Making Correct Parabolic Nozzles

by Richard Garcia, Director of Research, RRS.ORG


I would like to set the record straight on a common mistake I see with designing parabolic rocket nozzles.

The simplest and most common nozzle used in amateur rocketry is the simple cone. Most commonly the cone is a 15° half angle. But the simplicity of the cone comes at the cost of reduced performance and increased length. The next step up in complexity is a parabolic bell nozzle.

The parabolic bell nozzle design is a little more complicated and much more difficult to manufacture. Machining a parabolic shape out of metal on a lathe by hand is doable with a digital read-out (DRO) device but the process would likely be very tedious.  In this case, I would recommend looking into some sort of computer-controlled system like CNC or 3D printing. If you’re using an ablative composite like silica-phenolic then tuning a wood mandrel on a lathe is very doable. Beyond the parabolic nozzle the next step up in complexity and efficiency is using the method of characteristics to design a rocket nozzle. This method was pioneered by Gadicharla V.R. Rao. 

https://en.wikipedia.org/wiki/G._V._R._Rao

.These are sometimes called “Rao” nozzles. The parabolic nozzle is a decent approximation of a Rao nozzle, a method also proposed by Rao. Going further to get the last drop of performance would require a detailed computational fluid dynamics (CFD) based optimization.

I’m writing this report to set the record straight on a common mistake I see with determining the proper geometric and dimensional features of specific parabolic rocket nozzles. Almost all the amateur parabolic nozzles I’ve seen up to this point, including several college teams, and even my own early designs, made this mistake. It’s not really their fault though. The parabolic nozzle is commonly mention in rocket literature but rarely is one particular key fact needed to make them correctly explained. The two of the most prolific textbooks on the topic are, “Rocket Propulsion Elements” by Sutton and Biblarz, and “Modern Engineering for Design of Liquid-Propellant Rocket Engines” by Huzel and Huang. Both give their readers charts to help design the parabola and both also neglect to mention this one important fact.

That there is a problem at all will only become apparent when you try to solve for the exact equation of the parabola. If you read through the common literature (a task left to the reader) and go through the process you will see that the parabola is defined by two points and two angles. The two points are the beginning and the ends of the parabola. The beginning point is where the curvature at the throat, typically a circular arc, transitions into the parabola. The end point is the exit of the nozzle. The two angles, θn at the beginning of the parabola and θe at the exit are the angles between the centerline of the nozzle and lines tangent to the parabola at the two points. These angles are normally found by using graphs given in the literature. These graphs are the angles vs expansion area ratio as shown below. Note that they present the length of the bell nozzle as a percentage, Lf, of the length of an equivalent conical nozzle with the same nozzle expansion area ratio and a 15° half angle.

Graph of theta angles from Rao’s paper

Footnote about the θn and θe vs expansion area ratio graphs.

I’d like to note that the graph I shared is the in the commonly presented form. However, if you look at the original paper on parabolic nozzles “Approximation of Optimum Thrust Nozzle Contour” by G.V.R. Rao in 1960” you will find the information presented differently. Instead of theta versus epsilon at various percentages of nozzle lengths given in terms of the percentage of an equivalent 15° conical nozzle length of the same expansion ratio (epsilon), the axes of graph are in terms of the ratios re/rt versus Ln/rt given at various values of θn and θe. These lines are straight in comparison to the curved lines in the graph above. The more common graphs are supposed to have been derived from the originals. 

However, I have not gone through the math myself, but, I am interested in what the process and results would be. In turn, the originals are supposed to have been derived from full calculations of the Rao method. Again, something I have not tried replicating but it is something I would be interested in seeing the details of. 

Another thing worth mentioning is that I recall reading online that someone had tried to reproduce these results using the method of characteristics. (Unfortunately, I cannot recall exactly where I read this.) The original graphs were based on using a fixed specific heat ratio (gamma) of γ=1.23 and the original paper stated “A different value of γ does not appear to appreciably change the nozzle contour when the area ratio and length are prescribed.” however the testimony I found online stated that they found that the gamma value did significantly affect the angles. I am in doubt over this anonymous internet testimony, but it is something I’m interested in exploring in more detail.  If you ever investigate this yourself, dear reader, be sure to send us what you discover.


In total, you will have six parameters to determine, at point 1 you will have x1, y1, and θn. At point 2, you will have x2, y2, and θe. The obvious route from here would be to take the equation of a parabola (more precisely a square root since the x-axis will be our chamber centerline. But hey, a square root is just a parabola on its side), and these six parameters and try to solve for the unknowns in the equation.

This is where you run into the problem. The parabola is over-defined! You can’t solve for a parabola that meets all six criteria. It is impossible. Five of the defining parameters can be correct, but at least one will always be wrong. This is where most people, after considerable effort and eventually the banging of one’s head against a wall, will give up and let one, or more, of the 6 parameters to be wrong, usually by letting the starting angle of the parabola to be wrong.

