Fixing Portable CMM Probe Compensation with Pulled Hits

Problem Details

Two points are measured on one part and the point-to-point distance comes out wrong by roughly one probe ball diameter. The first point is taken with the arm's last segment vertical, probing straight down onto an accessible face — that result checks out. The second point is on a surface that is under the part, so the operator swings the arm over and contacts the surface with the side of the ball, with the probe shaft lying roughly parallel to the surface. Probe compensation is enabled and calibration is current, yet the reported distance is offset by the ball diameter.

The symptom is not a calibration fault and not a software bug. It is a vector problem: the software knows the ball center coordinates and the ball radius, but on a single contact it does not know which direction to shift the center to get the surface point.

Root Cause: Where the Vector Comes From

On a bridge/DCC CMM the approach vector is known because the machine drives the probe along a programmed direction, and a touch-trigger probe latches at contact along that move. On a portable articulating arm the operator can arrive at the surface from any direction, so there is no motion vector to trust. Manual/portable measurement software therefore falls back on a substitute:

• Assumed vector from the arm's last-segment axis (probe shaft direction) — this is the default behavior that makes straight-down hits work correctly.
• Vector supplied by a feature fit (plane, circle, cylinder) built from several hits.
• Vector captured explicitly by a pulled hit.

When the operator contacts with the side of the ball and the probe shaft points into the material rather than away from it, the assumed vector is rotated relative to the true surface normal — in the worst case a full 180°.

Error magnitude

For an assumed vector misaligned by angle θ from the true surface normal, the component of error along the measurement direction is:

e = R × (1 − cosθ)

θ = 0°   -> e = 0            (correct hit)
θ = 30°  -> e = 0.134 × R
θ = 90°  -> e = R            (compensation lost entirely)
θ = 180° -> e = 2R = D       (full ball diameter, the reported symptom)

A 6 mm ball gives up to 6 mm of error; a 3 mm ball gives 3 mm. That matches the classic complaint that the measurement "adds in the probe diameter."

Second-order effects on the same distance

Renishaw's probe operation notes make the pre-travel point explicitly: a trigger stylus deflects a finite distance after contact before the signal fires, so the electronic ball radius stored by qualification is not the nominal physical ball radius, and substituting the catalogue ball diameter builds pre-travel error into every dimension (Renishaw probe operation knowledge base). Their calibration paper adds that the stored radius is only valid for the feedrate, orientation and approach directions used during calibration (Renishaw TE415 probe calibration, PDF).

Solution 1: Orient the Arm So the Shaft Points Outward

The simplest fix costs nothing and requires no new technique. Position the last arm segment so the probe shaft points outward, away from the material, approximately along the surface normal, and contact with the tip of the ball.

1. Identify the outward normal of the surface to be probed (the direction air is, not metal).
2. Rotate the arm wrist so the stylus axis lies within roughly ±30° of that normal (keeps residual error ≤ 0.134 · R).
3. Take the hit with the leading face of the ball, not the side.
4. Repeat for both points of the distance so both are compensated consistently.

This is why the first point in the reported case is correct: probing straight down on an upward-facing surface puts the shaft exactly on the outward normal. It fails on the under-part surface only because the table or the part body blocks the required arm posture.

Solution 2: Pulled Hits for Blocked Access

When the required posture is physically impossible — the feature is under the part, close to the table, or inside a pocket — use a pulled hit. A pulled hit lets you contact the surface from any convenient direction and then declare the compensation vector by the direction you withdraw the probe.

1. Enable the pulled-hit / vector-hit mode in the measurement software before taking the point (check the exact command name and hotkey in your software's help; it is a distinct hit type, not a setting on a normal point).
2. Bring the ball into contact with the surface using whatever arm posture is reachable, including side-of-ball contact.
3. Hold contact and take the hit.
4. Pull the probe straight away from the surface along the true outward normal before releasing. The software derives the compensation vector from this withdrawal path.
5. Confirm the reported point vector before continuing. It should read close to the expected surface normal (for example, a downward-facing surface in a Z-up alignment gives roughly i=0, j=0, k=-1).

Solution 3: Measure Features, Not Bare Points

If operators are "quick trigger, fire and forget" and will not reliably perform pulled hits, remove the vector decision from them entirely:

• Measure a plane (3+ well-spread hits) on each of the two surfaces and dimension plane-to-plane. The fit derives its own normal, and each hit is compensated along that normal instead of along the arm posture.
• Measure circles/cylinders/spheres instead of individual points where geometry allows; the fit resolves comp direction radially.
• Build a guided routine with nominal vectors pre-loaded on each point so the software applies the programmed normal rather than an inferred one.
• On a surface too small for a plane, probe a larger co-planar surface and offset by the known nominal step.

Verification

Prove the fix before releasing parts, using artifacts of known size:

1. Ball-bar or gauge block check. Measure a known length once with good posture and once with side-of-ball contact and no pulled hit. The delta should be close to the ball diameter — that confirms the diagnosis rather than a calibration fault.
2. Repeat with pulled hits. The same length should now agree with the certified value within the arm's stated volumetric accuracy.
3. Vector audit. Open each measured point and read its I/J/K vector. Any point whose vector does not match the physical surface normal is suspect.
4. Two-sided artifact. Measure a ring gauge or a boss of known diameter. A size error equal to twice a radius delta on both bore and boss points to a stale or mis-qualified probe radius, not to a vector problem — re-qualify against the reference sphere.
5. Sign test. Reverse the pull direction deliberately on one hit. The distance should change by 2R. If it does not, the software is not using the pulled vector and the hit type is wrong.

Calibration and Probe Hygiene

• Qualify every stylus configuration you will use — different ball diameters, extensions and wrist adapters each get their own qualification record.
• Re-qualify after any stylus change, collision, or when the arm has been transported.
• For touch-trigger setups, qualify with the same feedrate and approach directions used in measurement; changing them invalidates the stored effective radius. Modern controls that recalculate the effective radius per move and auto-select feedrate remove the classic fast-vs-slow touch discrepancy (Renishaw SupaTouch).
• Log the ball radius used in the current routine and compare it against the artifact result during the verification step above.

Operator Training Checklist