. If the distribution is approximated in exponential form, the required integrals are easy and may safely be attempted by a lecturer at the board or screen and the anomalies eliminated. The results are perhaps less accurate than the Pohlhausen approximation but in either case would be replaced with exact values of the ow equations, obtained with considerably more difculty. Keywords at plate; heat transfer; Pohlhausen; laminar ow; exponential approximation

Introduction Teachers of convective heat transfer will generally wish to develop the boundary layer formulation for a at horizontal plate in laminar ow along the lines of the classic work of Prandtl [1]. It is desirable to limit the analytical complexity of the blackboard work and this paper is directed at such a teaching need. The exact solution, as given numerically by his pupil Blasius [2], has substantial difculties that makes it, I believe, unsuitable for lecture work at a second- or third-year undergraduate level, and more appropriately dealt with by means of reading, written example or problem work. It is natural therefore to employ one of the standard approximations for the velocity distribution in the boundary layer, typically a thirdorder Pohlhausen polynomial [3], to obtain the ordinary differential form of the integral approximation. However, such a nite polynomial approximation (or the sinusoidal approximation) is not only sufciently complicated to do at the blackboard as to invite errors, but suffers two further disadvantages: (1) it satises only a nite number of boundary conditions, whereas the exact Blasius solution satises innitely many; (2) as a consequence, it leads to a nite boundary layer thickness, whereas the exact Blasius solution thickness is innite, calling for the denition of an approximate nite boundary layer thickness, say 99% recovery of free-stream velocity along the plate. There is an alternative, a horizontal speed distribution based on an exponential fall-off from the plate. This distribution is readily integrated over and, therefore, suitable for blackboard work. It has the same property as the exact solution of not having a nite boundary layer thickness (except in the sense of a displacement thickInternational Journal of Mechanical Engineering Education 32/4

316

J. D. Lewins

ness) and it has a development for the thermal boundary layer that is considerably more satisfactory than the polynomial approximations normally used. Although it has the correct functional trend, it does not have the accuracy desired for the boundary layer itself, but this may be thought acceptable at a pedagogic level, certainly if the exact Blasius coefcients are available for quotation anyway. The thermal boundary layer equation derived from the integral approximation, typically based on the same nth-order polynomial t, allows a comparison with the thickness of the momentum boundary layer. The integral approximation equation for the (momentum) boundary layer is readily solved, but the equation for the thermal boundary layer is more complicated. Considerable simplication is possible, especially if the exponential distribution is assumed in the solution of the thermal equation, if the common shape factor is exploited, which is possible for the plate at uniform temperature. What is the boundary layer? The general concept is that it is the layer close to the surface in which viscous laminar (and later turbulent) effects are signicant, so that outside it viscous losses can be ignored. By an order-of-magnitude analysis, Prandtl [1] established that for a sufciently large Reynolds number (based on distance from the leading edge of the plate), the horizontal component of the momentum equations dominates over the vertical component, and we can expect a similar velocity distribution for the horizontal speed as we pass through the boundary layer from the plate upwards. In practice there is no exact envelope to dene the boundary layer and a practical denition uses a 1% approximation, that is, that surface where the horizontal speed that was brought to zero at the plate rises to 99% of the free stream speed. More generally we consider boundary layers dened as contours of constant horizontal speed, expressed as some fraction of the free-stream horizontal speed. Just as there is taken to be a (momentum) boundary layer, of thickness d(x), we suppose that there is a thermal boundary layer dt(x) representing the limit of the region in which the temperature difference between free stream and plate is signicant. Again, a practical denition is in terms of a 99% value. In this paper we develop the integral approximation for the third-order Pohlhausen polynomial and the exponential approximation in parallel, to contrast the algebraic complexity and compare the accuracy achieved against the exact Blasius results. In addition, a full account of the solution for the exponential approximation of the transient thermal boundary layer problem is provided, together with comparisons with the other Pohlhausen-type approximations. Fundamental equations The semi-innite plate, in a steady state, incompressible, laminar ow with free stream speed u and no pressure gradient, is measured along the positive x-axis. We have a two-dimensional problem with velocity components u and v in the x and y directions respectively (Fig. 1). Continuity requires u + =0 x y
International Journal of Mechanical Engineering Education 32/4

(1)

Thermal boundary layer equations

317

Fig. 1

Semi-innite plate in viscous ow.

Since there is no ow vertically through the plate and allowing that the horizontal ow is brought to rest at the plate by the surface drag, u and v are zero along the positive x-axis. Prandtls suggestion for the boundary layer [1] involves a thin layer in which viscous effects are signicant, close to the plate, with a free stream that can be idealised as viscous-free above it. The boundary layer grows in thickness downstream of the leading edge of the plate. This assumption requires a large value for the Reynolds number, based on the distance along the plate1. It is next argued that, within the boundary layer, the shape of the distribution of the horizontal component of speed u in the vertical direction y is similar at every x, but scaled as the boundu ary layer thickness (however dened). It follows that, at xed y, is negative, x whereas u itself is positive. Hence continuity requires the second term in equation 1 to be positive and, as the vertical component of velocity cannot penetrate the plate, v is necessarily positive, directed upwards. One of the features of any boundary layer solution, therefore, is to determine how this velocity relates to the (sloping) boundary layer itself, that is, whether the boundary is a stream line or is crossed by the ow. The force or momentum equation is obtained by assuming zero pressure gradient and neglecting vertical or y-direction forces, as follows from the thin-layer assumption. Momentum, measured in the x-direction, assuming constant Newtonian dynamic viscosity or thermal diffusivity , then satises the reduced NavierStokes equation: 2u u u 2u +u + = x y y y2 (2)

See Lewins [4] for a discussion of the Blasius paradox and the necessary size of the Reynolds number.
International Journal of Mechanical Engineering Education 32/4

318

J. D. Lewins

Fig. 2

Flat plate at wall temperature Tw, heated by free stream at temperature T.