When plotting a nozzle contour most put the centerline of the nozzle along the x-axis. The mistake everyone makes, including myself, is to make the natural assumption that this parabola is aligned with the nozzle’s same x and y axis in the same way as all of the parabolas in your algebra classes. That is to say the centerline of the parabola is around the x axis, the same axis of axial symmetry as the nozzle. But what everyone misses, including in the common literature, is that the parabola should be canted! That is to say, it should be rotated by some angle in relation to the x and y axis.

The way I found this fact out is by being a prolific reader. I’ve spent a lot of time gathering papers and books on rockets. I’ve only ever found one source that explicitly states the nature of these parabolasand that is in “Liquid Rocket Engine Nozzles NASA SP-8120” It simply refers the them as a “canted-parabola contour” and offers no further explanation. But that clue was enough for me to figure out what was going on. Adding the rotation of parabola adds a degree of freedom in the math. This allows for the equations to be solved with all the required constraints.


* Footnote on the NASA SP-8000 publication.

By the way, the NASA SP-8000 series of publications is a fantastic resource for the aspiring rocket scientist. Back in the 1970’s, they tried to write a comprehensive review of the state of the art in rockets and spaceflight. I remember reading somewhere that they fell short of their goal, since there was too much know-how to catalog. But what they did manage to publish is great. Whenever diving into an unfamiliar design topic, I’ll read the relevant SP-8000 article on it first as a good overview.  Some topics may be more outdated than others, but much of it is still great and relevant today.


When trying figure this out, I went digging and found the original paper on parabolic nozzles. “Approximation of Optimum Thrust Nozzle Contour, by G.V.R. Rao, 1960” I had hoped it would directly state the parabola should be canted, but it doesn’t. However, it does give enough information to deduce this conclusion.  In the short paper, its one page, Rao describes “a simple geometric construction for the parabolic contour.” It is something that would have been more familiar with the designers and draftsmen of that time. Today, we are more likely use a series of points and connect them with a spline curve, or use mathematical equations in CAD software to draw similar curves. But back then, designers would have to do it by hand on paper. The method suggested by Rao is called the “tangent method” of drawing a parabola. As the name suggests it involves drawing a series of lines that are tangent to the parabola.

Since we are interested in the mathematical equations, I won’t be describing that method any further.  But it is plain to see from figures that the method does not directly depend upon the x, and y axis. The contour is simply plotted on the x-y plane and then the contour is revolved around to x-axis to create the three-dimensional nozzle. Since we cannot fit the desired parabola with the normal parabolic equations, we are left with the question: 

How does one mathematically rotate the equation of a parabola?

Using parametric equations is how we will be able to create aparabola rotated on the x-y plane. Parametric equations also make some of the math easier when solving for circular arcs that are used in the rest of the nozzle contour. 

First let’s start with vertex form a parabola.

Equation 1; vertex form of the parabola equation

To parameterize this, let us set:

Equation 2: parameterization or variable substitution

And solve for x and y.

Equations 3.1 and 3.2: Solving for x and y variables

This pair of equations is now a parametric form of a parabola. But it is just a regular parabola and we need to rotate it. Recall from your study of trigonometry, that the formulas for rotating a point around the origin by angle θ are:

Equations 4.1 and 4.2: Rotating the frame of reference by angle, theta

We can combine equations 3 and 4 by substituting the parametric equations into the place of x0 and y0. Doing this, we get:

Equations 5.1 amd 5.2: Combining prior equations 3 and 4

From this, we can get:

Equations 6.1 and 6.2: further derivation of x and y equations

Note that for our use case, because the rotation angle θ is a constant, h∙cosθ-k∙sinθ and h∙sinθ+k∙cosθ are also constants and can be replaced by single terms, c1 and c2 respectively. But we can name c1 and c2 whatever we like so I shall name them h and k respectively. (Yes, I know this seems like an odd maneuver, but it works fine for us and makes our equations a bit simpler.) With a little rearranging this then gives us the parametric form of a rotated parabola:

Equations 7.1 and 7.2: rearranging equations 6.1 and 6.2

Now we are ready to solve for the parabolic nozzle contour. Before we do this for the above equations, we shall discus the rest of the contour as well.  We will use a piecewise set of equations to describe the entire contour. For each piece we will need to solve for parameters and the split points between the different pieces of the function. We will go through each of them in order.  How to size a thrust chamber is beyond the scope of this paper and is an exercise left to the reader. You should already have calculated the radius of the chamber, throat, and nozzle exit, rc, rt, and re, respectively and the length of the chamber and nozzle, Lc, and Ln respectively.  The figure below shows a typical form of a thrust chamber. You’ll note that I’ve used z and r instead of x and y. This is a notation that I adopted early in my work writing a code for thrust chamber design. I figured z is up and rockets go up, so that’s where I’ll put the axis of symmetry of the chamber. I put the throat at z=0 for convenience. I decided to keep my notation in this report, instead of adopting something more conventional, for my own convenience of not having to rewrite all of my equations.

