CNC emulator: what it proves before the first cut

• 1
  What a CNC emulator actually computes
• 2
  Why the simulation matters on 5-axis parts
• 3
  When to trust the simulation and when to cut air
• 4
  What the model needs to be worth anything
• 5
  How we sequence verification at GreatLight
• 6
  Simulation versus dry run versus first article

A CNC emulator is software that reads a toolpath and applies the same logic a machine control would apply: feed moves, rapid moves, tool changes, work offset shifts, cutter compensation and canned cycles. The output is not a picture of a finished part. It is a record of whether the machine could physically follow the commanded path without hitting itself or the fixture.

Two engines usually sit inside one package. The first is the interpreter, which parses the G-code and flags syntax errors and modal-state problems. The second is the geometric engine, which sweeps the tool, holder and spindle nose through the stock and fixture volume. A good emulator reports the exact block number where a problem starts. That is what makes a crash reproducible on screen instead of on the machine.

Machine kinematics is the part most people underestimate. A 3-axis mill has three linear axes, so the swept volume is easy to compute. A 5-axis machine adds two rotary axes, and the tool tip position depends on the pivot distance, the trunnion offset and the order in which the control applies the rotations. If those numbers are wrong in the simulation model, the virtual machine will look correct while the real one cuts in the wrong place.

Simulation also handles the non-geometric side: spindle warm-up blocks, tool length offsets, coolant commands and M-code sequencing. Some of these matter more than geometry on a first run. A program that rapids to Z 50 mm before the tool length offset is active will drive the holder into the stock no matter how clean the toolpath looks.

The boundary is simple. The emulator knows what you told it. It does not know that the vise jaw was re-clamped 0.1 mm off the datum since the last setup, or that a 12 mm end mill has 0.03 mm of runout. Those are shop-floor facts, not simulation inputs.

Complex geometry is where a CNC emulator earns its keep. On a 5-axis job, the tool and holder tilt through the part envelope while the table rotates. A holder that clears the stock at one tilt angle can clip a fixture at another. Static checking in CAM does not catch that. Simulation does, because it steps through the motion in time order.

Undercut and reach problems show up early. If a tool needs 62 mm of reach to hit a pocket floor and the holder body is 40 mm in diameter, the simulation will show the shank rubbing the wall. You can then decide between a longer tool with more chatter risk, a different tool axis, or a design change. Each of those decisions is cheap on screen and expensive after the first scrapped part.

Setup and re-fixturing is the second payoff. On a mill-turn center or a 4-axis job with multiple work offsets, the emulator verifies that each setup uses the correct datum and that the rapids between operations stay clear. We run the full program, not just the finishing pass, because most collisions happen during positioning moves.

Training is the third. New operators can run a program end to end, watch the toolpath, and see what a wrong offset does to the part. No spindle time, no scrap. That shortens the learning curve on our 16 simultaneous 5-axis machining centers and the 12 four-axis mills.

The limits are real, though. A CNC emulator does not predict chatter, tool deflection, thermal growth or surface finish. It tells you the path is possible, not that the cut will be good.

Trust the simulation when the risk is geometric and the inputs are current. New programs, first runs, tilted 5-axis toolpaths, deep pockets, long holders, and any job where a crash would damage a spindle or a fixture. In those cases the emulator is faster than dry running, and it shows the failure without a repair bill.

Cut air on the machine when the risk is dynamic or the model is stale. If the fixture was modified since the last simulation, or the tool stick-out changed, the virtual model no longer matches reality. A dry run with the Z offset raised by 50 mm verifies the same motion with the real control, the real offsets and the real kinematics.

For one-off prototypes, simulation plus a single dry run is usually enough. For a 10,000-part run, the cost of a 20-minute verification is negligible against the cost of stopping the line. The economics favor more checking as volume rises, not less.

A practical rule we use: any program that touches a rotary axis gets simulated first, and any program with a tool change inside a pocket gets simulated and dry run. Everything else gets a standard first-article check with the probe.

Neither method replaces the other. Simulation catches logic and interference. The dry run catches what the model got wrong. The first article catches what the process got wrong.

The simulation is only as good as its machine model. You need the actual travels, not the catalog numbers. Our large travel is 4,000 × 400 × 150 mm, the medium group runs 750 × 1,150 × 550 mm and 600 × 600 × 600 mm, and the compact group is 500 × 500 × 450 mm and 500 × 310 × 200 mm. A model built for the wrong travel group will miss over-travel.

Tool data is the second input. Real stick-out, real holder geometry, real flute length. A 6 mm end mill in a shrink-fit holder behaves differently from the same cutter in an ER collet chuck, because the nut diameter changes the clearance envelope. Put the actual assembly in the library, not a generic block.

Fixture and stock models are the third. A vise, soft jaws, a vacuum plate, a tombstone. The stock model should match the blank you will actually load, including the saw-cut face and any pre-machined datum. If the blank is 2 mm thicker than the model, every clearance number is wrong by 2 mm.

Control-specific behavior is the fourth. Fanuc, Siemens and Heidenhain handle plane selection, rotary axis clamping and canned cycles differently. A post-processor that emits code for the wrong control will simulate fine and alarm on the machine, or worse, run the wrong cycle.

Finally, keep the model versioned. When a machine is re-calibrated or a fixture is replaced, the simulation file changes with it. An out-of-date model is worse than no model, because it gives false confidence.

We start with DFM. Quotation and free DFM analysis come back within 12 hours, and that review already flags features that will be hard to hold. The engineer notes which setups need simulation before the program is even written.

Programming comes next, then simulation against the machine model for the specific machine that will run the job. Five-axis work goes through the full motion check. Four-axis and 3-axis work goes through interference and offset checks. If the simulation flags a collision, the fix happens in CAM, not on the floor.

Then a dry run on the machine with the work offset raised, followed by a first article. We inspect 100% of parts before shipment, with raw material checks, in-process monitoring and final inspection. Reports are available on request. Production can start within 24 hours, and parts ship in 3–5 days.

Tolerance targets drive how much verification is warranted. A bracket at ±0.1 mm needs less than an optical mount at ±0.005 mm (±0.0002 in). The tighter the tolerance, the more the simulation and the metrology loop matter, because a small kinematic error in the model becomes a real dimensional error in the part.

We hold ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022. Uploads are secure and confidential, and an NDA is available on request.