In this we may substitute the continuity equation (1) to obtain the working form of the momentum equation: u u u 2u + = x y y2 (3)

Temperatures, T, are measured (Fig. 2) relative to the plate, at temperature TW, which is being heated by the free stream at constant temperature T. We then write a non-dimensional temperature that has the same form as the non-dimensional horizontal speed distribution: q= T - Tw T - Tw

That is, at the wall the non-dimensional temperature is zero and in the free stream it is unity. The linearity with temperature allows such a solution to be adapted to other cases. The uid has constant thermal diffusivity a. The energy equation is simplied by ignoring the release of thermal energy in dissipation and considering conduction only in the y-direction, thus obtaining: u q q 2q u + +q + =a 2 x y x y y (4)

where a = u

l is the thermal diffusivity. Substituting the continuity equation yields: r cp (5)

q q 2q + =a 2 x y y

Inspection of equations 3 and 5 for the case where the Prandtl number Pr =

=1 a shows that these equations are identical as between u and q. If therefore the boundInternational Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

319

ary conditions are identical (implying a uniformly heated plate), the distribution of temperature, q, will be the distribution of horizontal speed, u. More generally for Pr 1 but for similar boundary conditions, we can expect that the shape of the two boundary layers, momentum and thermal, will be similar along the x-axis, that is, dt the ratio of their heights, z = , will be a constant, Z, independent of x. It is this d ratio, z, which determines heat transfer properties in relation to uid ow and wall stress or surface friction properties. Unity of the ratio at unit Prandtl number will not follow if the boundary conditions are not similar; for example, if the velocity is brought everywhere to zero on the surface of the plate but only part of the plate is cooled (or heated) relative to the free stream temperature, then we must revert to z(x). But this similarity will simplify some later analysis. When the Prandtl number is greater than unity (e.g., water, where Pr = 7), the momentum diffusivity should exert its inuence further into the free stream and the ratio, Z, is expected to be less than one, as the momentum effect is felt deeper at higher viscosity. Vice versa, when the Prandtl number is less than unity (just less for most gases and substantially less for liquid metals) the ratio Z should be greater than one, as the thermal effect is felt deeper at higher conductivity. It is to be expected that the nature of the solutions, therefore, might change around unit Prandtl number. Some exact considerations similarity The expectation is that the horizontal velocity, u(x, y), will vary vertically in a similar fashion at every horizontal station x. To demonstrate this we can use the Blasius similarity variable: h=y u y = x x xu = y 1 2 Re x x (6)

u = g(h) can be represented as a function of this variable alone, u thus reducing the partial differential equations to ordinary differential equations. A tedious substitution shows this to be true; it is essentially Blasiuss treatment of the stream function, which is available in various texts [5, 6] and not detailed here. In outline, one can use the assumed form to integrate the continuity equation to nd v(h). Then the momentum equation reduces to the form: and suppose that
h

- gg h + g g (z)zdz = 2 g
0

(7)

which reduces to Blasiuss equation putting g = f . Hence with suitable boundary conditions, the momentum equation has a similarity solution. A neater way is to employ group theory [7] to express the invariance of the system under dilation. To do this, express the independent and the dependent variables as dilated to a new variable set as X = xegx, Y = yegy together with U = uegu, V = vegv, Q = qegq. We note that the velocity components are zero on the plate but that the
International Journal of Mechanical Engineering Education 32/4

320

J. D. Lewins

horizontal speed is unchanged deep into the uid, away from the plate. Hence we immediately have gu = 0. Substitute the dilated variables into equations 3 and 5 and require the form of the equations to be invariant: ue -g x u u 2u + veg v -g y = ve -2g y x y y2 (8) (9)

Hence g y = g x 2, g v = -g y and ueg q -g x q q 2q + veg v +g q -g y = ae 2 (g q -g y ) 2 x y y

(10)

Hence indeed gq = 0. Now it follows from equation (10) that any function of y/ x is an invariant under dilation. Not only can we now formulate a symmetry variable, say y/ x , made u non-dimensional as the Blasius variable h = y , but we can argue in an order-ofvx magnitude sense from the second and third terms of equation (10) that an approxiu mate solution for will fall off exponentially with y. This might be taken as the y motivation for the exponential approximation.

Integral equations Blasius was able to develop the exact solution to the boundary layer equations numerically by employing the similarity variable. There are, however, substantial numerical difculties in integrating the ordinary but third-order and non-linear equation this leads to for the velocity distribution and associated boundary layer thickness. Hence it is customary to use an empirical approach in which a form of velocity distribution is assumed, based on an unknown boundary layer thickness, and to employ the same spirit of the boundary layer approximation to formulate an ordinary differential equation (of much simpler form) that can be solved to yield the unknown thickness as a result. The technique is to integrate the continuity and horizontal momentum equations, later the energy equation, over the vertical face of a control volume to obtain a rstorder differential equation for the unknown thickness. Fig. 3 shows the construction for these integrals. To do this at the blackboard runs some risk of algebraic error unless the assumed velocity distribution is sufciently simple. One simple assumption is for a linear fall-off. But although this is acceptably simple, it does not give satisfactory results; the exact distribution in the boundary layer certainly does not look linear, from zero at the plate to free-speed at the boundary. The most usual form is the third-order polynomial (one of a family suggested by Pohlhausen) and we

International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

321

Fig. 3

Construction for integral equation derivation.

compare this in the present paper with the exponential fall-off. Order-of-magnitude analysis (which no doubt has been demonstrated before reaching this point in the syllabus) shows that the boundary layer thickness goes as the square root of distance along the plate, leading to the parabolic form of boundary layer thickness2. We need to examine both of the proposed boundary layer velocity distributions to obtain the coefcients for this thickness, followed by similar calculations for the thermal boundary layer. The exact Blasius solution is developed with the aid of the stream function concept. Approximate solutions express the velocity similarity by means of an assumed function, independent of distance x: u y = g d u (11)

The connection with the exact Blasius formulation is that his similarity function, 1 1 y f(h), has g = g h = f (h), where d = Cx Re x 2 , in terms of the Reynolds C d u x number based on distance x (i.e., Re x = ). Then, putting g = y/d, the momentum integral for the integral approximation method becomes:

As Bejan remarks, it would be good if textbooks drew the parabola accurately at the tip, i.e. rising vertically. It is this behaviour that ultimately leads to an innite local heat transfer coefcient at the tip. Admittedly, the assumptions of the Blasius model are unfounded at the tip, since here the Reynolds number is not large. [8]

International Journal of Mechanical Engineering Education 32/4

322

J. D. Lewins

d u (u - u)udy = v y dx 0

2 = u y=0

d d [1 - g(g )]g(g )dg = u g (0) dx 0

(12)

where the right-hand term originates from the stress at the wall. The integral can be stopped at the boundary layer thickness d when this is nite in the approximation. The evaluation of this integral yields an equation for the boundary layer thickness. y Suppose the temperature distribution is given in the same form as q = g , dt where q is the non-dimensional temperature rise into the stream from the plate, with dt the thermal boundary layer thickness. Then, in similar fashion with g = y/d, an energy integral yields: q d (1 - q )udy = a y dx 0

