Mould Flow Analysis: What the Simulation Gets Right and What It Misses

What a simulation is actually for

A filling simulation solves the flow of a polymer melt through a cavity, and it produces a picture of where the melt is at every moment, where it last fills, what pressure it took, and where it is hot. It is a design tool, and like any design tool its value is in the questions it lets you ask before the steel is cut.

The questions it answers well are the geometric ones. Where is the last point to fill. Is there a flow front that will trap air. Is the runner balanced. Does the part fill at a pressure the press can deliver. Those are all consequences of geometry and they are robust to the details of the material model.

The questions it answers badly are the ones that depend on things the model does not know: the actual melt temperature, the actual tool temperature distribution, the actual gate freeze time, and the actual behaviour of the compound at the shear rates in a real gate. Those are where a simulation and a mould diverge.

So the useful posture is to trust the shape and check the numbers. A simulation that says the last point to fill is in the corner is probably right. A simulation that says the pressure is 82 MPa is probably right to within 15 percent, and 15 percent on a pressure that is near the press limit is the difference between a tool that runs and one that does not.

Where it is reliable: fill pattern and fill time

Fill time prediction is the strongest output, because it depends mostly on geometry and volume. Measured against eight production tools, predicted fill times tracked measured fill times closely across a range from 0.6 to 2.6 seconds, with the points sitting near the diagonal.

That matters because fill time is what decides the balance question. If a simulation says cavity 7 fills 0.2 seconds after cavity 1, that is a real result and it is worth acting on. The absolute value may be 5 percent off; the difference between cavities is not.

Fill pattern is similarly reliable. The last point to fill, the location of a weld line, the place where two flow fronts meet and trap air: those are geometric and the simulation gets them right, which is why it is so useful for finding a venting problem before the tool is cut.

What it does not get right about the fill is the fine detail near the gate. The flow in the first millimetre after the gate is dominated by viscoelastic effects and by the actual gate geometry, and a simulation that uses a 1D gate model will not predict jetting or a gate blush.

Mesh size decides the pressure answer

A simulation divides the part into elements and solves the flow in each one. The element size is the mesh size, and a finer mesh resolves the flow better and costs more computing time. The relationship between mesh size and pressure error is steep.

At a 0.5 mm mesh the pressure error is about 3.1 percent. At 1.0 mm it is 4.4, at 1.5 mm 6.2, at 2.0 mm 8.8, at 3.0 mm 14.6 and at 4.0 mm 22.4. A mesh that is eight times coarser gives seven times the error.

The error is not random. A coarse mesh over-predicts pressure, because it smooths out the thin sections that would have carried high shear and it underestimates the flow resistance in the details. So a coarse simulation is conservative on pressure, which is the safe direction, and it is still wrong by enough to mislead.

The practical rule is a mesh with at least three elements across the thinnest section. On a 1.5 mm wall that is a 0.5 mm mesh, which is exactly the size that gives a 3 percent error. A simulation run on a 2 mm mesh for a 1.5 mm wall has less than one element through the thickness and its pressure number is a guess.

The other mesh issue is the runner. A runner modelled with beam elements is fast and inaccurate at the branch; a runner modelled with 3D elements is slow and accurate. On a multi-cavity tool the runner is where the balance is decided, so a beam model is the wrong choice for the question being asked.

Table 1 — Mesh size against the number it produces

What it misses

A simulation does not know the tool temperature. It assumes a constant cavity wall temperature, usually 40 to 80 C, and a real tool has a temperature map that varies by 15 to 25 C across a cavity. That map is what decides where the part warps, and it is the input the simulation is least able to supply.

It does not know the actual melt temperature. It is given one, and the real melt is 10 to 20 C hotter in the middle of the channel than the barrel setpoint suggests. A simulation run at the setpoint will under-predict the fill and over-predict the pressure.

It does not model the packing phase well. Packing is a compressible, transient problem with a moving boundary at the gate, and most filling simulations reduce it to a pressure hold at a point. The result is that sink marks and warpage are predicted qualitatively at best.

It does not know the compound. A generic grade from a database is not the lot in the machine. A 15 percent difference in viscosity between two lots of the same grade is normal, and it moves the pressure by more than the mesh error does.

It does not know the tool. Gate freeze time depends on the actual gate geometry and the actual cooling around it, and a simulation that assumes a freeze time will predict a packing profile that the tool does not deliver.

Using one without being misled

Ask the geometric questions. Where does it fill last, is the runner balanced, is there an air trap, does it fill at all. Those answers are reliable and they are worth the run.

Do not accept a pressure number without knowing the mesh. A pressure prediction on a coarse mesh is a number with a 20 percent error and no error bar, and it will be quoted in a meeting as if it were precise.

Run the sensitivity yourself. Change the melt temperature by 10 C and the tool temperature by 10 C and see how much the answer moves. If the answer moves more than the mesh error, the material inputs are the problem, not the mesh.

Use the simulation to design the experiment. A simulation that says the last point to fill is in the corner tells you where to look on the first shot, and that is worth more than the pressure number.

Table 2 — Which outputs to act on

Where it fits in a tool project

The useful point to run a simulation is after the part design is frozen and before the tool layout is released. At that point the geometry is known, the alternatives are still open, and a wrong answer costs a day rather than a tool.

Running it earlier, on a part that is still changing, produces an answer to a question that will not be asked. Running it later, on a tool that is already cut, produces a confirmation of a decision that has been made.

The other useful point is after the first trial, when the measured fill pattern can be compared with the predicted one. That comparison is how the material inputs get calibrated, and a calibrated model is worth ten uncalibrated ones.