Vous êtes sur la page 1sur 12

THE EQUATIONS OF MOTION (Chapter 2)

In this course, we will mostly deal with constant property (c-p) Newtonian fluids and will derive and discuss the Navier-Stokes (N-S) equations for this specific case. Continuum Assumption: We will deal with continuum flow case only, in which in a given point we have a statistically significant number of molecules that permits us to define properties such as density [(x,t)], pressure [p(x,t)], etc. Note: in standard atmospheric conditions in air, the average spacing between molecules is 3 nm, the average mean free path () is 60 nm, and the mean time between successive collisions is 0.1 ns. Knudsen number (Kn): Kn is the ratio of the mean free path to a typical length scale in the flow (Kn = / l ). If we take l to be 100 m (105 nm or 0.1 mm), then Kn is about 0.0006. For an l of 100 mm, Kn is about 0.0000006 (or 6x10-7). Kn < 0.01 0.01 < Kn < 0.1 0.1 < Kn < 3 Kn > 3 Continuum flow Slip flow Transition flow Free-molecule flow

Eulerian and Lagrangian Coordinates/Fields In defining flow properties such as (x,t), U(x,t), or p(x,t), we express them in Eulerian field with respect to an inertial (stationary or moving with a constant speed) coordinate system x. Now if we follow a fluid particle, then we define flow properties such as (t), U(t), or p(t). Here we express them in Lagrangian field or material field with respect to time only. While Lagrangian coordinate system is often used in solid mechanics & dynamics, we use Eulerian coordinate system in fluid dynamics, where it is not feasible to follow tremendously large number of fluid particles. We often need to relate the rate of change of properties in these two coordinate systems. Material derivative, substantial derivative, or total derivative provides such a relation. D()/Dt = ()/t + U.() = ()/t + Ui()/xi where the term on the left is substantial derivative of the Eulerian field; the first term on the right is partial derivative of the Lagrangian field (temporal term) and the 1

second term on the right is the rate of change due to convection of particles (convection term). Conservation of Mass or the Continuity Equation You have seen the derivation of this equation many times before. In the general case: /t + .(U) = 0 In terms of specific volume [(x,t) = 1/(x,t)]: D(ln)/Dt = .U The term .U is called dilatation term. For constant density (incompressible flow: either Mach number less than 0.3 or variations in density less than about 5%): .U = 0 or Ui/xi = 0 The flow is called divergence free or solenoidal.

Momentum Equation The momentum equation is derived using Newtons 2nd law: particle acceleration = surface forces + body forces. Define a gravitational potential force in such a way that g = -. Then for a constant gravitational field ( = gz), the general momentum equation becomes: DUj/Dt = ij/xi - /xj For specific cases of flows with constant property and Newtonian fluid, stress tensor is given by: ij = -Pij + (Ui/xj + Uj/xi) where P is the pressure, is the coefficient of dynamic viscosity, and ij is the Kronecker delta (=1, for i=j, = 0 otherwise). Since we are dealing with a solenoidal field (Ui/xi = 0), the first term on rhs is isotropic normal stress, the second term is non-isotropic shear stress. Substituting ij in the momentum equation and using the continuity equation of .U = 0, we get the Navier-Stokes equations for constant property Newtonian flows: 2

DUj/Dt = 2Uj/xixi - P/xj - /xj Further simplification is possible if either we can neglect the effect of gravity or lump the pressure and the gravitation term (p = P + ): DU/Dt = -(1/)p + 2U This momentum equation together with the conservation of mass, along with appropriate boundary and initial conditions are used to determine properties of a constant property (cp), Newtonian flow. For a stationary solid wall for example, the impermeability and the no-slip conditions yield U=0. If either viscous effects are negligible or we assume ideal inviscid flow, then the shear stresses can be ignored, which makes the stress tensor an isotropic tensor (ij = -Pij). The simplified momentum equation is then called the Euler equation: DU/Dt = (1/)p For a stationary solid wall then, impermeability (n.U=0) is a valid boundary condition at the surface, but NOT U=0 condition. Note that one can use = 0 in the Navier Stokes equations or use (ij = -Pij) to derive the Euler equation. However, the solution for the Euler equations and the Navier Stokes equations with approaching zero, will give different results because of the differences in the boundary conditions. In fact, = 0 is a singularity point for the N-S equations. The role of pressure in c-p Newtonian flows Earlier, we lumped the gravitational force [which is a conserved (no source or sink) body force] with the pressure and defined p = P + , where g = . This means: The effect of body force and the isotropic stresses (ij = -Pij) is the same Also, since the N-S equations derived for a c-p flow and p appears as p, therefore the body force does not have any effect on the velocity or the modified pressure field (it adds a constant). This would NOT be the case if we let the density, which is a property, to vary (e.g. in a buoyancy driven flow or a compressible flow). As a result of what was said in the above paragraph, we can call p pressure. Note that since we have assumed constant properties, therefore, there is NO connection between this pressure and density or temperature (hence the equation of state does not hold). Now if we take the divergence of [.()] of the N-S equation, without assuming the velocity field to be solenoidal (.U = 0), the result is: 3