CAD drawing of nozzle geometry with refernce points and radii, note the parabola axis is canted

The first part of our piecewise function f1 is the chamber section representing by red in the figure. f1 is a simple horizontal line with the equation. 

f1 Chamber Section

Equation 8 descrbing the straight chamber section before the converging section that follows

f2 Concave Curved Converging Section

The exact converging section shape typically does not strongly affect engine performance. So, the designer has a lot of freedom here butthe following geometric convention is common. Three parts are used, f2 (blue in figure) is a concave circular arc, followed by f3(green in the figure) is sloped line and f4 (orange in the figure) is convex circular arc. I’ve found that using the parametric form of a circle makes the math easier. The parametric form of this circular arcin function 2 is the following. 

Equation 9.1 amd 9.2 for a concave-type converging section

Were rf2 is the radius of curvature for function 2. 

f3 Linear Converging Section

The third piece is a straight line and we will use the point slope formof a line.  The designer provides the converging angle and a typical rule of thumb value is 30°. Using a bit of trigonometry, we can see that the slope will be tan(-θc). For the point we will use (z4, r4) which is where f3 intersects f4.

Equation 10: linear converging nozzle

f4 Convex Curved Converging Throat Section

The fourth piece is another circular arc. The radius of curvature is typically given by the designer as a multiple of the throat radius. A typical a rule of thumb value of 1.5 time the throat radius is commonly used. We will call this scalar rc2. Again, we will use the parametric form.

Equation 11: convex curved converging section

About solving for the unknowns upstream of the nozzle throat

Solving the converging section is where we will notice a hiccup (minor problem). Say that you have a large throat and a small contraction ratio, Ac/At. This could mean the point where f4 meets the given converging angle, θc, of f3, could be equal to or larger than the chamber radius. A mathematical solution that uses the given parameters would fail to produce real geometry. A second problem could arise even when a valid solution does exist because the result could be an undesirably sharp radius if the point is too close to the chamber radius.

So, we have two solution paths. For the conventional case, we don’t run into the previously mentioned hiccup. Here, functions f1, f3, and f4 are sufficiently defined by the given parameters mentioned so far. But to define where the function f2 is located, we will need an additional parameter.  In my work, I’ve decided to do this by setting the point r3 to be a percentage of the vertical distance between the chamber radius and r4. Where the concave f2 curve meets the linear converging portion, f3, is the location of r3. Where f3 meets the convex f4 curve is r4. My default value is 50% of the way between r4 and rc. I’ve named that parameter f2vp and its value should be between 0 and 1. The letters used mean: function 2, vertical percentage. (You are probably asking yourself “Why not just define rf2 in the same way as rf4, with a scaler multiple of rt?” Well good question. That’s the way I started doing it. It’s been a while but if I remember correctly, I found that it was hard to pick a default value that worked for a large variety of engines and if picked poorly results were either undesirable or didn’t solve. So, I came up with the above method which I found much more satisfactory.)

For the second solution path an alternat method for defining the converging section must be used. It will be convenient to set the radius of both of the concave, f2, and convex, f4, sections equal. Then we define the curved converging sections, f2 and f4, to each take up a percentage of the radial distance between the throat and the chamber radius. The linear section is the remaining percentage. I’ve named this parameter f2vp_alt. The name means: function 2, vertical percentage alternate method. Since there are 3 functions that take up this vertical distance, and two of them are equal the value for f2vp_alt must be between 0 and 0.5. If it is 0 that means that the linear portion takes up all the vertical distance and the curved portions would disappear. If it is 0.5 then the linear portion would disappear. The default vale I use is 1/3. This means that f2 takes up 1/3 of the distance, and f4 takes up 1/3 of the distance and the linear section, f3, takes up the remaining 1/3. It makes for decent looking converging contours.

This will always be solvable with real geometry, but means that theeither the converging angle or the radius of curvature for the converging portions cannot be pre-defined and is instead solved for. At first, I tried to let the converging angle, θc be derived. However sometimes the solution would be a very shallow angle creating an overly long converging section. Instead, I found better results by keeping the designer given θc and instead solve for the radius of rf2and rf4, which would still be kept equal. I was resistant to this at first because it would remove the ability to use the rule of thumb value for rf4, or any specified value for that matter. This will only work if the assumption that the designer will provide reasonable values for θcis met. To summarize, the primary method lets the radius of function 2, rf2, be an unknown to be solved for, and the converging angle of f3is given and the radius of f4 is given. With the alternate method the converging angle is given along with the vertical space the functions occupy. This leaves the radiuses of curvature rf2 and rf4, which are equal, to be solved fore.

When I have designed thrust chambers, I use the first method and then only switch to the second method if the first method does not work. The remaining question is when to switch to avoid a sharp corner. My solution was to specify that the concave radius, f2, take up a minimum percentage of the vertical distance between the throatand chamber radius. My default value for this is 15% but this was somewhat arbitrary. You can also avoid the second method by increasing the thrust chamber contraction area ratio if possible.

