Three Coordinate Measurement Machine Simulation in a Remote Measurement System
This page explains how a CMM simulation environment is built, what it actually predicts, and where it stops being useful. It is written for quality engineers and manufacturing engineers who need to approve a remote measurement setup before it touches a real part. After reading it you can judge whether a simulation is trustworthy for a given job, and what still has to be verified on the real machine.
How a Remote Measurement System Is Put Together
A remote measurement system has three parts: the local end where the operator sits, the remote end where the CMM stands, and the network between them. The operator issues commands on the local end. Those commands travel to the remote end, where sensors and actuators drive the machine and report position back. That is the whole architecture, and it is worth keeping in mind because every failure mode traces back to one of the three parts.
The local end is not just a screen. It holds the measurement plan, the alignment data, and the nominal CAD model. If the local end and the remote end disagree about which coordinate system is active, every reported value is wrong by a constant offset. We have seen this cost a full day of re-measurement on a batch of 40 aluminium housings.
The remote end is a three coordinate measurement machine with its own error budget. Squareness, straightness, and scale error all live here. A simulation can model them, but the model is only as good as the calibration file you feed it. Feed it a stale calibration and the simulation will confidently predict the wrong number.
The network is the part most people underestimate. Typical industrial links run 20–100 ms round trip. That is fine for a slow touch probe moving at 5 mm/s, but it is not fine for a scanning probe collecting thousands of points per second. For scanning work, keep the point cloud generation on the remote end and send only the results.
- 1Local endPlan, alignment, nominal model, operator decisions.
- 2Remote endMachine, probe, controller, local point capture.
- 3NetworkCommand latency and data volume. Keep raw data local.
Geometric Modeling: Building the Virtual CMM
Geometric modeling is the step where you create a 3D representation of the machine itself, not the part. The model contains the granite table, the bridge or gantry, the Z ram, the probe head, and the stylus. Each body gets a local coordinate frame and a set of nominal dimensions. Get the stylus length wrong by 2 mm and the simulation will still run, but the reported probe center will be offset from the real contact point.
The model is usually built from the machine drawing plus a laser tracker or ballbar survey. A 1,200 mm bridge machine might be modeled to within 0.02 mm of nominal geometry. That sounds tight, but the machine's own volumetric error is often ±0.005 mm or better on a good day, so the model geometry is not the limiting factor. The limiting factor is how you map the model to the real machine's error map.
Rendering libraries such as OpenGL handle the drawing. They give you fast 3D display, not accuracy. A common mistake is to treat the visual match as proof of correctness. A simulation can look perfect on screen and still be wrong by 0.1 mm because a joint frame was defined at the wrong end of the ram.
Practical check: load a known master artifact, a step gauge or ball plate, into the simulation and compare the virtual result against its certified values. If the virtual result is off by more than 0.01 mm on a 500 mm artifact, fix the model before trusting anything else.
- 1Model the machine, not the partTable, bridge, ram, head, stylus each get a frame.
- 2Verify with a master artifactStep gauge or ball plate with certified values.
- 3Rendering is not accuracyOpenGL draws pixels; it does not validate geometry.
Kinematic Analysis: Probe, Boom, and Rotary Joints
Kinematic analysis studies how the machine's components move and where they end up. For a bridge CMM that means three linear axes and a probe head that may index in two rotations. For an articulated arm it means six or seven joints in series. The math is a chain of transforms. Each transform has a nominal value and an error term.
The output you care about is the position and orientation of the probe tip in part coordinates. Everything else, motor speed, acceleration profile, joint torque, is secondary for measurement accuracy. If the kinematic chain is correct, you can predict where the tip will be for any command, and you can predict whether a given feature is reachable without collision.
Reachability matters more than most people expect. A 500 × 500 × 450 mm work envelope does not mean every point in that box can be touched from a usable probe angle. Deep bores and undercuts force the head to index, and each index adds a small angular error. Simulating the index positions ahead of time lets you choose an approach that keeps the head at fewer index steps.
Temperature is the quiet variable. A steel part grows about 11 μm per metre per °C. A 500 mm aluminium part in a 2 °C drift moves roughly 23 μm. The simulation can add a thermal compensation layer, but only if the part and machine temperature sensors are fed into it in real time. Without that, the model is describing a machine at 20 °C that may be running at 23 °C.
- 1Chain of transformsEach axis and joint contributes position and error.
- 2Check reachabilityEnvelope size is not the same as accessible angles.
- 3Thermal layerNeeds live temperature data to be meaningful.
- 4Probe tip is the outputEverything else is intermediate.
Measurement Plan and Probe Path Simulation
Before running a plan on the real machine, the simulation executes the full sequence: datum setup, alignment, feature measurement, and reporting. It checks that the probe can reach each feature, that no collision occurs along the path, and that the number of touch points is enough for the tolerance you are claiming.
Point count is a real engineering decision. A circle measured with 4 points gives a diameter that is sensitive to form error. The same circle with 16 points and a least-squares fit gives a much more stable result. For a bore held to ±0.005 mm, we typically plan 12–16 points per circle and 3 circles per bore. The simulation tells you the total cycle time before you commit to it.