[DU/Dt = -(1/)p + 2U] (D/Dt - 2)(U) = (1/)2p + (Ui/xj)(Uj/xi) If and only if the rhs is equal to zero everywhere, then this c-p flow is solenoidal. This means that: 2p = - (Ui/xj)(Uj/xi) This is the well-known Poisson equation. Therefore, we conclude that the satisfaction of the Poisson equation is a necessary and sufficient condition for a solenoidal velocity field to remain solenoidal. Another word, Poisson equation must hold in a solenoidal field (or incompressible flow) The Poisson equation can be solved (it is a bit involved). The final answer is: p(x,y) = p(h)(x,y) + (/4)(Ui/xj)(Uj/xi)dy/|x-y| Without going into details, it is sufficient to say that the first term on the rhs depends on the boundary conditions, and therefore, the solution to the Poisson equation gives the pressure filed at a given point in a given time, p(x,t), in terms of an integral over the entire fluid volume of the velocity field. This means that the pressure is a global property and its value in a given point and time depends on the entire velocity field at that time. This creates a major problem to turbulent modelers, who try to relate the pressure to velocity (correlation of pressure and velocity appears in Reynolds averaged equations).

The conserved passive scalar equation Conserved means that there is no source or sink. For example, if we add a trace species for measurement, visualization, etc. purposes. Passive means it does not have measurable effect on material properties (e.g. , , ) For c-p flows, the equation is: D/Dt = 2 Where is the diffusivity coefficient (which is a fluid property and is constant, as we are dealing with constant property cases) and is a conserved, passive scalar. If the scalar is the concentration of a trace species, then is the molecular diffusivity. The Schmidt number is defined as: 4

Sc = / (the ratio of viscous diffusion to molecular diffusion) If the scalar is a small excess temperature, then is the thermal diffusivity. The Prandtl number is defined as: Pr = / (the ratio of viscous diffusion to thermal diffusion) Note: Pr ~ 1 (0.72) for air (momentum & thermal diffuse at abut the same rate). Pr ~ 6 for water (momentum diffusion is much faster than thermal diffusion) there is potential for large temperature gradient in regions with small velocity gradient. Pr << 1 (0.024 for mercury) for liquid metals (momentum diffusion is much slower than thermal diffusion) - there is potential for large velocity gradient in regions with small temperature gradient. A note of caution: in turbulent flows, fluid properties may not play a major role in diffusion, as most of the flows are dominated by large scale turbulence structures that dominate the diffusion process.

The Vorticity Equation Turbulent flows are inherently rotational (which means that fluid elements spin around their own center of mass) with 3-d fluctuating velocity and vorticity. Take a point in the flow, pass two mutually perpendicular lines through the point, calculate the average rotation rate of the two lines around the axis passing through the point and is perpendicular to the plane of the two lines. Twice this average rotation rate is called vorticity at that location. =xU where 1 = U3/x2-U2/x3 2 = U1/x3-U3/x1 3 = U2/x1-U1/x2 In turbulent flows, which are rotational, = x U 0. If = x U = 0, then flow is called irrotational. To obtain the equation for the evolution of the vorticity, we can take the curl of N-S equations: x [DU/Dt = -(1/)p + 2U] 5

D/Dt = 2 + U The first thing to notice is that the pressure term has dropped out: - x (1/)p = -(1/) xp = 0 This is because of the constant density assumption and also because the curl of gradient of any scalar is zero. If we did not assume constant density, then: - x (1/)p = -(1/) xp -(1/)xp = -(1/)xp The term on the rhs is called the baroclinic torque, and could be a major vorticity source, if the density and pressure gradients are misaligned. Now getting back o the vorticity equation: D/Dt is the rate of change of vorticity 2 is the diffusion of vorticity by viscosity U is the stretching and tilting of vorticity, which is a very important term Now we will expand and look at one component of the vorticity equation to better understand the stretching and tilting term: D1/Dt = 21 + U1 Expanding the vortex stretching and tilting term: U1= (1i+2j+3k)(iU1/x+jU1/y+kU1/z) = 1U1/x+2U1/y+3U1/z 1U1/x is the stretching of 1 itself by the strain rate (U1/x) in the x direction (from the conservation of angular momentum, the stretching, thus decreasing the radius of vortex line, will increase vorticity, and compressing the vortex line will do just the opposite). 2U1/y is the tilting of 2 due to gradient of U1 in y direction 3U1/z is he tilting of 3 due to gradient of U1 in z direction