Converging section solution, Method 1

The following is a mathematical solution for the converging section when using the regular method for the converging section described above. From f4 we need to solve for z4 and its corresponding radial point r4. For that we need to find when the circular arc is at the givenconverging angle. This is when its derivative is equal to the slope of f3, the linear section, which is tan(-θc). We will take the derivative ofthe parametric equations of function 4. In case you forgot from your calculus class the derivative of a parametric equation is found by taking the derivative of each equation separately with respect to the parametric independent variable, in this case t, which will give us the equations for and . Then you divide the equation for by the equation for and cancels out and you are left with an equation for . Using this method on the equations 11.1 and 11.2 for function 4 we get

Equation 12:

We solve this at t4 by setting the derivative equal to tan(-θc) which gives us

Equation 13:

Solving for t4 we get

Equation 14:

Note that this solution is periotic so it mathematically solves with any added to it. Through quick trial and error, we find a value n=1will put point (z4, r4) in the correct quadrant on the x-y plane. We then plug t4 into f4 to get the z4 and r4 coordinates. 

Equations 15.1 and 15.2:

Now that we know that value of r4 we can now check if the criteria needed for switching to the alternate method is met. If not, we can carry on to find r3. Recalling our assertion that r3 shall be some given fraction of the vertical distance between rand rc we see that it would be given by the following equation.

Equation 16:

Next, we plug r3 into the f3 function [eq 10] and solve to get z3.

Equation 17:

Now all that remains to be solved is z2 and the radius of curvature, rf2, for the concaved converging function f2. If we take f2 [equation 9] and plug in the known point z3, and r3, which both occur at t3, we get the following:

Equation 18.1 and 18.2:

We can see that the there are three unknowns z2, rf2, and t3, but there are only two equations. So, we need a third equation to solve this system. We know the derivative of this function is the same as for the function f4, [equation12] since they are both the equations of a circle. And we know the value of the derivative at this intersection pointfrom the slope of f3, the linear converging section, which is tan(-θc).Setting equation 12 to tan(-θc) gives us the third equation we need. 

Equation 19:

We then solve this system of three equations, 18.1, 18.2, and 19 for the three unknows z2, rf2, and t3. This gives us:

Equation 20:
Equation 21:
Equation 22:

We now have all the parameters and split points needed for each of the function to fully define the converging section. Equation 20 like in equation 14 also has a periodic portion of the solution, but here n=0.

Converging section solution, Method 2 

With the second method everything is the same as the first method up to calculating t4 [equation 14] then the methods diverge. In method 2, after calculating tthe next things to calculate are r3 and r4. This method defines the percentage of the radial distance between the throat and the chamber that the two curved converging functions, f2and f4, take up. This definition allows us to create, by reasoning, the equations for r3 and r4. They are:

Equation 23:
Equation 24:

Next the radius rf4 can be found by taking f4r [equation 11.2] at t4 when it will equal r4,

Equation 25:

and then solving for rf4 and we get 

Equation 26:

Then we set rf2=rf4. For the remaining unknowns that define the geometry, we can use the same math from the other method. Those remaining unknowns are z4, z3, t3, z2 and are defined by equations 15.1, 17, 20, and 22, respectively.

f5 Diverging Throat Section

For the diverging section function 5, magenta in the figure, is again in the parametric form of a circle.

Equations 27.1 and 27.2:

The radius of this function, rf5, is the diverging radius of the nozzle throat, it is given by the designer and is usually a rule of thumb value of 0.382 times the throat radius. But we still need to find the point z5, and r5. To do this we first note that the derivative of f5 is thesame as for f2 and f4 given in equation 12. We know what the angle of the tangent should be at z5, where f5 meets the parabolic portion. It should be the θn mention above and gotten from the θn vs expansion area ratio graph. From the same trigonometry used for the slope of f3we can set the derivative [eq 12] equal to tan(θn) at t5. Then we solve for the parameter twhich gives us:

Equation 28:

Like in equation 14 this also has a periodic portion of the solution, but here n=0. We then plug t5 into f5 [eq 27] to get z5 and r5.

Equation 29:
Equation 30:

f6 Parabolic Nozzle Section

The parabolic section of the nozzle, function 6, is yellow in the figure. As derived above in equation 7 we use a pair of parametric equations for a parabola with a rotation applied. I like to use ψinstead of θ here for the sake of symbolic variety.

Equations 31.1 and 31.2:

Solving for the unknowns downstream of the nozzle throat 

All that’s left is to solve for the remaining six unknowns in function 6. These unknowns are, t5, t6, m, h k and ψ. Note that the parameter variable t is unique for each section, so it’s technically a different t5 than for the t5 for function 5.  We know six key data points on the parabola. We know the starting coordinates z5, and r5 from our prior work. We know the end coordinates z6, r6 from the exit radius and the length of the nozzle provided by the designer. If we substitute these coordinates into function 6, we would have the following four equations:

Equation 32:
Equation 33:
Equation 34:
Equation 35:

But it’s not possible to solve these four equations for six unknowns. We need at least two more equations. We know the slope at the beginning and end of the parabola from θn and θe which we got from the θn and θe vs expansion area ratio graphs. This means we can use the derivative of function 6 to give us the last two equations. To find the derivative function first we take the derivative of each parametric equation with respect to the parameter variable, t.

Equation 36:
Equation 37:

And then we divide by the equation for and we get:

Equation 38:

Taking this function at t5 and t6 and using trigonometry to find the slope of the tangent lines at beginning and end of the parabola, we see that the derivative should be tan(θn) and tan(θe) respectively. This gets us last two equations we need for a system of six equations to solve for our six unknowns. 

