Open menu Close menu
Particle

An uncalibrated DEM model is an animation, not a calculation

Which bench tests constrain each contact parameter, the order they are fitted in, and why a calibration only holds inside the envelope it was tested on.

Published

Rotating drum test and its DEM twin · matching angle of repose
Rotating drum test and its DEM twin · matching angle of repose

In finite elements, Young’s modulus for steel comes off a table. 200 GPa is a measurable material property and it transfers from one model to the next without argument.

DEM has no such privilege. Contact stiffness, restitution, sliding friction, rolling friction and cohesion are not properties of the ore. They are properties of the numerical particle, and that particle — a sphere or a clump a few centimetres across — does not exist in the pile.

Why you cannot just measure it

A production run of a transfer chute uses particles far larger than the real fines and stiffnesses reduced by orders of magnitude. That is the only way the model finishes in days rather than months. But the moment you scale, any measurement taken on a single grain stops transferring.

The practical consequence: calibration does not fit particle properties, it fits bulk responses. And it needs more than one response, because the map from parameters to a single measurement is not injective. A repose angle of 37° is reproduced by many combinations of rolling and sliding friction, and each of them discharges differently.

The tests and what each one constrains

Material is sampled on site, quartered, characterised by particle size distribution, and its moisture is measured on the day of the test. A sample that spent two weeks travelling in an open drum is no longer the material under discussion.

TESTRESPONSE MEASUREDPARAMETER IT CONSTRAINS
Repose angle · lifted cylinder or drumStatic and dynamic angleRolling and particle-particle friction, coupled
Shear cell · Jenike, annular or rotationalYield locus, internal friction, cohesion under consolidationCohesion and internal friction
Bulk density and compressibilityPacking as a function of stressClump shape and friction
Wall friction against the real liner plateWall friction angleParticle-wall friction
Drop test onto a liner couponCoefficient of restitutionNormal restitution
Hopper discharge · flow rateMass flow rate and flow patternNone → reserved for validation

The last row is the rule most often broken. Once the discharge test is used to fit, it can no longer verify, and the calibration loses the only independent check it had.

The order of the fit

  1. Sensitivity screening. A design of experiments over the plausible range shows which parameters actually move each response. In dense slow flow, restitution barely registers; in free fall and impact, it dominates. Normal stiffness has little influence as long as overlap stays small.
  2. Simultaneous fit against two or more responses. Fitting one at a time hides the coupling: you close the repose angle and open up the bulk density.
  3. Validation against a test that never entered the fit. If the discharge rate lands 20 % off, the calibration failed even when the repose angle closes within 1°.

Reduced stiffness and time step

Real rock stiffness drives the Rayleigh critical time step down to the order of nanoseconds. So contact stiffness is reduced and the model runs at 20–40 % of the critical step.

That reduction holds as long as the flow is not stiffness-dominated and overlap stays small. The usual working limit is a mean overlap below 1 % of the particle radius, checked as a distribution at the end of the run rather than assumed at the start. The reduction factor is a documented modelling decision, not a setting buried in the input deck.

Every calibration has an envelope

A calibration is valid for the size distribution, moisture, consolidation range and wall material it was fitted against. The same ore at 8 % moisture is not the same material as at 2.5 %: cohesion moves considerably more than the tolerance of the fit. Winter and summer, ROM and crushed, a steel liner and a UHMW one are different cases. Either you recalibrate, or you state the extrapolation and the risk that comes with it.

That is why the deliverable carries the fitted parameter table, the tests with their conditions, the residual on each response, the validation result and — explicitly — what was not calibrated.

Without that backing the material still moves and the image still looks plausible. What stays a free parameter is everything the model was meant to decide: the forces on the liner, the wear rate, the flow rate and the segregation.