Now if we assume we have a two-dimensional flow, say in the x-y plane. This means, we have 1= 2= U3 = ()/z = 0. As a result, U1 = 0 in the above equation. One can similarly expand U2 and show that it is zero. By definition U3 = 0. Therefore, U = 0 in the general vorticity equation and the vorticity equation becomes: D3/Dt = 23

which means that there is no vorticity source to sustain turbulence. Therefore, the nonzero component of the vorticity, 3, will evolve as a conserved scalar. This means that there is no true 2-d turbulent flow. Only in the atmospheric turbulent flows, one could get an approximately 2-d turbulent flow, which is quite different than typical turbulent flows.

Helmholtz Theorem: For an infinitesimal line element of material, s(t), one can derive the equation of the evolution (see the text book): ds/dt = sU This is identical to the vorticity equation in the absence of the viscous term (D/Dt = U). Therefore, in an inviscid flow, the vorticity vector behaves in the same way as an infinitesimal line element of material. This is called Helmholtz theorem. If the strain rate generated by the velocity gradients aligned with the element, then it will either stretch or compress the element. Similarly, the strain rate generated by the velocity gradients aligned with the vorticity vector will either stretch or compress the vorticity. Circulation The circulation is defined as the line integral around a closed curve C of the velocity component tangent to the curve, times the arc length around the curve: = C Uds where s is the unit vector tangent to the curve C. Using the Stokes theorem to relate the integral over a closed curve C to the integral over the surface A bounded by the curve: C Uds = A (xU)dA where dA = ndA (n being normal unit vector to the surface. The circulation can be defined as: = C Uds = A (xU)dA = A dA which shows that circulation is the surface integral of vorticity. Kelvins theorem states that for an inviscid, incompressible flow, the circulation is time invariant for any closed path: d/dt = 0. To get this, you need to differentiate = C Uds, then substitute for dU/dt from the momentum equation. 7

Using equation = A (xU)dA = A dA, one can infer that the integral of angular velocity or the integral of vorticity over an area bounded by a closed curve is also time invariant. (note the notation change on the figure from U to V)

Rates of strain and rotation One can decompose the velocity gradients Ui/xj, which is a second-order tensor, into three components (isotropic, symmetric, and antisymmetric): Ui/xj = 1/3ij + Sij + ij where = U is the dilatation rate, which is zero for incompressible flow, ij is the Kronecker delta, Sij is the symmetric rate of strain tensor: Sij = (Ui/xj + Uj/xi) [(1/3)ij] where the last term is for compressible/variable density flow, and ij is the antisymmetric rate of rotation tensor: ij = (Ui/xj - Uj/xi) Recall that we discussed the Newtonian stress law as

ij = -Pij + (Ui/xj + Uj/xi) This equation can now be written as: ij = -Pij + 2Sij which shows that the viscous stress depends linearly on the rate of strain and, as one would anticipate, is independent of the rate of rotation. As we had discussed earlier, the vorticity in a point in a flow can be obtained from the average rotation rates of two mutually perpendicular lines through the point, which is perpendicular to the plane of the two lines and twice this average rotation rate (i = ijkUk/xj). It can be written as: i = -ijkjk ij = -ijkk where ijk is the alternating symbol, where ijk = 1 if 1, 2, 3, (or x, y, z) are in this circular order, ijk = -1 if the circular order is reversed (3, 2, 1), and ijk = 0, for all the other cases. Note that while i is a vector, ij is a tensor. Transformation or invariance properties Here we will look at the effects of various transformations of the coordinates on the Navier-Stokes equations. What we want to find out is that how general these equations are and what kind of scaling do we need to use in order to get similar flow fields using different facilities, different scale models, different fluids, etc. This exercise will provide us some interesting information on Reynolds number similarity, invariance under fixed rotations and reflections of the coordinates, Galilean invariance, and the lack of invariance under the coordinates rotation. Lets define to be the length scale and to be the velocity scale, then two independent nondimensional space and time variables are: x = x/, and t = t/ Note that for nondimensional variables I use an underscore and the book uses a hat.

