Showing posts with label water hammer. Show all posts
Showing posts with label water hammer. Show all posts

Thursday, October 26, 2023

Governing equation for transient pipe flow, part 4

From the last part we ended up with these two equations:


These equations are on a form that is suitable for FSI analysis when combining them with the equations of motion for piping systems (longitudinal, rotational and bending vibrations). They could also be used as is for pure hydraulic transients, but the normal starting point is to use pressure head, H, instead of the pressure, P, and velocity, V instead of flow rate Q (even though Q usually is eventually used in the end) and the diameter, D, instead of the radius, R.

Neglecting the Poisson coupling (sturdy piping structure), the last term in equation 3.19 and reorganizing we get:

Here the wave propagation velocity, a, becomes from the outset (for thin walled pipes):

For most cases this is a good approximation, but Wylie & Streeter have elaborated this for lots of different conditions in their books. Based on experience and practicalities however, the wave propagation velocity is a very fuzzy thing to determine. It can hardly be experimentally verified for any real piping system within +- 20%, mostly due to small, and changing, amounts of gas content, gas pockets and so on. Wylie & Streeter writes a lot about that too, and in the MOC (Method of Characteristics) it is used as a parameter that is changed within limits to discretize large systems. Even though it can be calculated theoretically at very high accuracy, that accuracy simply does not exist in real systems. Having a good starting point, is a good idea nonetheless though, because the difference between polymer based pipes and steel pipes can be more than 100% for instance.

What's missing now is the pipe friction. Diving into this, one discover it is super complex material. To this day, no good general transient friction term exists as far as I know. I have written a lot about it in my thesis. For steady state, Darcy-Weisbach and the Moody diagram is what is used. In lack of a better general non steady state model, Darcy-Weisback is also used for transient simulations, even though "everybody" knows this is far from correct. The best we can say about this, is that at least the calculations becomes conservative. Calculating water hammers that are smaller than in real life, does not normally happen. The exceptions are when FSI becomes important and for fluid cavity collapse. If this is not taken into account, then the calculations are usually not conservative anymore.

The Darcy-Weisbach equation:
Here f is the Darcy-Weisbach friction factor we normally obtain from the Moody diagram. The minus sign is there because of integration when positive x is in positive flow direction. This is simply the steady state result of the momentum equation above, and we can "reverse" it back into that equation to achieve exactly that. The final equations for liquid filled transient pipe flow analysis therefore becomes:
The wave propagation velocity, a, is as shown above.




Wednesday, October 25, 2023

Governing equation for transient pipe flow, part 3

 To get further from part 2, Hooke's law in cylindrical coordinates is introduced.


Then by combining Equation 3.13, 3.15 and 3.16 we get:

We have already discussed that the radial inertia term is unimportant (Long wavelength approximation). We can therefore substitute the tangential stress term for the quasi-stationary relation for a thin walled cylinder found in most text books, and assuming the pipe wall is much thinner than the pipe diameter.
A bit reorganizing, and Equation 3.17 becomes:

This is now fully 1D, and the stress term can be substituted using Hooke's law in 1D:

We now get the final fluid equations:
 

This is the same equations for transient pipe flow as can be found in any text book (when re-organized a bit), except the last term in Equation 3.19. This is the so called Poisson coupling of the fluid to axial motion of the pipe. The fluid wave propagation velocity is approximated by parenthesis before the second term (when multiplied by the fluid density), and is also essentially the same as found in most text books.

An interesting point is that when deducing the Water-hammer equations from Navier-Stokes equations, the FSI coupling pops out automatically, even for fully 1D equations. The only way to remove that coupling, is to force the axial movement of the pipe to be small compared with the pressure forces. We simply say that axial displacement of the pipe wall, u, is "small". In other words, it is assumed that the piping system is in practice 100% rigid. The other way is to set the Poisson ratio to be zero, which it isn't. The Poisson ratio of steel is a constant, and about 0.3.

For normal engineering purposes we simply say that the piping system is "rigid enough" for the transient fluid simulations to be correct. This is usually the case for the vast majority of systems. The only way the piping system is normally included, is in the calculation of the pressure wave velocity, see for instance Equation 3.18 and 3.19 where it can change the wave propagation velocity substantially. This will be correct also for thick walled pipes, except in cases where the pipe wall material becomes very "soft".

For the pipe itself, the axial motion ends up being described by:

The pressure is coupled to axial displacement by the Poisson ratio. When we say that the axial displacements are small compared with pressure forces, we can flip it around and say that the pressure forces are large compared with the axial displacements. Thus, vibrational analysis of the mechanical piping system without including the fluid transients, is therefore fundamentally wrong from the start. We see that the pressure forces are proportional to R/e, the radius of the pipe over the wall thickness. By increasing the wall thickness we will also reduce FSI effects, which fits well with intuition in this case.

For a full FSI analysis, junction coupling, bending and torsion also must be included, as well as all the clamping structures. It is a bit funny however, that transient simulations of the fluid inside the pipe can be done in most cases by disregarding the piping structure elasticity, but doing a vibrational analysis of the piping structure cannot normally be done correctly without including the fluid transients.

Monday, October 23, 2023

Governing equation for transient pipe flow, part 2

In the previous part, the so called extended water hammer equations was discussed (FSI), and a reduction of the Navier-Stokes equation was started.