= u
y=0

g d 1 - g z g(g )dg = ag(0) dx 0

(13)

where a = l/rcp is the thermal diffusivity and the right-hand side expresses the heat transfer by conduction through the uid at the wall. Into these equations we need to substitute an approximation for the vertical distributions. Approximations for the integral equations Several approximations of the Pohlhausen type are in use. They include: y y rst-order, g = d d y 3 y 1 y3 third-order, g = d 2 d 2d y y y 3 y 4 fourth-order, g = 2 - 2 + d d d d together with the sinusoidal approximation that the similarity function is y p y g = sin d 2 d In all these, the similarity function is undened beyond the nite boundary layer thickness. Fig. 4 shows the various shapes of these approximations. It will be seen that the linear approximation is poor, the sinusoidal approximation is close to the third order, and these and the fourth order bracket the true value. These various approximations are motivated by an attempt to match the boundary conditions of the ow speed. These include, of course, that u vanishes on the plate and reaches u at innity. Note that the Pohlhausen polynomial approximations and the sinusoidal approximations have a nite boundary layer thickness and so have a recovery to the free stream speed at nite distance. In addition, equation 3 shows that if the rst-order differential of u vanishes at innity, so does the second. Furthermore, dn y by repeated differentiations, the derivatives n 0 at innity (or at the nite dx d
International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

323

Blasius exact 99% together with exponential and Pohlhausen approximations for the horizontal speed distribution in the boundary layer

Fig. 4

Approximate horizontal speed distributions. (The Blasius exact and the exponential are drawn for the 99% empirical thickness denition.)

boundary layer thickness in the Pohlhausen approximations). Hence the exact solution implies an innite not a nite polynomial to match all boundary coefcients. An alternative which we promote here is the exponential distribution. The exponential approximation offers a t to the velocity distribution that falls off exponentially, that is, with a boundary layer parameter that measures an e-fold reduction in the discrepancy between the free-stream ow and the zero velocity at the wall: we y have g = (1 - e - y d ) which takes the speed from zero at the plate to the free speed d at innity. This means that there is an innite boundary layer in the sense that the exponential term takes the discrepancy to zero only at innity; the parameter d is therefore not a total thickness as such but the e-folding thickness, analogous to a time constant of the decay in velocity difference. However, the exponential solution, while being an innite polynomial that can match all boundary conditions, does not
International Journal of Mechanical Engineering Education 32/4

324

J. D. Lewins

satisfy the equation exactly and indeed falls off faster than the correct asymptotic Blasius solution. In representing the exact Blasius and the exponential approximation in the same gure as the nite boundary layer approximations, we have drawn the former in terms of their empirical 99% thickness denition. It is seen that the exponential is a poorer approximation within the empirical boundary layer than Pohlhausen forms. We develop the third-order Pohlhausen approximation and the exponential approximation in parallel, to compare their didactic properties. The integral equation for the boundary layer thickness is: d u (u - u)udy = dx y 0

(12)
y=0

Correspondingly the integral equation gives both a third-order Pohlhausen approximation and an exponential approximation. The equation for the third order (Pohlhausen) polynomial is:
2 u

d 3 y 1 y 3 3 y 1 y 3 d y + g dy = u 1 dx 2 d 2 d 2 d 2 d dy d 0

=
y=0

3 u 2d

We note that the integral is not to be taken to innity but only to the nite boundary layer thickness. As a result, the equation integrates to give: d 1 9 3 3 1 3 3 d 2 g - 2 g 3 - 4 g 2 + 4 g 4 + 4 g 4 - 4 g 6 dg = 2ud dx 0 = or 2d d 1 3 1 d d 7 1 d 39d d - 8 + 10 - 28 = dx 4 10 - 7 = dx 280 dx dd d 2 280 = (d ) = 3 dx dx 39 u
1 1 280 Re x 2 = 4.641 x Re x 2 13 1

(14)

and since d(0) = 0, d=x (15)

The equation for the exponential approximation is:


2 u y y y - - - d d 1 - 1 + e d 1 - e d dy = u 1 - e d y=0 dx dy 0

(16)

This integrates to give:


y y -2 d d 1 e d - e d dy = dx d 1 - 2 = ud dx 0

(17)

or

d 2 (d ) = 4 ; d (0) = 0 dx u

(18)

International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations


1

325

Hence d = 2 x Re x 2 The exponential derivation is seen to be more concise. But we also need to compare this coefcient of 2 for the exponential solution and 4.64 for the Pohlhausen to the exact Blasius solution and the other Pohlhausen-type approximations. At rst sight, this coefcient of 2 is very different to the more usual factors of approximately 5 seen in such expressions, but we must recollect that our exponential boundary layer thickness parameter is only for an e-fold diminution of the effect of the plate on horizontal velocity. Like the exact solution, therefore, the effect vanishes only at innity, whereas the Pohlhausen approximations have a nite boundary layer thickness. If we compare the distance at which the exact Blasius solution has a reduction by e-1, we nd the exact coefcient is 1.7208, instead of our coefcient 2, an improved comparison. If we compare the conventional 99% recovery of free-stream horizontal speed, the exact solution has a coefcient 4.918 and the exponential approximation 9.2103. The Pohlhausen approximations are compared in Table 1 and are seen to be more accurate in general in this respect. Because of the conceptual difculty with a boundary layer thickness that extends in principle to innity, it is customary to dene the displacement thickness, d*, where this is the displacement of the stream lines of the original potential ow innitely far from the plate. This thickness is given by:

d* = 1 0

u dy u

or, for the exponential distribution

d * = e d dy = d
0

(19)

It is notable, then, that the e-folding boundary layer thickness is identical to its displacement thickness. Table 1 lists corresponding displacement thicknesses. The nal column gives the (one-sided) wall stress. Comment Only the exponential approximation has the exact Blasius form of the boundary extending to innity. The exponential coefcients are of the right order of magnitude but not as accurate as the Pohlhausen approximations. The exponential is not a good approximation for the conventional 99% boundary layer thickness3. The displacement thickness is not as well given by the exponential solution (15% error) as by the Pohlhausen approximations (some 12% error). On the other hand, the required integrations are notably easier to evaluate in the exponential approximation

The exponential form would match the exact Blasius better than the Pohlhausen forms if the convention was changed to, say, 99.9%.
International Journal of Mechanical Engineering Education 32/4

326

J. D. Lewins

TABLE 1
Solution

Comparison of exact and approximate Blasius solutions


1 2 C = d1 Re x x 1 2 C99% = d 99% Re x x 1 2 C* = d* Re x x

CW =

1 tW 2 Re x 2 ru

Blasius exact Pohlhausena rst order third order fourth order sine Exponential approximation
a

3.464 4.641 5.836 4.795

4.918 3.430 4.105 4.838 4.363 9.2103

1.7208 1.732 1.740 1.752 1.741 2

0.332 0.289 0.323 0.343 0.327 0.5

Pohlhausen values adapted from Schlichting [5].

than in the third-order Pohlhausen approximation. But if only the (momentum) boundary layer is to be considered, it is doubtful that the advantage of algebraic simplicity would outweigh the loss of accuracy in taking the exponential over the Pohlhausen forms, unless the exact results were available for reference anyway. The feature that the exponential displacement thickness is the same as the assumed efolding thickness may be taken as an additional conceptual advantage, however. Vertical velocity solutions By integrating the continuity equation vertically and noting that the vertical velocity component starts at zero on the plate, we have: u u dd ( x, y)- = dyy = g g (g )d g x 2x 0 0 Taking d = Cx Re
1 2 y d y