Equation 39::
Equation 40:

Now all we have to do is simultaneously solve the system. Here are the six equations fully written out.

Equation 41:
Equation 42:
Equation 43: exit axial position
equation 44: exit radius value
Equation 45: nozzle tangent angle
Equation 46: nozzle exit angle

As far as I can tell this can’t be solved symbolically. I’ve tried several methods including a few computer algebra systems, but always failed. (Give it a try yourself and let us know what you come up with.) 

Instead, I solve the system of equations 41 through 46 numerically for m, h, k, phi, t5 and t6. This can easily be done with a calculator like the Texas Instruments, TI-89. But the way I solve this and the rest of the calculations is with a piece of custom software I wrote. My programing language of choice for doing this has been Octave, but it is possible to do with any Turing-complete programming language.  Once you’ve done this and solved for those unknowns you now have a fully mathematically defined parabolic nozzle contour.  One final note worth mentioning is that I typically use these equations to find the value r at a particular z location. If this z is in one of the parametric sections, then I first solve the equation fz(t)=z for t. Then I take the value for t and use it to solve fr(t)=r for the value of r.

Hopefully, that clears up any confusion and set the record … canted.


List of symbols

Ac​ = Cross sectional area of the thrust chamber

Ac/At​ = Thrust chamber contraction area ratio

Ae​ = Cross sectional area of the nozzle exit

At​ = Cross sectional area of the nozzle throat chamber

c1-c2​ = Numerical constants

f1-f6​ = Parts of the piecewise function that defines the nozzle contour

f2vp​ = Fraction of the vertical distance between r4 and rc occupied by f2

f2vp_alt​ = Fraction of the vertical distance between rt and rc occupied each by f2 and f4

fr1-fr6​ = Radial portion of the of the piecewise functions that defines the nozzle contour

fz1-fz6​ = Axial portion of the of the piecewise functions that defines the nozzle contour

h​ = Horizontal translation of parabola

k​ = Vertical, or radial, translation of parabola

Lc​ = Length of the thrust chamber, measured from the injector to the nozzle throat

Lf​ = Length of a bell nozzle given as a percentage of an equivalent 15° half angle conical nozzle

Ln​ = Length of the nozzle, measured from the nozzle throat to the nozzle exit

m​ = Scalar that is the stretch factor of a parabola 

n​ = scalar for periodic solution, n can mean any whole number

r​ = Radial coordinate

r1 – r6​ = Radial coordinate of points 1 through 6, see figure for location of points

rc​ = Radius of the cylindrical portion of the thrust chamber

rc1​ = Scalier for the concave converging radius rf2 = rc1 ∙ rt

rc2​ = Scalier for the convex converging radius rf4 = rc2 ∙ rt rule of thumb value of 1.5

rd​ = Scalier for the diverging arc radius rf5 = rd ∙ rt rule of thumb value is 0.382

re​ = Radius of the nozzle exit

rf2​ = Radius of curvature for function 2 

rf4​ = Radius of curvature for function 4

rf5​ = Radius of curvature for function 5

rt​ = Radius of the nozzle throat

t​ = Parametric independent variable, it is unique for each pair of parametric equations

t1 – t6​ = Parametric independent variable at points 1 through 6, and unique for each parametric function

x​ = Horizontal coordinate

y​ = Vertical coordinate

z​ = Coordinate perpendicular to the x-y plane

z1 – z6​ = Axial coordinate of points 1 through 6, see figure for location of points

ε​ = Nozzle expansion area ratio, At/Ae

θ​ = Rotation angle of a parametric parabola

θc​ = Converging angle of the linear converging section, function 3

θe​ = Angle between the nozzle centerline and parabola tangent line at the the nozzle exit

θn​ = Angle between the nozzle centerline and parabola tangent line at the beginning of the parabola

ψ​ = Rotation angle of the parabola of the nozzle contour


For questions and comments:

research @ rrs.org

MTA Launch Event, 2021-10-16, First Update

by Bill Claybaugh, RRS.ORG


In a remarkable demonstration of persistence and luck, RRS President Osvaldo Tarditti was able to find the spent booster rocket. A few photos were captured of the recovered rocket.

Bill Claybaugh’s recovered spent booster casing brought back to the Mojave Test Area (MTA)
Closeup on the bulkhead shoved into the aluminum case of the booster from the impact.
The fins look great and the nozzle was recovered.

Based on the impact location, it was possible to reconstruct a possible flight trajectory by assuming the motor performed as designed and further assuming the front of the vehicle was a flat plate and that the mass did not include the mass of the payload.  We know from video, telemetry, and recovery of the payload that the payload separated from the booster about one second into the flight.

The recovery location on the map shows a northeast trajectory as confirmed by launch footage.

This analysis suggests a burnout velocity of about 1550 feet/second with a peak altitude of about 21,200 feet given the known range of about 14,300 feet. This analysis gives a flight time of about 74.5 seconds and an impact velocity of about 820 feet/second.