Speed matters for throughput but hurts accuracy on a touch probe. Approach and retract at 10 mm/s is normal. Touch speed is usually 1–3 mm/s. If the simulation shows a cycle time that looks too good, check the touch speed setting first. It is the most common place where a plan gets quietly optimized past the point of usefulness.
The simulation also flags stylus bending. A long stylus on a small tip deflects under contact force. A 50 mm stylus with a 2 mm ruby tip can deflect 1–2 μm at 0.1 N. On a ±0.005 mm tolerance that is a real fraction of the budget. Simulate with the actual stylus, not a generic one.
- 1Plan the full sequenceDatum, alignment, features, report.
- 2Point count drives stability12–16 points per circle for tight bores.
- 3Watch touch speed1–3 mm/s. Faster numbers usually mean a hidden error.
- 4Model the real stylusLength and tip diameter affect deflection.
What Simulation Cannot Tell You
A simulation is a model. It predicts the machine, not the part. It cannot tell you that the fixture moved 30 μm during clamping, that a chip sat under a locating pad, or that the operator wiped a surface with the wrong cloth. Those are process errors, and they live outside the model.
It also cannot validate the probe calibration. Ruby wear, shank contamination, and a loose stylus are physical conditions. The simulation assumes a clean, correctly calibrated probe. If the real probe has 5 μm of wear on one side, the simulation will still report a perfect result.
Environmental effects beyond temperature are usually ignored. Floor vibration, air currents from a nearby door, and acoustic noise all show up in real data. A simulation run in a quiet office will not predict a machine sitting next to a stamping press.
So the honest boundary is this: simulation is good for reachability, collision, cycle time, plan completeness, and gross error checking. It is not a substitute for a first-article inspection on the real machine. Use it to avoid wasting machine hours, then confirm with a certified artifact.
- 1Process errors are invisibleFixture shift, chips, handling.
- 2Probe condition is assumed goodWear and contamination are physical only.
- 3Environment is simplifiedVibration and air currents are often ignored.
- 4Always confirm on the machineSimulation narrows the risk. It does not remove it.
When Simulation Is Enough and When It Is Not
Match the task to the right verification method.
| Task | Simulation is enough | Verify on the real machine |
|---|---|---|
| Probe reachability and collision check | Yes, standard use | Not needed unless geometry changed |
| Cycle time estimate for quoting | Yes, within about 15% | Only for high-volume runs |
| Plan completeness and point count | Yes, good first pass | Confirm on first article |
| Feature size held to ±0.005 mm | No, not sufficient alone | Required, with certified artifact |
| Stylus deflection under contact force | Partly, if stylus is modeled | Confirm with known step gauge |
| Thermal drift during a long run | No, needs live sensor data | Log temperature and re-check |
| Fixture and clamping effects | No | Required, inspect after clamp |
The Practical Takeaway
If you need to prove reachability, catch collisions, and estimate cycle time before booking machine hours, three coordinate measurement machine simulation is the right tool and it will save you real time. If you need to certify a tolerance near ±0.005 mm on a production part, simulation only narrows the risk. The release decision still comes from a calibrated artifact on the real machine.
Questions Engineers Ask
How accurate is a CMM simulation compared with the real machine?
For geometry and reachability, a well-built model matches the real machine to about 0.02 mm. That is enough to catch collisions and confirm probe access.
For measured values, the simulation is only as good as the error map and calibration data you load into it. Without a current calibration file, treat the predicted values as a sanity check, not a result.
Can we run the simulation over a normal internet connection?
Yes, for touch-probe work. Command latency of 20–100 ms is acceptable when the probe moves at 1–10 mm/s and only a few thousand points are collected.
For scanning or high-density point capture, keep the point cloud generation on the remote end and transfer only the fitted results. Sending raw scan data over a shared link will slow the whole plan.
What software is typically used to build the simulation environment?
Most environments use a 3D rendering library such as OpenGL for the display layer, plus a separate geometry and kinematics engine. The rendering library draws the machine; it does not compute accuracy.
Some systems build the whole environment around OpenGL calls and keep the kinematic chain in the same codebase. That is simpler to maintain but harder to validate, so artifact-based verification becomes more important.
Does the simulation account for temperature changes during a long measurement run?
Only if you feed it live temperature data. A static model assumes one temperature for the whole run.
On a 500 mm aluminium part, a 2 °C drift moves the material about 23 μm. That is larger than many tolerances, so for long runs we log part and machine temperature and apply compensation separately.
How many touch points should a plan use for a tight bore?
For a bore held to ±0.005 mm, plan 12–16 points per circle and measure at least 3 circles at different heights. That gives a least-squares fit that is stable against local form error.
Four-point circles are fine for rough location, but the diameter they report swings with every small bump on the surface.
Can simulation replace first-article inspection?
No. Simulation cannot see fixture shift, chips under a locating pad, or probe wear. Those are physical conditions that only show up on the machine.
Use the simulation to prepare and de-risk the plan, then run a first article on a certified artifact before releasing the plan for production.
Need Parts Measured to ±0.005 mm?
Send us your drawings and tolerance callouts. We review the measurement plan, confirm what can be verified, and quote within 12 hours. Uploads stay confidential, and an NDA is available on request.
12-hour quote100% inspectionISO 9001 / IATF 16949NDA on request
More from GreatLight
We publish setup notes, tooling trials and inspection data from the factory floor.