(20)

in any approximation gives:


y 1 d 2

( x , y) =

u C Re 2

zg(z)d z
0 y y d ( )d zg z 0 - g(z)d z 0

(20)

so that the slope of a stream line is given by:


y

C -1 1 y C -1 1 d zg (z)d z = Re x 2 = Re x 2 y y u 2 2 g g 0 d d

(21)

y On the other hand, the slope of the family of boundary layers, g = gn, = Cn, const, d is given by: y d = Cn x n x
International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

327

TABLE 2
Function Vertical speed v Limit ata

Vertical velocity components


Exponential
1 - y -y u Re x 2 1 - 1 + e d d (y = : C = 2)

Pohlhausen, third-order
1 3C y 2 1 y 2 u Re x 2 1d 2 d 8 (y = d, C @ 4.64) 1 1 3C u Re x 2 = 0.87u Re x 2 16 1 y 2 14.84 y 2 d 1 1 y 2 d 2 14 Re x 3 d 2.32 y < 1 d 2 Re x

()

()

( )

u Re x 2 1 Re x 2 < 1 Re
1 2 x 1

Slope of stream line

Slope of boundary contour

() () () ()

y d 1 - y e d - 1

( dy )

The coefcient for the exact Blasius solution is 0.86 . . . .

These forms are compared for the exponential and Pohlhausen forms in Table 2. It is seen that (as in the exact solution) although there is a rising component of velocity, stream lines enter the boundary layer contours from above. The exponential form mirrors the exact Blasius solution in predicting this vertical component out to innity (with paradoxical results) whereas the Pohlhausen forms are undened beyond the nite boundary layer. But at the Pohlhausen boundary layer, the Pohlhausen forms also predict a nite vertical component and hence a paradoxical increase in the free stream. Locally exact solutions The exact Blasius energy equations were rst solved by Pohlhausen. His technique may be applied to the approximate equation but gives poor results. The method supplies answers of the correct form for the exponential approximation but it fails altogether for the polynomial approximation, which is therefore treated separately in the Appendix to this paper. The vertical velocity is given by equation 20, that is:
y

u u dd ( x, y) = dy = gkg g (gkg )d gkg x 2x 0 0 Then the momentum equation becomes:


y

(22)

y y y d y d 1d y 2 2 2 -u g g 2 zg (z)d z = u 2 g + u g d d d 2x d d 2x d d 0 or
y g ud y d (z)d z = - g + zg g 2x d 0

(23)

(24)

International Journal of Mechanical Engineering Education 32/4

328

J. D. Lewins

For the energy equation, the non-dimensional temperature is also a function of y the similarity variable or q = q , a function to be determined. The equation then d becomes:
y d u q y (z)d z = a 2 q - g + zg 2 xd d d 0 q g whence and since q(0) = 0, for the geometrically similar case, = Pr q g

(25)

q (0) y q = d [g (0)]Pr

y d

[g(z)]
0

Pr

dz

(26)

Substituting the exponential approximation gives g(z) = e-z; g(0) = 1 so that q () = 1 - q (0) e - Pr z d z =
0

q (0) Pr

(27)

and
y - Pr y q (0) = Pr;q = 1 - e d d

(28)

giving the result that the thermal boundary layer is a fraction, 1/Pr, of the (momentum) boundary layer, dt = d/Pr, z Z = 1/Pr. Now, this behaviour is only approximately that of the exact solution given by Pohlhasuen to the energy equation, based on the Blasius exact solution. Certainly the thermal boundary layer is bigger than the boundary layer when the Prandtl number is smaller than one, but the behaviour far away from unity is not the standard result. We conclude that this method, although it is valid for the exact Blasius problem, is not helpful here. (It is certainly not suitable in the Pohlhausen approximation.) One might well ask why this exact technique gives poorer results when applied to an approximate solution than does the integral equation development, given next. It might be expected, however, that an integration averages out the errors inherent in any approximate solution based on an assumption of the exact velocity distribution. This suggests that an even better approximation might be gained by using a variational procedure. Energy equations and the thermal boundary layer The integral energy equation is obtained from equation 5 in the form d q (1 - q )udy = a y dx 0

(29)
y=0

where the right-hand side is proportional to the conduction at the wall. The thermal boundary layer is dened again as a region where the thermal inuence of the wall
International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

329

is felt in terms of a non-dimensional distance, y/dt. This can be approximated by the Pohlhausen forms and by the exponential form. We again develop the two forms in parallel. Note that the Pohlhausen approximation has an integral taken to a nite limit. This limit differs as the Prandtl number is larger or smaller than unity. Hence this requires the development of two cases for completeness, even if this often neglected in textbooks. And in the case where the thermal boundary layer is thicker than the (momentum) boundary layer, it is necessary to break the integral into the two parts: where the horizontal velocity is changing with y, and where it has reached dt the free-stream velocity, u, In both forms we put z ( x ) = Z , the ratio of the d boundary layer, which is smaller than unity for Prandtl numbers greater than unity. Energy integral equation formulation Third-order Pohlhausen: y 1 y 3 y q = g = - dt 2 dt 2 dt
2

The energy integral equation is: u


3 3 H d 3 y 1 y 3 y 1 y 3a + - dy = 1 2 d t 2 d t 2 d t 2 d t 2 d t dx 0

Case (a), Pr > 1, dt < d, H = dt: = d 3 2 3 4 3 a d 20 z - 280 z = 2 udz dx dd -140 280 x ,d 2 = dx 13 m 13 m

where d =

Case (b), Pr < 1, dt > d, H = d: d 3 1 3 9 2 3 4 3 1 g + g + 3 g 4 - 3 g 6 dg d g - g dx 0 2 2 4z 4z 4z 4z


1 3 1 3 3 1 d 3 1 3 + dz 1 - g + g 3 dg = 2 dx d 4 - 8 - 4z + 20z + 20z 3 - 28z 3 2 1 z 1

1 3 1 1 1 d 5 3 4 + dz 1 - - 1 - 2 + 1 - 4 = + d 3 z 4 z 8 z dx 8 5z 35z 3 1 3 a 3 + d z -1+ = 8 4z 8z 3 2 udz


