Three Coordinates Sirui Croma: How a Bridge CMM Turns Points Into Verdicts

• 1
  What a three coordinates machine actually does
• 2
  Alignment decides whether the numbers mean anything
• 3
  Where the measurement uncertainty comes from
• 4
  How many points, and where to put them
• 5
  Fitted values, outliers, and what a report hides
• 6
  Where a CMM fits in a CNC shop
• 7
  Three coordinates sirui croma versus other gauges

A three coordinates Sirui Croma machine is a bridge-type coordinate measuring machine. A granite table carries the part. A bridge moves in X and Y above it, and a vertical ram moves in Z. At the end of the ram sits a touch probe with a ruby sphere, usually 1 mm to 5 mm in diameter.

When the sphere touches the surface, the machine reads the scale position of all three axes at that instant. One touch gives one point in space. The controller logs X, Y and Z to three decimal places or better, and the software keeps the list.

The software then fits geometry to that list. Three points define a plane. Four or more points let the software average out surface roughness and give a more stable plane. A cylinder needs a minimum of six points, and a real bore is usually sampled with 12 to 40 points around two or three levels.

That is the whole idea. The machine does not measure a diameter directly. It measures points, then computes the diameter from a fitted cylinder axis. Every number on a CMM report is a fitted number, not a raw reading.

• 1
  Points first, features secondThe report shows derived values, not the raw hits.
• 2
  Probe tip mattersA 1 mm ruby on a thin stylus deflects more than a 5 mm one.
• 3
  Granite baseThermally stable, but it still drifts with the room.

A CMM has no idea where the part is. You have to tell it. That process is alignment, and it is the step that most often goes wrong in a first article inspection.

The usual approach is a 3-2-1 alignment on a prismatic part. Three points on the primary datum plane lock Z and two rotations. Two points on the secondary datum lock the remaining rotation. One point on the tertiary datum sets the origin in the last direction. Now the software knows the part coordinate system.

For parts with no flat faces, a best-fit alignment is used instead. The software moves the measured point cloud until the sum of the deviations is as small as possible. This works well on a casting or a curved housing. It also hides things. A best-fit alignment distributes error everywhere, so a bore that is 0.03 mm off-centre may look acceptable.

On a drawing with datum callouts, always align to the datums. If the drawing is GD&T controlled, the alignment must match the datum reference frame. Otherwise the report is internally consistent and externally wrong.

• 1
  3-2-1 for prismatic partsPlane, line, point. Fast and repeatable.
• 2
  Best-fit for freeformGood for castings, but it spreads error.
• 3
  Follow the datum frameNever align to whatever face is convenient.

Machine accuracy is quoted as MPE, maximum permissible error, usually written as a formula such as 2.5 + L/300 μm, where L is the measured length in millimetres. At 100 mm that is about 2.8 μm. At 1,000 mm it is about 5.8 μm.

Temperature changes that number. Most CMMs are specified at 20 °C. A shop floor at 28 °C with a warm part can add 5 μm to 15 μm of apparent error on aluminium, because the part and the scale grow at different rates. A controlled room at 20 ± 2 °C removes most of it.

Probe and stylus geometry add more. A long stylus bends under its own contact force. The bending is repeatable, so it can be calibrated out, but only for the direction it was calibrated in. A stylus that is 100 mm long may lose several micrometres of accuracy and much of its stiffness.

Surface finish matters too. A turned surface with Ra 3.2 μm scatters probe hits. A ground surface at Ra 0.4 μm gives a much tighter plane fit. On rough surfaces, take more points and expect a wider spread.

• 1
  MPE is length dependentShort features measure better than long ones.
• 2
  20 °C is the referenceEvery degree away costs you, especially on aluminium.
• 3
  Stylus length is a trade-offLonger reaches deeper, but bends more.

Point count is a decision, not a default. The minimum geometric definition of a plane is three points. That gives no redundancy. One bad hit on a burr tilts the whole plane, and a plane fit with almost no redundancy has nothing to check itself against. The software will still print a flatness value, and that value will be meaningless.

For a production plane, use 9 to 25 points spread across the full face. For a bore, use at least two levels with 8 points each. For a roundness check, four levels with 12 points is a reasonable starting point.

Probing strategy also matters on the contact side. A touch trigger probe fires when the stylus deflects, so it reads slightly past the true surface. That pre-travel is calibrated, but it varies with approach speed. Approach at a constant 2 mm/s to 5 mm/s and keep it the same across the program.

Scanning probes behave differently. They drag the stylus along the surface and take thousands of points per second, which gives much better form data for roundness and cylindricity. On a bridge machine, scanning is slower than touch probing but far more informative.

• 1
  Minimum is not a strategyThree points per plane tells you almost nothing.
• 2
  Spread points wideClustering them in the middle hides tilt.
• 3
  Constant approach speedPre-travel drift shows up as scatter.

A CMM report gives one number per feature. Behind it is a point cloud and a fit. Least-squares fitting is the default. It minimises the sum of squared deviations, which means one outlier can pull the result.

Consider a bore with 24 points. If a chip sits on the wall and the probe hits it, that single point can shift the fitted centre by several micrometres. The report shows a diameter and a position. It does not show that one point was 0.05 mm off the trend.

That is why filtering matters. Most software offers outlier removal, typically discarding points beyond two or three standard deviations. It works, but it can also hide a real defect. If the outlier is a burr, filter it and note it. If the outlier is a dent, the part is bad and the filter is lying to you.

For position tolerances, the software must also apply the maximum material condition bonus when the drawing calls for it. That bonus is calculated from the actual size of the feature. Get the feature size wrong and the bonus is wrong, and a borderline part passes.

• 1
  Least-squares is defaultSensitive to single bad points.
• 2
  Filter with a reasonLog every removed point.
• 3
  MMC bonus needs sizeWrong diameter, wrong bonus.

A bridge CMM is slow. A 30-point inspection on a simple bracket takes a few minutes of probing plus setup and alignment. That is fine for first article inspection, for a process audit, or for a customer who asks for dimensional reports. It is not fine for checking 5,000 parts one at a time.

In our own workflow, the CMM sits at the front and the back. On the front end it verifies the first part off a new program, on a part that may be up to 4,000 mm long. On the back end it confirms critical features before shipment, alongside the 100% inspection step we run on every order.

For volume work, the CMM validates the gauge. If a go/no-go gauge and the CMM agree on the same parts, the gauge can run production. That is the useful division of labour. The CMM proves the gauge, the gauge proves the parts.

The trade-off is clear. A CMM gives you numbers you can put on a report and trace back to datums. It costs time per part and needs a controlled room to hold its accuracy. For a shop-floor check of a single diameter, a micrometer is faster and just as good.

• 1
  FAI and auditsWhere a full dimensional report is required.
• 2
  Gauge validationCMM proves the gauge, gauge proves the run.
• 3
  Not for 100% high volumeToo slow per part on simple features.