Given the observation that the vehicle stopped in about 2 inches (based on the depth of the depression in the hardpan) before falling on its side; we can estimate the impact deceleration.  Given an average velocity during impact of about 410 feet/second because the final velocity is zero and it took only 0.167 feet to come to rest, it follows that the impact occurred over 0.000407 seconds.  This, in turn, indicates an average deceleration of about 31,250 g’s.

The reason for the vehicle turning to the Northeast starting at about 0.20 seconds into the flight remains unclear. There is no evidence either in video or in images of the recovered hardware of any hot gas leak nor of any transient thrust vector anomaly.  The wind was less than 5 miles per hour and from the Northwest; if it had caused a turn, we would expect it to be toward the Northwest, not the Northeast as observed.  The only plausible theory at this time is that part of the belly-band became embedded between the nose of a fin and the rocket body causing the turn via differential drag and then fell away from the vehicle, causing the resumption of normal flight.  Once the recovered hardware is available for inspection, we will test each fin nose to see if a gap exists that might have caught the 0.020-inch thick belly-band.

The recovered payload segment was examined after it was found just north of the launch site.

It also remains unclear as to why the payload separated about 1 second after launch.  The recovered payload showed that both initiators had fired (by design, if one fires the other is ignited; thus, only one signal is required to fire both) but did not show any evidence of having been “swaged” or otherwise subject to being forced off the rocket by aerodynamic or other forces. Neither does the matching front end of the rocket show any evidence for the payload having been forced off. We thus conclude that one of the flight computers ordered the firing of the initiators.

The bellybands being fit checked in the launch rail.
Recovered bellybands have evidence of tearing from what is likely fin impact.

However, the main flight computer stopped working just after 0.80 seconds into the flight for an unknown reason after recovery it was still connected to its battery, which showed the expected 3.87 volts. Further, the limited data recovered from that computer shows that it did not initiate separation of the payload: the firing circuit shows continuity throughout the period that the computer was operating and separately records that no signal was sent by that computer.

Still image of the rocket just after launch making the unexpected hard turn.

This points to the backup flight computer.  That hardware is currently at the manufacture for repair, after which we hope to extract continuity data with regard to its firing status.  Hopefully, once that and other data is available from the backup computer we will be able to establish when it ordered the separation of the payload, and why.

Recovered payload with the main and backup computer.

A second update to this firing report is expected. The booster has been packaged up for a more detailed inspection.


Tank Blowdown Math

by Prof. Dean R. Wheeler, Brigham Young University


EDITOR’S NOTE

This posting is reprinted from the original article written March 13, 2019 with permission from the author. This article was intended for chemical engineering students to size relief valves for pressure vessels, but it applies well to amateur liquid rocketry as many use a pressure fed system to deliver propellants to the engine.

The PDF of this white paper can be found below.

https://www.et.byu.edu/~wheeler/Tank_Blowdown_Math.pdf

The RRS has several members engaged with liquid rocket projects. An important part of analyzing the performance of those systems is the pressurization system that drives the propellant into the engine. The tank blowdown problem is useful to designing the system and estimating performance. This derivation goes through the thermodynamics of the general tank blowdown problem and should be a useful starting point for a pressure-fed liquid rocket project.


INTRODUCTION

This document provides a mathematical model for computing the rate of expelling gas through a small orifice or nozzle attached to a tank. Furthermore, two models are described for how fast the tank will depressurize. Related material on compressible flow can be found in fluid mechanics and thermodynamics textbooks and web pages.

Figure 1 shows the tank and associated nozzle. The narrowest diameter of the flow path in the orifice or nozzle is known as the throat region. The tank and throat regions are described with their own sets of equations.

Provided the tank is large and the throat is small, it will take many seconds to empty the tank and gas velocities in the main part of the tank will be much smaller than the speed of sound. This means that gas pressure, temperature, and density in the tank will be spatially uniform, though they will be changing in time. Thus, we describe the tank using a transient mass balance. One can compare this to a model in heat transfer known as lumped capacitance.

In the nozzle region however, gas velocity is large and there are large spatial variations in the gas properties. In addition, there is relatively little gas contained in the nozzle region. Thus, flowrate in the nozzle adjusts rapidly to match current conditions in the tank, making it seem as if the nozzle is operating at steady state. This approximation for the nozzle is known as quasi-steady state.

Figure 1: Schematic of a task with nozzle or orifice, allowing gas to exit. Italicized are variables that pertain to twokey regions. During blowdown every variable depends on time,

EQUATIONS OF STATE

The P, T, and rho variables in Figure 1 denote absolute pressure, absolute temperature, and density in the tank or the narrowest part of the nozzle or throat (denoted by an asterisk,*, subscript), respectively. Note that if tank pressure is given experimentally as a gauge quantity, it must be converted to absolute to be used in the equations below.

The first relationship between gas variables is given by an equation of state. The ideal gas law is a fairly accurate representation for air when pressure is less than around 10 atmospheres or 150 psia. It states that:

Figure 1: The ideal gas equation

where “V” is the volume of the gas, “n” is the number of moles, and “R” is the universal gas constant (8.31446 J/mol/K). With the introduction of the molecular weight, M (effectively 0.028964 kg/mol/K for air), and the substitution that density is mass over volume, rho = n M / V, the ideal gas law is changed to

Equation 2: Density calculated from the ideal gas equation

This equation could be applied separately to the tank variables or to the thrust variables.