International Journal of Mechanical Engineering Education 32/4

330

J. D. Lewins

where d dd 140 280 x = ; d2 = d x 13 u 13 u


y

Exponential:
y q = g = 1 - e dt dt

The energy integral equation is:


y - - d a U e dt 1 - e d dy = d t dx 0 y

or
y y 1 1 y y - y + d + d t - d + dt a d d d d2 -d t e dt + e d t - e d t d dy = e = = dd t dx 0 dx d x d t + d ud t 0

or

d dz 2 a = d x z + 1 udz

where d dd x = 2 ;d 2 = 4 dx u u

We conclude that the exponential form is considerably simpler than the Pohlhausen polynomial when it comes to the thermal boundary layer. Even at this stage, the polynomial has two distinct ranges: we have not spelt out the integrations needed in case (a) as the results may be found in most textbooks. The solution to these equations is readily obtained in the case of geometric similarity by assuming a constant ratio of thermal to momentum boundary layer, dt z = Z , const. This is known to be exact for unit Prandtl number in the Blasius d model. The assumption satises the boundary condition. We obtain an equation for Z and its limiting values: Energy integral equation solution for boundary layer ratio Z Third-order Pohlhausen: Case (a), Z 3 1 5 13 1 Z = : 14 14 Pr
1 13 - 1 Pr 3 = 0.9756 Pr 3 14

(30)

Limiting cases, Pr = 1, Z = 1: Pr , Z 3

2 13 2 1 Z = 0: Case (b), Z 4 - Z 3 + 5 35 Pr 35
International Journal of Mechanical Engineering Education 32/4

(31)

Thermal boundary layer equations

331

Limiting cases, Pr = 1, Z = 1: Pr 0, Z Correspondingly,


1 1 1 1 dt 13 14 =3 2 5 Pr 3 Re x 2 = 4.58 4 Pr 3 Re x 2 x 14 13 1 13 - 1 Pr 2 = 0.609 4 Pr 2 35

for large Prandtl and


1 1 1 1 dt = 2 2 Pr 2 Re x 2 = 2.828 4 Pr 2 Re x 2 x

for small Prandtl. Expansion around the solution Z = 1, Pr = 1 gives the rst-order approximation 39 - 1 Z = 1 - Pr 3 - 1 with 5% accuracy at Pr = 0.75, representative of real gases. 43 Exponential: Z3 = 1 [ Z + 1] 2 Pr (32)

Limiting cases, Pr = 1, Z = 1: Z 0, Pr , Z 3 Z , Pr 0, Z
1 1 = 0.7937 Pr 3 2 Pr 1 1 = 0.7071 Pr 2 2 Pr

These have a break point, where the asymptotes are equal, at Pr = 0.5. Correspondingly,
1 1 1 1 dt 3 = 4 Pr 3 Re 2 = 1.587 4 Pr 3 Re x 2 x

for large Prandtl and


1 1 1 1 dt = 2 Pr 2 Re 2 = 1.4141 Pr 3 Re x 2 for small Prandtl. x

Expansion around the solution Z = 1, Pr = 1 gives the rst-order approximation 6 -1 Z = 1 - Pr 3 - 1 , with 5% accuracy at Pr = 0.75, representative of real gases. 5
International Journal of Mechanical Engineering Education 32/4

332

J. D. Lewins

Comment The polynomial expression is now considerably more complex than the exponential, since it has two ranges, for Prandtl greater and smaller than one, whereas the exponential form is evaluated readily over the whole range of Prandtl numbers. The fth-order polynomial in the rst Pohlhausen range is often approximated by assuming that Z is small and hence dropping the term in Z5, giving Pr , 13 - 1 Z3 Pr 3 . This is valid for large Prandtl numbers, and indeed is often further 14 approximated as Pr , Z Pr 3 with the virtue of giving the correct result at unit Prandtl number. But since the expression is commonly used for Prandtl numbers around unity, this seems unsound (as well as unnecessary), even if the resulting error Z 1 is small in heat transfer terms. The exponential equation Z 3 =0 2 Pr 2 Pr necessarily has a real root and inspection shows that this root has the same sign as the Prandtl number (i.e. it is positive). Table 3 shows the exponential solution in exact and asymptotic forms. The table will give some indication of the accuracy and range of validity of the asymptotic approximations; the Pohlhausen solution has much the same sensitivity. Note that neither asymptotic approximation is accurate for unit Prandtl number. It is also notable that for a realistic Prandtl number for typical gases (0.7) the large Prandtl number approximation is substantially more accurate than the small number approximation, a fortuitous range of validity for values representative of real gases. Even more strange is the common practice of using the Pohlhausen asymptotic result for a large Prandtl number to predict values in the range of Prandtl numbers less than unity. Again, the expressions are not sensitive to such a cavalier treatment.
1

TABLE 3

Coefcient of boundary proportionality as a function of Prandtl number in the exponential approximation


Z asymptotic

Pr 505 103 550 40 6 1.5 1 0.7608 0.09375 0.03906 5.5 10-3 50.5 10-6 0

Z exponential exact 0 0.01 0.1 0.25 0.5 0.8514 1 1.15 2 4 10 100

Z expansion around Pr = 1

1 2 Pr

1 2 Pr 0

0.0195 0.9248 1 1.1045 1.6549

0 0.009967 0.09687 0.2321 0.4368 0.6934 0.7937 1.1538 1.7471 2.3393

0.5774 0.7071 1.2455 2.3094 3.5777 9.5346 99.5037

International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

333

TABLE 4
Prandtl number 550 40 15 10 7 6 1.5 1.1 1 0.9 0.8 0.761 0.7 0.6 0.094 0.039 0.005
? a

Thermal to momentum boundary layer thickness ratio z


Exponential 100% 0.1 (0.086)a 0.250 0.356 0.414 0.474 0.500 0.851 0.963 1 1.043 1.094 1.116 1.154 1.229 2.700 4.033 10.467 (10)a Third-order exact 0.120 0.286 0.397 0.455 0.513 0.541 0.868 0.967 1 1.038 1.082 1.102 1.135 1.199 2.451 3.562 9.110 Pohlhausen trunc. 0.119 0.285 0.396 0.453 0.510 0.537 0.852 0.945 1 0.642 0.681 0.699 0.728 0.787 1.988 3.086 8.619 Pohlhausen asymptotica,b (0.026) (0.096) (0.157)

Blasius exact 99% ? 0.299 0.394 0.454 0.503 0.536 0.862 0.963 1 1.039 1.089 1.110 1.142 1.212 ? ? ?

(2.151) (2.877) (5.705)

Digital integration fails. Asymptotic value in parentheses. b Pr < 1, 3 13 14 Pr ; Pr > 1, 13 35 Pr .

We will nd, however, that the approximation of restricting the expression to the third order is almost essential in the further use of the Pohlhausen theory for the general case. We are here assuming uniform temperature along the plate and hence constant ratio z = Z. Table 4 compares the predictions for the ratio of boundary layers for a variety of Prandtl numbers. Integrations of the Blasius exact results were based on Schlichtings values and second-order quadrature.4 Since the exact Blasius ratio of boundary layers is not generally available, Fig. 5 plots these results together with the exact and asymptotic exponential values for Z. We can also compare the approximations, Pohlhausen and exponential, for accuracy against the exact Blasius result in terms of the prediction of the local heat transfer coefcient (Table 5). These results are derived, of course, from the derivative of the temperature distribution at the plate wall. It can be noted that the asymptotic

4 Our integration of the velocity distribution at a spacing of Dh = 0.4 agrees with Schlichtings improved Blasius values within 1 part in the fth decimal place. Subsequent comparison with Pohlhausens integration for Nusselt numbers, however, agrees only within 0.3%, but his values are presumably based on the unimproved Blasius gures. Outside the values used by Pohlhausen (Pr = 0.615) numerical integration degenerates presumably which is why he limited the range of Prandtl numbers and the asymptotic behaviour is more reliable.

