Module 5 · Calibration and validation
This module is still locked
To unlock it, pass the Module 4.5 quiz with at least 70%.
Go to the quiz · Module 4.5Bulk density
Lesson transcript
And Bulk density is the independent test. It fills a container of known volume, lets the bed settle, and adjusts the solid density until the bulk density matches your target. The relationship is simple. Bulk density is the packing fraction times the solid density. Packing is usually between zero point five five and zero point six four. If it comes out too high, friction is too low. If it comes out too low, there is too much friction or cohesion.
Check what you learned
5 multiple-choice questions. Get at least 4 right to complete the module. You can try as many times as you like.
Transcripts of every lesson in this module
1. Parameters you can trust
You built the case, and the simulation ran. But where do the numbers come from? Contact friction and restitution are not in any catalog. They depend on the material, the shape and the chosen model. The way out is to run simple tests and adjust the simulation until it repeats the test. That is calibration. Keep three things apart. A material property belongs to the body: density, Young's modulus, Poisson's ratio. An interaction belongs to the pair of materials that touch. And a calibrated parameter is the value fitted to a test, for that model. It can be quite different from the literature value. The literature has a surprising case. A study from two thousand eighteen modeled soybeans as clumps and arrived at a restitution of zero point one seven. Another, from two thousand twenty-two, modeled them as spheres and arrived at zero point six zero. The grains and the tests are not the same, but much of the difference comes from the model: the parameter makes up for what the model does not represent. So a calibrated value holds for that shape, that size and that model.
2. Verify, calibrate, validate
Three words that sound alike, but are not. Verifying means checking that the program solves the equations correctly, by comparing with analytical solutions. Calibrating means adjusting parameters until a test is repeated. Validating means predicting something else, which was not used in the fitting. Two rules to keep. If you calibrate and validate with the same test, you have validated nothing: you have only shown the fit worked. And the acceptance criterion is written before you see the result. No test measures everything. The drop test sees restitution. The inclined plane sees friction and rolling against the wall. The pile responds to friction between particles. But different pairs of friction and rolling values can give the same angle. The drum sees everything together. That is why order matters: first fix what an isolated test can measure. Three findings show up in many sources. Restitution barely affects the angle of repose. Don't spend fine calibration on it. Static friction dominates the angle, and rolling comes second. And the gain diminishes: beyond a certain value, raising friction changes the angle less and less.
3. Drop test and restitution
The simplest test is the drop test. You drop a particle and measure the rebound height. The higher it comes back up, the higher the restitution. Here is an example. The particle falls from thirty centimeters. And bounces back up to nineteen. Restitution is the square root of nineteen over thirty: zero point seven nine six. That means thirty-seven percent of the energy was lost in the impact. Open a new project, with no geometry, and go to the Calibration menu. The list is already in the right order: first the drop test, then the inclined planes, and finally the angle of repose. I open the first one, the particle drop. At the top of each wizard there is this button: it shows how to run the real test, on the bench. The test goes like this: you drop a particle from a known height onto a plate made of the wall material, without spin, and measure how high it bounces back, ideally on Then, the three blocks. Under Material, the Particle and Wall buttons open only what this test uses. Test data holds your measurements: thirty centimeters for the drop, nineteen for the rebound. Fixed parameters shows the values frozen from the other tests. And under Calibrate, Max simulations limits the attempts. I click Calibrate, and confirm. The program builds the test, runs it, compares the simulated rebound with your measurement, and corrects the restitution. Each line is one attempt. Done. In five simulations, it reached a restitution of zero point seven nine eight, and saved the value in the material. Check it with the formula: the square root of nineteen over thirty gives zero point seven nine six. The wizard got it right. Four pitfalls. A particle that leaves spinning bounces off the vertical. Restitution changes with speed: pick a height that gives a speed close to your equipment's. An irregular grain bounces sideways: repeat many times. And on the bench, use the same material and finish as the real wall.
4. Inclined plane: sliding or rolling
The inclined plane measures how the particle behaves on the wall. Tilt it slowly, and note the angle at which it starts to move. But there are two ways to start moving: sliding or rolling. And they are two different tests. Here is the difference. In sliding, the sample goes down without turning: the same face stays against the plate. What resists is sliding friction. In rolling, the particle turns. The contact point does not slip, and what resists is rolling friction. The sliding angle is usually larger. And the rolling angle, much smaller. A practical tip: put a mark on the particle and film it. If the mark turns, it is rolling. If you simply release a round grain, it will roll before it slides, and you will never measure sliding friction. So in the sliding test, glue three or more particles to a small plate, to stop them from turning. The rolling test is the opposite: a single loose particle. Now the second test, the sliding inclined plane. Look at the help. The key to the test is here: glue three or more particles to a small plate. That way the sample cannot roll, it can only slide. You rest that sample on a plate made of the wall material, and tilt it very slowly. The angle at which it starts to slide is the test result. Repeat five to ten times, and use the average. In the simulation, rolling is locked, just like the glue. I enter the angle, here twenty-five degrees, and calibrate. The result is the sliding friction against the wall. Done. In six simulations, sliding friction against the wall came out at zero point four five, and the simulated angle matched the twenty-five degrees Notice that the tangent of twenty-five degrees is zero point four seven. The quick math already gets close. Third, the rolling plane. Now it is the opposite: a single loose particle on the plate. No glue. Tilt slowly. The particle may settle a few times: ignore those moves. Record the angle at which it rolls continuously off the plate. This angle is usually low, five to fifteen degrees. In the simulation, wall friction is already fixed, from the previous test. That is why order matters. I enter twelve degrees, and the program adjusts the rolling friction. Done. In six simulations, rolling friction came out at zero point one nine, and the particle now rolls at the same twelve degrees. Theory says this value is close to the tangent of the angle, which is zero point two one. And the value was saved in the material. The quick math is the same for both. For sliding, friction is the tangent of the angle. For rolling, rolling friction is also the tangent of the angle. But there is a limit: for a sphere to roll without slipping, the tangent of the angle cannot exceed three point five times the friction. Beyond that, it rolls and slips at the same time. Here is rolling measured in the solver, in a test with a rolling friction of zero point three. Below sixteen degrees, nothing happens. From seventeen on, the sphere starts to roll. The tangent of sixteen is zero point two nine, and the tangent of seventeen is zero point three one. The threshold falls between the two, just as the math predicts.