TEMPERATURE AND PRESSURE DURING EXPANSION

The second important relationship comes from figuring out what happens when gas in the tank or nozzle expands. When a gas expands, its internal energy is used to perform work on the surroundings, and the gas therefore tends to cool off. If the gas expands slowly, there is time for itmto absorb hest from its warmer surroundings and the expansion is essentially isothermal, meaning the temperature stays at its initial value or that of the surroundings.

On the other hand, if a gas expands quickly its temperature will drop dramatically. This is called adiabatic expansion, where adiabatic means no noticeable heat transfer from the surroundings (i.e. the walls of the tank). In adiabatic expansion, the pressure drops more rapidly than it would for an isothermal (slow) expansion. Adiabatic expansion could haolen inside the tank if it is emptying rapidly, but this depends on the relative sizes of thr tank and nozzle. On the other hand, adiabatic expansion certainly occurs when a gas moves from the tank through the nozzle region. In other words, here the gas is moving quickly and therefore expanding quickly.

The thermodynamic relationships for pressure and temperature for reversible adiabatic expansion of a constant heat capacity ideal gas are:

Equation 3A: Adiabatic pressure and density relationship
Equation 4A: Adiabatic temperature and density relationship

where the subscript, “o” indicates the initial state of the gas before the expansion started. This means if we know how the density is changing from an initial state to some later state, we can compute P and T as well. In the case of the nozzle, we apply the above equations as the gas travels between the tank and the throat. In the case, they become

Equation 3B: Adiabatic pressure and density relationship between tank and throat regions
Equation 4B: Adiabatic temperature and density relationship between tank and throat regions

The parameter, “gamma” , is the dimensionless ratio of specific heats ( gamma =. Cp / Cv ), and by statistical theory of gases, gamma = 7/5 = 1.4, for low temperature diatomic molecules, nitrogen (N2) and oxygen (O2) and so that value is used here.

CHOKED FLOW

Next, we need to determine the gas density in the nozzle when the tank is at a specified conditions. Recall that that the nozzle is treated as if it instantaneously responds to whatever state the tank is in. A fuller discussion of the nozzle flow equations can be found in other sources like textbooks that cover ideal compressible flow in nozzles.

Choked flow means that the flow is exactly at the speed of sound in the throat region. A higher speed cannot be achieved in the throat, regardless of upstream or downstream conditions. Thus, choked flow acts to limit how much gas flow can pass through a given size orifice, This is the reason why pressure relief valves on tanks must be properly sized to accommodate sufficient flow.

Choked flow happens for a large pressure drop across the nozzle or orifice, specifically if the upstream tank pressure meets the following condition relative to atmospheric pressure downstream from the nozzle:

Equation 5: Choked flow condition

Equation 5 is the origin of the rule of thumb or approximation that choked flow occurs for upstream pressure that is more than twice the value of downstream pressure (absolute). If the tank pressure drops below this limit, the speed of gas in the throat is subsonic, and less gas will flow than in the choked flow regime. The solution to subsonic flow in the nozzle is complicated and is less important to know because it is at the end of the tank’s discharge when pressure is low, and so will be neglected here.

The solution to choked flow in the throat region follows a simple relationship, derived from energy and mass balances:

Equation 6: Throat to tank density ratio

This can be substituted from Equation 3B and 4B to determine pressure and temperature in the throat in terms of tank conditions.

For choked flow the throat velocity is exactly the speed of sound, which is what makes it easier to analyze. For ideal gases, speed of sound, c, is determined solely by temperature. Thus, we can relate throat velocity to throat temperature, and in turn to tank temperature:

Equation 7: Speed of sound at the throat

For example, if T_tank = 294 Kelvins, then c_o = 314 m/sec for air.

MASS FLOW RATE

Now we can determine the mass flow rate, “m_dot”, through the nozzle or orifice. This comes from the following standard relationship, applied at the throat, because that is where conditions are known:

Equation 8: Mass flow,rate at the throat

where “A_*” is the throat cross-sectional area given by

Equation 9: Area of a circle

and where “d_*” is throat diameter.

Dimensionless parameter, Cd, in Equation 8 is the discharge coefficient, accounting for friction between fluid and walls and a phenomenon known as vena contracta. In essence, Cd, is needed in Equation 8 because the effective area for fluid at speed, v_o, is somewhat smaller than actual throat area. Cd would be equal to 1.0 for a perfect (frictionless or thermodynamically reversible) nozzle: in practice for a smoothly tapering nozzle it might be as high as 0.98, while for a sharp-edged orifice it might be as low as 0.60. Anything that causes separation of flow from the nozzle wall or increases frictional contact will decrease Cd.