International Journal of Mechanical Engineering Education 32/4

334

J. D. Lewins

Fig. 5

Thermal to momentum boundary layer ratio Z for exponential approximation with exact Blasius 99% values.

approximation for the exact Blasius solution taken for large Prandtl numbers actually predicts the Nusselt number for Prandtl numbers down to Pr = 0.6 with little more than 1% error, whereas the asymptotic expression for vanishing Prandtl number is some 20% in error at this value. For the geometrically similar problem, therefore, the thermal boundary layer thickness, as a function of Prandtl number, is: d t = Z (Pr)d (33)

for dt as a ratio to the momentum boundary layer. The more conventional form (which we think less convenient) is to write:
1 dt = CZ (Pr) Re 2 x

(34)

where C is the appropriate coefcient (see Table 1). One can dene a thermal (or energy) displacement thickness by analogy with the displacement thickness and it is readily shown that this is equal to the thermal boundInternational Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations


1

335

TABLE 5
Asymp. Pr 15 10 7 1.1 1 0.9 0.8 0.7 0.6 Blas. exact 0.835 0.730 0.645 0.344 0.332 0.320 0.307 0.293 0.276 Fitted
1 1 3

Accuracy of heat transfer predictions: Nux Re x 2


Asymp. Pr
1 2

Asymp. Expo. exact 1.401 1.208 1.057 0.519 0.500 0.479 0.457 0.433 0.407 Pr
1 3

Asymp. Pr
1 2

Pol-3 exact in Z 5 0.814 0.711 0.630 0.334 0.323 0.312 0.298 0.240 0.270 truncated Z = Pt 0.797 0.697 0.619 0.334 0.323 0.312 0.300 0.287 0.272
-1 3

Asymp. Pr
1 3

Asymp.
1

Pr

Pr 2 0 2.053 1.677 1.402 0.556 0.531 0.502 0.474 0.444 0.411

Pr 3C 0.819 0.715 0.635 0.343 0.332 0.321 0.308 0.295 0.280

0.836 0.730 0.648 0.350 0.339 0.327 0.315 0.301 0.286

0 3.090 2.523 2.111 0.837 0.798 0.757 0.714 0.668 0.618

1.554 1.357 1.205 0.650 0.630 0.608 0.505 0.559 0.531

0 2.739 2.236 1.871 0.742 0.707 0.671 0.633 0.592 0.548

0.817 0.714 0.634 0.342 0.331 0.320 0.308 0.294 0.279

ary thickness itself in the exponential approximation. In contrast, the third-order 5 Pohlhausen approximation gives d * = d t . t 8 With these results, we have an expression for the local Nusselt number: Nu x = Nu x = hx l
1 q 1 2 Re x y 0 CZ (Pr)

(35)

Blasius exact. Pohlhausens numerical integration is given in Table 5. He took as a good t to the exact Blasius solution over the range of Prandtl numbers 0.6 to 15 the expression:
1 1 2 Nu x = 0.3323 Pr 3 Re x

(36)

Asymptotic forms (Schlichting) are:


1 1 2 Nu x = 0.339 Pr 3 Re x , Pr , Nu x = 1 1 1 2 1 2 2 Pr 2 Re x = 0.7979 Pr 2 Re x , Pr 0 p

These have a break point, where the asymptotes intercept, at: p Pr 0.339 2 * = 0.00588 2 Exponential distribution. The temperature derivative at the wall in equation 35
1 1 2 is unity. For large Prandtl numbers we have Nu x = 0.6300 Pr 2 Re x . This is seen to be too high by about a factor 2. Around unit Prandtl numbers we have 1 1 6 1 1 2 2 Nu x = 1 - Pr 3 - 1 Re x , going to 0.5 Re x at Pr = 1. 2 5 3

International Journal of Mechanical Engineering Education 32/4

336

J. D. Lewins

Third-order Pohlhausen approximation. The temperature derivative at the wall is 3 . Taking the truncated form of the third-order polynomial to be valid over all 2 Prandtl numbers, a common form would be5:
1 1 2 Nu x = 0.323 Pr 3 Re x

(36)

Table 5 shows that the Pohlhausen third-order polynomial approximation is generally a good estimate of the Nusselt number as given by the exact Blasius numerical formula, and that the exponential approximation overestimates signicantly. The various asymptotic expressions are given and that based on large Prandtl numbers is of wider applicability outside its range than that based on small Prandtl numbers. General solution of the thermal boundary layer equation A more general problem has the wall temperature varying. The most common form of this is to have an unheated entry length at the free-stream temperature, followed by a uniform temperature along the remainder of the plate. (An alternative is to prescribe the heat ux along the plate.) The solution of the thermal boundary layer equation is elementary for the special case of geometric similarity with the (momentum) boundary layer on the at plate; the two boundary layer thicknesses have a constant ratio along the plate. The general solution of the thermal boundary layer equation, which would be required for problems lacking geometric similarity (e.g. an unheated leading portion followed by a constant-temperature portion) are more complicated. A solution for the case of an initially cold (at free-stream temperature) section followed by a constant plate temperature is sufcient; because of the linearity of the temperature in the energy equation, other cases may be solved by a linear combination of such a basic solution.6 Since the techniques to be used differ substantially between the Pohlhausen case and the exponential approximation, we treat them serially, the exponential approximation rst. Exponential transient solution We start with the equation: d d t2 a d + d = u d dx t t (37)

Writing as before dt/d = z and utilising the solution for the (momentum) boundary layer thickness, d = 2 x , we have:

Holman [6] is somewhat disingenuous in claiming that his equations 5.41 and 5.43 follow from his third-order Pohlhausen solution; the coefcient he quotes of 0.332 is Pohlhausens exact t. The thirdorder coefcient from Holman would be 0.323. And at his equation 5.36 the coefcient should be 1.0250, not 1.026. 6 We omit the adiabatic wall solution describing the effect of viscous self-heating.
International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

337

Fig. 6

Root locus diagram for the Z ratio with Prandtl number.

d dz 2 z 2 dd z 2 + 2z dz a = 2 z + 1 = z + 1 d x + d dx (z + 1) d x uzd

(38)

which has, as before, a particular integral, a solution of constant ratio Z, if this satises the polynomial equation Z3 = Z +1 2 Pr (39)

The equation has, of course, three roots, as shown in the root locus for Z(Pr) (Fig. 6). There is a real positive root for real positive Prandtl numbers, the only case of interest. The remaining two roots lie in the left-half plane and may be real or a 3 complex pair. These roots are duplicated for the break value Z = 3, Z2,3 = - for a 2 2 , for which, of course, we might expect the thermal boundary layer Prandtl value 27 to be outside the (momentum) boundary layer. If the problem at hand has, for example, an unheated leading strip followed by a constant elevated temperature in the plate, the thermal problem is not geometrically similar to the boundary layer problem and a constant ratio is not to be expected. Having found a particular integral, however, the equation can be reduced by the dt = Zz( x ), where Z satises equation 39.7 We obtain on utilising substitution z = d the expressions for d:

It is natural but not essential to employ the positive real root.