5. Angle of repose
The most popular test is the pile. Fill a cylinder, lift it slowly, and the material forms a heap. The slope of that heap is the angle of repose. It sets the friction between particles. The fourth test is the angle of repose, which sets the friction between particles. On the bench, the material flows out of a tube that rises slowly, and forms a pile. You measure the slope of the heap. The measured angle goes here. Heap base, the base diameter, is optional, and helps separate the two frictions. Test tube is the pouring tube: zero means automatic, about twenty particle diameters. Fixed parameters shows the values inherited from the previous tests: the restitutions, the frictions and rolling. You can check and correct them without leaving the screen. Reproduce the bench copies your real test: tube lift speed, rest times and tube friction. Unchecked, the test is almost static. Use it when your bench test was not that slow. Max simulations and Angular error decide when to stop. The default of half a degree is already tighter than the noise of the pile itself. This is the slowest test in the menu, because it uses many particles. How does BrixDEM measure the angle? It looks at the pile from eight directions and measures the flanks in each one. Only settled particles count, and the result is the average of the last frames. The spread between sectors is reported too. If it is high, the pile came out lopsided, and that point deserves suspicion. Here is a trap. A pile is built by avalanches, which depend on the initial arrangement. Repeating the same test, a variation of two to three degrees is normal. Don't waste time chasing a smaller difference: you would be fitting noise. And always report the value with its uncertainty. Another common surprise. You raise rolling friction, and the angle grows less and less. Studies show that, with low sliding friction, rolling barely changes the pile: sliding is what limits the angle. In that case, sliding friction is the one to adjust.
6. Bulk density
And Bulk density is the independent test. It fills a container of known volume, lets the bed settle, and adjusts the solid density until the bulk density matches your target. The relationship is simple. Bulk density is the packing fraction times the solid density. Packing is usually between zero point five five and zero point six four. If it comes out too high, friction is too low. If it comes out too low, there is too much friction or cohesion.
7. Calibrated cohesion
The fifth test is the cohesive pile. In Pile calibration mode you choose what the optimizer is allowed to change. Dry adjusts only the frictions. Cohesive, cohesion only, is for when the frictions were already calibrated dry. Rolling plus cohesion is the default for cohesive material, because those two move the angle the most. And the advanced mode moves all three, but it is more expensive, and can give equivalent solutions. In Cohesion model, choose linear or JKR. That value goes into the pair's interaction. With cohesion, the challenge is bigger. Several different parameter sets can give the same pile. So calibrate at rest. And confirm with a second test, ideally with the material flowing, at several stress levels. Each model has its own parameter. For linear, the energy density k. For JKR, the surface energy gamma. For the liquid bridge, most of it comes from measurement: the liquid content and the surface tension. Keep the contact angle at zero, the recommended range for the program's correlation. Also check the liquid balance in the results. Breakage has its own path. Start with drop-weight tests, at several energies and sizes. Fit the breakage probability first. Then t ten and the fragment size distribution, with sieving. And compare against energies you did not use in the fit. A chart in the preview is not validation.
8. Independent validation
Calibrated? Now comes the proof. Compare with something you did not use in the fitting. Here, the discharge of a silo with glass beads, from the test by González-Montellano and co-authors, published in two thousand eleven. The simulation gave twenty-nine point six five seconds, and the experiment, twenty-nine point two seven. The difference is one point three percent. Be honest about this result. The record is still provisional. A total time is not the discharge curve. The flow pattern still has to be compared. And the full project has to be archived, so anyone can repeat it. How good is good enough? These are starting points from the manual. Angle of repose, within two to three degrees. Bulk density, within five percent. Hopper flow rate, within ten percent. Drum, the right trend and within three degrees. Write your own criterion before running. If it does not match, don't start changing the material. Check that the time step has converged: simulate with half of it. Check that the mesh is adequate. Repeat the case with a slightly different initial arrangement, to separate noise from trend. And compare the scaled-up case with a fine one. Finally, keep everything. Which test, under which conditions, and on what date. Which parameter you obtained, and with what uncertainty. Which program version. And the full project in one folder. Without that, nobody can repeat your result, not even you.
9. Pitfalls
A calibrated parameter holds for the setup in which it was measured. If you calibrated with one size and use another, recalibrate. Calibrated spheres and will use clumps? Recalibrate. Changed Young's modulus or the contact model? Same thing. Coarse-graining is different: calibrate at the real size, and the program scales the parameters on its own for the larger particles. Only recalibrate if you check Reduce E by CGF. And always compare with a fine case. Four more mistakes. Calibrating out of order carries the error into the next tests. Forgetting the wall: the pile sits on it. Copying a value from a book without checking shape, size and model. And trusting a result that changes when the time step is halved.
10. The right order
To wrap up, the order that works. First, restitution. Then friction against the wall. Next, rolling against the wall. Then friction between particles. And last, cohesion. Each test uses the result of the previous one, so the order matters. Bulk density is independent. And finally, validate with a dynamic test, such as a drum or a hopper.