Making the appropriate substitutions into Equation 8 leads to an equation for mass flow in terms of readily determined quantities:

Equation 10: Mass flow rate in terms of readily determined quantities

Frequently in industrial situations, mass flow rates are expressed instead as volumetric flow rates corresponding to a gas at a standard temperature and pressure (even though the gas is not actually at that temperature and pressure). For instance, a mass flow meter used for gases may express mass flow as standard liters per minute (SLPM) or standard cubic feet per minute (SCFM). In other words, even though m_dot (mass flow) is the key value being measured, it is expressed as

Equation 11: Standard volumetric flow and mass flow rate

which requires knowing what rho_std value is programmed by the manufacturer into the flow meter. This can be determined from the ideal gas law, given specified P_std and T_std values. As an example, the American manufacturer, Omega, assumes a standard temperature “T_std” of 70 degrees Fahrenheit (294.26 Kelvins) and a standard pressure “P_std” of 1 atmosphere which equals 14.696 psia (101,325 Pscals) thus by the ideal gas law, the standard density “rho_std” would equal 1.2 kg/m3 for air (molecular weight 28.97 g/mole).

Combining Equations 10 and 11 and the ideal gas law leads to

Equation 12: Combining Equations 10 and 11 for standard volumetric flow rate

where “c_std” is the speed of sound at the standard temperature:

Equation 13: Standard volumetric rate and mass flow rate relationship

Makers of valves and orifices may provide an experimentally determined size parameter known as flow coefficient, Cv. For gases this dimensionless parameter can be converted to Cd*A_* by

Equation 14: Discharge area relationship4 to valve coefficient (metric units)

The key design principles resulting from the above analysis are, provided tank pressure is large enough to generate choked flow, that (1) mass flow rate of a gas through an orifice is proportional to throat area and tank pressure and (2) flow rate does not depend on downstream pressure.

TWO MODELS OF TANK BLOWDOWN

Equation 10 gives the rate of mass loss from a tank at a given gas density and temperature. To determine how long it will take to depressurize the tank, we must do a transient mass balance on the tank. The ordinary differential equation for this is:

Equation 15: Change of mass in time

where “m_dot” comes from Equation 10 and “m” is the mass of gas in the tank. This in turn is:

Equation 16: Mass in the tank

where V_tank is the fixed tank volume. With these substitutions we get for the governing equation

Equation 17: Mass flow rate from the tank

To make things more manageable, let us create a discharge time constant called “tau”

Equation 18: Time constant for blowdown of a tank

where “c_o” is the speed of sound at the initial temperature “T_o” (i.e. at the beginning of blowdown)

Equation 19: Speed of sound at initial conditions

With this new time constant, Equation 17 becomes:

Equation 20: Mass flow rate change in the tank

The last thing to do before solving this equation is figure out what to do with T_tank. We have two options:

ISOTHERMAL TANK ASSUMPTIONS

Assume gas temperature in the tank does not change in time, based on blowdown taking a long time so that heat can be readily absorbed from the walls. Thus, T_tank = T_o. This leads to Equation 20 becoming

Equation 21: Tank density change in time

which can be separated and integrated to give the solution

Equation 22: Tank density as a function of initial conditions

where “rho_o” is initial density in the tank. We then convert densities to pressure using the ideal gas equation.

Equation 23: Tank pressure as a function of initial conditions

The equation tells us how tank pressure varies with time, for an isothermal tank and choked exit flow.

ADIABATIC TANK ASSUMPTIONS

Assume the gas cools as it expands in the tank, due to no heat transfer from the walls, based on the blowdown taking a short time to complete. Thus, T_tank is given by Equation 4A. This leads to Equation 20 becoming

Equation 24: Mass flow rate from the tank

which can be separated and integrated to give a solution.

Equation 25: Density of the tank as a function of time

We then convert densities to pressures using Equation 3A for adiabatic expansion.

Equation 26: Tank pressure as a function of time

This equation tells us how tank pressure varies with time, for an adiabatic tank and choked exit flow. The tank temperature can likewise be predicted from Equation 4A.

Equation 27: Tank temperature as a finction of time

COMPARISON OF THE TWO MODEL ASSUMPTIONS

The isothermal and adiabatic models of tank blowdown can be considered two extremes, with the correct answer (i.e., with the true amount of heat transfer) lying somewhere in between them. Figure 2 shows an example of the respective blowdown curves (Equation 23 and 26). As noted previously, adiabatic tank conditions lead to more rapid pressure loss than do isothermal conditions.

The curves predict that the tank will have lost 80% of its original pressure at a time in the range of 1.3*tau < t < 1.6*tau. This shows the value of evaluating the variable, tau, to get an approximation of the time it takes to depressurize the tank.

Figure 2: Comparison of isothermal and adiabatic blowdown curves.

izmir sex shop
izmir sex shop
izmir sex shop
izmir sex shop
izmir sex shop
izmir sex shop samsun sex shop
samsun sex shop
samsun sex shop
samsun sex shop
samsun sex shop
samsun sex shop
malatya sex shop antalya sex shop antalya sex shop sex shop sakarya sakarya sex shop sakarya erotik shop sakarya sex shop erotik shop erotik shop erotik shop erotik shop erotik shop erotik shop erotik shop vibrator vibrator vibrator izmir sex shop izmir sex shop ankara sex shop ankara sex shop antalya sex shop penis pompasi penis pompasi balıkesir sex shop