A Accordingly , the two dep pendent nondimensional l velocity an nd pressure variables v are: U (x,t t) = U(x,t)/ and p(x,t)/ /(2) Now if we su N ubstitute thes se nondimen nsional variab bles into the continuity, the NavierStokes, and th he Poisson equations, e we e get: Ui/xi = 0 Uj/t + Uj Uj/xi = (1/Re)2Uj/xixi - p/xj 2p/xixi = (Ui/xj)(Uj/xi) w where the Rey ynolds numb ber is define ed as: Re = / . When you compare these W e nondimensi ional equatio ons to the di imensional equations e discussed ear rlier, they are e identical, except e for the appearance of the Rey ynolds number. R Reynolds num mber similar rity Now we run an N a experime ent that has different d leng gth (b) and velocity v (b) scales (and d thus different time e scale, (b/b)), and also o different flu uid propertie es (b, b, etc c.) than the one o ove. Now if we w normaliz ze all the dep pendent and independent t variable in the discussed abo ame fashion as before, th he boundary y conditions and a the N-S equations would w be the sa sa ame, except the Reynold ds number. The T Re = / in the first t case and Reb = bb/b. If w make Reb = Re, then when we w we me easure or calculate the sc caled velocit ty and pressu ure, th hey would be e the same in n the two ex xperiment. Th his is called Reynolds nu umber simila arity. W use the Reynolds We R num mber similarity regularly y in modeling g and in exp periments dea aling w fluid flow with ws. 10

Note that the Reynolds number appeared in the viscous term, when we nondimensionalized the equations of motion. Since the Euler equations do not have the viscous term, therefore, the normalized velocity and pressure field are the same, regardless of the scale. Therefore, the Euler equations are invariant with respect to scale transformation. Time and space invariance (c) If you shift the independent variables t and x by T and X, the equations of motion would not be affected and the velocity and pressure results in the two cases would be the same. Therefore, the N-S equations are time and space in variant. Whether you run the same experiment today or tomorrow, and whether you run in Ohio or California, the results better be the same. Rotational and reflection invariance (d & e) You can show that Navier-Stokes equations and the experiments that one run are rotational and reflection invariant. Galilean invariance (f) The coordinate systems that we typically use are inertial (with no accelerations). Now if the second coordinate system moves with a constant velocity V with respect to the first one. The transformation in x, t, and U would be: Xt = x Vt; tt = t; Ut(xt,tt) = U(x,t) - V

If we first normalize these variables, then substitute them into the N-S equations, you will not see any changes in the equations. This means that the N-S equations are Galilean invariant. If you run an experiment on a truck moving with a constant speed on a smooth road or run the same experiment in a laboratory, you would get the same results.

Extended Galilean Invariance (g) A very interesting property of N-S equations for incompressible flows is that the equations are invariant with respect to rectilinear acceleration of the coordinate system. The transformations in this case are similar to the one above, except that V=V(t). Xt = x Vt; tt = t; Ut(xt,tt) = U(x,t) - V

If we substitute these variables into N-S equations (but use underbars instead of subscript t for the transformed variables; (note that underbars do not mean normalized here) , we get:

11

Uj/t + Uj Uj/xi = 2Uj/xixi (1/)p/xj Aj where A = dV/dt. The last two terms on the rhs can be combined and written (note the acceleration term is absorbed into the modified pressure term) (1/)p/xj + Aj = (1/)(p+xiAi)/xj (Note that double I indices in xiAi imply summation). Now if we normalize the transformed velocity (Ut(xt,tt)) with and the modified pressure (p+xtA) with 2, and substitute them into the N-S equations, we get the same equations as those in the inertial frame. This is called extended Galilean invariance. Note that this is true only in incompressible flow, as in compressible flow we cannot absorb the acceleration term into the pressure term. This means that if you run an experiment on a truck moving with a variable speed on a smooth road or run the same experiment in a laboratory, you would get the same results.

Rotating frame (h) If we choose a non-inertial rotating frame with a rotation rate of , and follow the procedure that we carried out under the extended Galilean invariance section, we get Fj instead of -Aj for the forces on a fluid element: Fj = -xiikkj 2Ui ij - xidij /dt where x, U, and t are all transformed parameters (should have overbars on them), and the three terms are due to: the centrifugal force, the Coriolis force, and the angular acceleration force. The last term would be zero if the was not time dependent. However, the Coriolis term always plays a major role in rotating flows such as atmospheric and turbomachinery flows. While the centrifugal force term can be absorbed into a modified pressure term, the other two cannot. Therefore, the N-S equations will be significantly different in rotating and non-rotating frame. Therefore, they do not have material-frame independence properties. In the rotating frame, the equations for the evolution of vorticity, i = ijkUk/xj, become: i/t+Uii/xi=2i/xjxj+jUi/xj-2ijk(Un/xjnk)-ijkdjk/dt Again, the last two terms are due to the Coriolis and the angular acceleration forces, making these equations in a rotating frame significantly different than those in an inertial frame.

12

Vous aimerez peut-être aussi