Imagine a valve being slam shut at the left end. Water streaming from the right. The valve causes a pressure wave, a shoch wave (water hammer) propagating to the right at about 1200 m/s. The widening of the pipe wall causes axial stress, which will shrink the diameter through the poisson coupling. This causes the precursor wave propagating at about 5000 m/s.

The equations were 2D. We want them 1D. This is done by averaging the pressure and flow across the cross section. Mathematically this is done by multiplying with 2𝜋R, integrating with respect to r from 0 to R and dividing by 𝜋R^2.

V and P is the average velocity and pressure, and expressed as:


This is just a fancy way of saying V and P are the same all over the cross section of the pipe. This is of course not 100% correct physically, particularly not for the velocity V, but this is the simplification we must do to make them 1D.

But we had one more equation from the last part:

The assumption is that the characteristic fluid length scale, L is large compared to the characteristic diameter, D. That is L/D >> 1 under the constrain that the wavelength is "long". With this assumption, it can be shown that the radial inertia is negligible compared to the axial inertia. We can simply neglect the first term in equation 3.7. This is called "the long wavelength approximation", and is found to be correct for frequencies up to 63% of the first radial liquid filled pipe frequency (which normally is much higher than the first axial frequency). Proof of this is found in my thesis. Then we are left with the last term in equation 3.7 to be equal to zero, which simply states that the pressure is constant in the cross section. Thus "the long wavelength approximation" will also "one dimensionalize" the pressure, and remove one equation from the set.

On the pipe wall, the radial fluid velocity must be equal to the radial pipe velocity:

Inserting this, and we are left with only these two equations:

We have removed one equation. We have transformed them from 2D to 1D, but we have added a radial wall velocity term. This looks like a complication, but is a key point for the end result in part 3.

Sunday, October 22, 2023

Governing equation for transient pipe flow, part 1

There are several ways to deduce the governing equations for transient pipe flow. These equations are the heart of any transient flow simulations software. The final equations are often called the Allievi equations, or the water-hammer equations.

This will be a short series of how I did it in my PhD thesis, directly from the Navier-Stokes equations. Going from the NS equations, instead of any other method, we can see all the simplifications that is done to achieve the wanted result. These equations are the so called extended water-hammer equations. They also take into account the elasticity of the pipe itself, which causes the not so well known precursor wave. 

When a valve is shut very fast, this will cause a water-hammer wave in the fluid and a similar "steel hammer" wave in pipe wall, an axial shock wave due to the elasticity of the wall.

When a valve is shut very fast, the velocity of the fluid close to the valve is changed from the steady state velocity to zero. If this change of velocity is fast enough, as it often is, this causes a chock wave in the fluid, starting at the valve, propagating upstream with a velocity of about 1200 m/s, the approximate speed of sound in a water-filled regular steel pipe. This is the normal water-hammer. However, the sudden pressure rise at the valve, also causes the pipe to expand radially. This radial expansion of the pipe wall will in turn cause an axial tension in the pipe through the poisson coupling.

The poisson coupling of a homogeneous solid material is the effect that if it's stretched, it will also become thinner in the perpendicular direction. Stretching a rubber band, causes it to become thinner, proportional to the poisson number, or poisson ratio of that material.

This axial tension will also propagate as a shock wave, but much faster than the water hammer wave. The coupling goes both ways however. Axial tension in the pipe wall causes the pipe to decrease in diameter, like a rubber band. This decrease, propagating roughly at the speed of sound in the pipe wall, about 5000 m/s in steel, will in turn cause a smaller pressure wave, also propagating at 5000 m/s. This is the so called precursor wave, that actually can be measured to foresee the coming water hammer wave some small time step later. This could be a millisecond, or several seconds, depending on the length of the pipe.

This precursor wave and the coupling is usually of little practical interest when dealing with most fluid systems (the fluid transients), but can be of outmost importance when dealing with the pipe structure and particularly vibrations in this structure. This is called fluid structure interaction, FSI, and is the cause of most failures in piping systems. FSI in general is a wide area encompassing many different phenomena.

OK, back to the equations, but it is important to keep in mind that every single transient fluid pipe phenomenon has some level of FSI in it, it's impossible to look at fluid transients without also keeping the piping system in mind.

The starting point is the 2D Navier-Stokes equations in cylindrical coordinates.


by observing this fluid element in axis-cylindrical coordinates:
No simplifications are made except 2D flow, symmetry around the x axis and isothermal flow.

The next simplification is to assume very low compressibility. A typical gas has high compressibility, while a liquid has very low compressibility. Many people don't even consider liquids to be compressible, but the very fact that it has some compressibility is the crucial point in all transient piping flow analysis. For very low compressible fluids, the variation of density with respect to pressure can be approximated by a constant value, that is:
K is the bulk modulus, the elasticity of the liquid and c is the wave propagation velocity. 𝜌 is the density of the fluid, and p is the pressure.

For the sake of clarity, all viscous terms are neglected at this point. Friction will be added later in the form of Moody friction. By assuming very low Mach numbers, M << 1, all the convective terms are neglected. For clarity we also neglect all external body forces. These can also easily be added later. For more elaborate proof, especially this M << 1, one can look in my thesis. We then end up with a set of much simpler equations.
These equations are still not easily solvable, particularly due to the variation in radial direction, they are still 2D. This will be handled in the next post.