International Journal of Mechanical Engineering Education 32/4

338

J. D. Lewins

2 z 2 z + dz 1 dx Z = 2 x z + 1 z 3 - zZ + 1 Z Z +1

(40)

By inspection, the denominator on the right has roots z0 = -1/Z and z1 = 1. Dividing this second root into the third-order polynomial in z leaves the following quadratic equation for the remaining roots: z2 + z + so that Z -3 1 z2,3 = - 1 Z +1 2 (41) 1 =0 Z +1

Further inspection shows that these four roots are distinct with the exceptional cases when Z = 3 and z2 = z3 and in the limits of zero and innite Prandtl numbers. Thus, in general, the expression can be given in partial fractions obtained by the Cauchy residue theorem in the form: Z+2 1 1 dx 1 1 1 = + R2 +R + ( z - z 2 ) 3 ( z - z3 ) 2 x z + 1 2 Z + 1 (z - 1) Z (42)

where the nal residues are given by R2,3 = -

( Z 2 + 4 Z - 2) ( Z + 2) ( Z - 3) ( Z + 1) ( Z - 1)( Z - 3) ( Z 2 - 3Z - 3) ( Z - 3) ( Z + 1)

(43)

These are seen to be innite at the exceptional point. In general, then, the variables of the equation have been separated and it can be integrated. Applying an initial condition that the ratio z is zero at the end, x = x0, of the unheated initial plate length, we have:
Z +2 x z = ( Zz + 1)(1 - z) - 2 Z +3 1 - x0 z2 - R2

z 1 - z3

- R3

(44)

This is an implicit equation for the position where the ratio of boundary layers is some fraction z of the ultimate ratio Z. Of course, such an equation can be rearranged in a variety of ways. The special case of repeated roots at Z = 3 can be dealt with directly. We have: 1 1 5 1 13 1 1 dx 1 = d z + + + 2 1 6 1 9 z -1 9 1 2 x z+ z+ z + 2 2 3
International Journal of Mechanical Engineering Education 32/4

(45)

Thermal boundary layer equations

339

TABLE 6
z/Z = z x x0 0 1

Development of the ratio function with distance along the heated plate
0.2 1.013 0.3 1.037 0.4 1.078 0.5 1.143 0.6 1.244 0.7 1.407 0.8 1.703 0.9 2.532 0.95 3.512 1

0.1 1.002

Fig. 7 Partially heated plate and its boundary layers. Plate heated uniformly from x/x0 = 1, at Pr = 2/27 and Z = 3 exponential and third-order Pohlhausen approximated. (Note that a common boundary layer coefcient is used.)

Hence:
2z - (3z + 1) x 3( 2 z +1) = (46) 5 13 e x 0 (1 - z) 9 (2 z + 1) 9 This implicit expression for the ratio function is tabulated in Table 6 and plotted in 2 < 1 and a thermal boundary layer Fig. 7. For Z = 3, we have a Prandtl number 27 that tends to extend further than the momentum boundary layer. There is seen to be an immediate steep rise towards the asymptotic value, which is followed by a slower approach far down the plate. The theory is amenable therefore to exact solution in the general case and well illustrates the conventional thermal boundary layer results.

International Journal of Mechanical Engineering Education 32/4

340

J. D. Lewins

Pohlhausen third-order transient solution The general equation has the form: d 3 2 3 4 3 a d 20 z - 280 z = 2 Udz , z 1 dx and d 3 3 3 3 3 a d 8 z - 8 + 20z - 280z 3 = 2 Udz , z 1 dx (48) (47)

It can be noted that in the case of Pr > 1 but with a transient thermal boundary layer having a delayed start along the plate, it will rst grow according to the rst case, but once the boundary layers are equal will grow according to the second case. Hence the solution techniques are more complicated. Substituting the Pohlhausen values for the boundary layer gives: 1 5 13 3 2 dz 4 dz , z 1 z + 4 xz - z + 8 xz = d x 14 d x 14 Pr or z 3 + and 2 2 1 dz 4 6 13 + , z 1 z - z + -x 2z = 5 35z 2 dx 5z 35d 3 70 Pr (50) 4 dz 3 1 5 8 dz 5 13 - z + = 14 Pr 3 d x 14 5 dx (49)

(49)

These equations admit separation of variables, but we have found no simple analytical solution. Equation 42 is typically treated [6] by a further approximation, neglecting the terms of higher order, as in the analogous treatment of the particular integral giving Z, to write: z3 + 4 dz 3 13 = , z 1 3 d x 14 Pr (51)

so that we have a rst-order equation linear in z3 which can be solved by elementary means to yield:
1 3 3 dt 13 - 1 x 0 4 3 1 (52) z= =3 Pr x > x 0 , z << 1 d 14 x However, the resulting approximation is then used for Prandtl numbers of the order of unity, where we expect the ratio to be of the order of one, so that this approximation is not adequately justied. Furthermore, the approximation leads to a ratio at a Prandtl number of one that is incorrect, so that the expression is typically subject to a further inaccuracy. The nal result will commonly be used for gases with a

International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations

341

Prandtl number less than unity, using an asymptotic approximation for large Prandtl numbers. Comment Equation 52 indicates the general nature of things in the Pohlhausen approximation and in form is similar to the exponential approximation (Fig. 7). The many approximations made in coming to the simple form, however, make it dubious. The only feasible approach that avoids the sequence of dubious approximations would be to solve the problem numerically, unless we were limited to dealing with large Prandtl numbers. In nding a general solution, it must be noted that with small Prandtl numbers (less than unity), the transient solution has to be solved through both Pohlhausen ranges. The technique used successfully with the exponential approximation (substituting z = Zz) has not been found to be helpful, although correct in principle, because the resulting characteristic equations are higher than third order and do not readily yield explicit roots. Again, it is not obvious from the form of the equations that additional approximations are justied. These difculties are absent in the exponential distribution. Conclusion The exponential fall-off denition of a boundary layer is seen to have certain advantageous features: Like the exact solution for the underlying NavierStokes problem and the Prandtl boundary layer approximation solved exactly, but unlike Pohlhausen approximations, the boundary is innite in its extent. The integrations over an exponential in the integral approximation method are relatively easy to undertake in a teaching session. The e-folding boundary layer is found to be identical to the displacement boundary layer, which therefore gives further meaning to the e-folding. The corresponding thermal boundary layer is readily established for the usual case of geometric similarity and the thermal boundary layer is equal to its energy displacement thickness. The resulting expression for the ratio of the two boundary layers is continuous and gives the same asymptotic behaviour at large and small Prandtl numbers as the exact Blasius solution. The asymptotic values correspond to well known results in the literature. In contrast, the Pohlhausen approximation gives two ranges to be treated. The transient equation for the growth of thermal boundary layer relative to the (momentum) boundary layer is obtained exactly over the whole range of Prandtl numbers and is soluble analytically by elementary means. In contrast, the Pohlhausen approximation probably needs numerical solution expect for large Prandtl numbers. The ratio of thermal to momentum boundary layer thickness is readily obtained.
International Journal of Mechanical Engineering Education 32/4

342

J. D. Lewins

There are some disadvantages, of course: The exponential numerical factor for the boundary layer thickness at the conventional 99% recovery thickness is not accurate but is of the right order of magnitude. It is better given by Pohlhausen approximations. Of course, the exact (Blasius) value can be quoted and used in the subsequent thermal equations to improve the accuracy of the exponential approximation. The exponential approximation gives the correct functional dependence of the boundary layer displacement thickness on Reynolds number and Prandtl power coefcient. It does not, however, give the exact numerical coefcient. Although the geometrically similar heated plate problem is readily solved, the exact solution for (say) an unheated entry length to the plate is more complicated than the conventionally truncated third-order Pohlhausen approximation but nevertheless easier than the integration of the exact Blasius equation or indeed of the third-order system solved exactly. We think that the e-folding denition has some appreciable didactic advantages over the Pohlhausen approximation and the exact Blasius solution, when heat transfer is considered, and might therefore appeal to teachers of heat transfer. Of course, in place of either the Pohlhausen results or the exponential results, the nal form quoted is likely to be that of the exact t. In that case, ease of development at the blackboard should be a major factor and the exponential approximation has considerable advantages in this regard. Useful term project work might be to undertake numerical solutions of the transient equation approximations and compare the results, including predictions of the averaged Nusselt number over a partially heated plate. Acknowledgements I am grateful to Thiery Alboussiere for introducing me to the exponential distribution and discussing these results. Computations were undertaken with Mathcad 2.0. References
[1] [2] [3] [4] [5] [6] [7] L. Prandtl, Collected Works, Vol. II, (Springer, Berlin, 1961), pp. 575584. H. Z. Blasius, Math. Phys., 1 (1908), 56. E. Pohlhausen, Z. Angew, Math. Mech., 1 (1921), 115. J. D. Lewins, Beyond the boundary layers: the Blasius paradox, Int. J. Mech. Engr. Educ. (1998). H. Schlichting, Boundary-Layer Theory (trans. J. Kestin), 7th edn (McGraw-Hill, New York, 1975). J. P. Holman, Heat Transfer, 5th edn (McGraw-Hill, New York, 1981). E. R. Tuttle, S. Welch, Similarity solutions: a group theoretic approach for students, Int. J. Mech. Engr. Educ., 25(4) (1997), 243262. [8] A. Bejan, Heat Transfer, (Wiley, 1993).

Appendix We outline the difculty in using the Pohlhausen exact technique on his third-order polynomial approximation, starting from equation 26:
International Journal of Mechanical Engineering Education 32/4

Thermal boundary layer equations


y d

343

y q (0) q = d [g (0)]Pr Now we have g ( z ) =

[g(z)]
0

Pr

dz

(26)

3 3 [1 - z 2 ]; g (0) = 2 2

(A1)

Case (a). Thermal boundary layer thinner than the (momentum) boundary layer thickness: q (0) y q = z = 1 = d [g (0)]Pr
z

[g(z)]
0

Pr

dz

(A2)

dt < 1. Unfortunately, this involves two d unknowns, z and q(0). We have to make some assumption as to the functional 3 y 3 y 1 y behaviour of the temperature, for example q = - so that d 2 zd 2 zd 3 q (0) = , 2z to make progress. We would then have an equation for z as: where the integral terminates at z = 1= 3 2z
z

[1 - z ] d z
0

2 P

For example, at Pr = 2: z= 5 - 10 = 0.7827 and 3 q(0) @ 1.1740

in itself a reasonably accurate result. Case (b). Thermal boundary layer thicker than the (momentum) boundary layer thickness, Pr < 1, z > 1. Now the integral stops at the (momentum) boundary layer
1

and 1 = q (0) [1 - z 2 ] d z, allowing solution. For example, for Pr = 0.5,


0 1

Pr

1 = q (0) 1 - z 2 d z =
0

p q (0), and q (0) = 1.2732 4

(A3)

This is in itself a reasonably accurate answer but does not determine z or the form of the temperature distribution. For general Pr, a generalised binomial expansion with integration of a rapidly converging series is in principle possible but tedious. But one has to say this exact approach is unrewarding in the nite polynomial approximation in contrast with the exponential approximation, with its boundary layer stretching to innity.
International Journal of Mechanical Engineering Education 32/4

344

J. D. Lewins

A note on the numerical integrations To facilitate comparisons with the exact Blasius solutions, improved values of the Blasius function were taken from Schlichting and integrated by a Simpsons rule quadrature (exact therefore to second order) at twice the tabulated intervals of the similarity variable, f. The speed distribution was obtained to an agreement in the fth decimal place integrating over the tabulated f over some 44 points of the table at intervals of 0.2 reduced therefore to 22 points at intervals of 0.4. To obtain the boundary layer ratios and Nusselt numbers involved integrating
h

f (x)
0

Pr

dx numerically using the same scheme. The Pohlhausen (exact) results were

reproduced to within one part in the third decimal place. Whether this discrepancy is due to the numerical scheme or to Pohlhausens results using an earlier version of the Blasius exact results was not investigated. We did investigate extending the range of the exact results beyond Pr = 15 to Pr = 500 and found satisfactory agreement with the asymptotic form for large Prandtl numbers (as given in Schlichting). An extension to low Prandtl numbers showed agreement from Pr = 0.6 down to Pr = 0.05 but below that the scheme evidently failed, as can be seen in Fig. 8. Fortunately, this extension brings us to the range of validity of the low-Prandtl asymptote. Consideration was given to implementing an alternative exact scheme as given in Holman [6] (Appendix B, equation B14) but this involves a double integration of the stream function itself and correspondingly would have reduced the second quadrature to 10 points over an interval of 0.8 and so was not implemented since the whole range is adequately covered by the rst scheme.

Fig. 8

Range of validity of asymptotic ts to exact boundary layer Nusselt numbers.

International Journal of Mechanical Engineering Education 32/4

Vous aimerez peut-être aussi