Coordinate measuring machines and machine vision inspection

How coordinate measuring machines (configurations, probes, alignment, feature fitting, errors) and machine vision systems (lighting, optics, sensors, processing) inspect parts, with coordinate-geometry, thermal and pixel-resolution numericals.

Drafted with Aria, reviewed by the AiCanCode.org team. Spotted an error? Use Give Feedback at the bottom of the page.

Why it matters

Modern parts carry dozens of related features (hole patterns, bores, freeform surfaces, GD&T callouts) that hand instruments cannot check efficiently. A coordinate measuring machine (CMM) measures them all in one set-up against the CAD model, and machine vision inspects every part on a fast line without touching it. Both are core mechatronic systems: precision mechanics, sensors, motion control and software working together.

Key ideas

What a CMM does A CMM moves a probe along three orthogonal axes, each with a linear scale (typically 0.1–1 μm resolution). Every probe contact records the (x, y, z) coordinates of a point. Software fits geometric features (planes, circles, cylinders, cones, spheres) to those points and then evaluates sizes, distances, angles and GD&T tolerances. The CMM therefore measures points, and everything else is computed.

Structural configurations

  • Moving bridge: the most common; good accuracy and access for medium parts.
  • Fixed bridge (moving table): stiffer, highest accuracy, smaller parts.
  • Cantilever: open on three sides for easy loading; lower stiffness, small parts.
  • Horizontal arm: for large sheet-metal parts such as car bodies; lower accuracy.
  • Gantry: the bridge runs on elevated rails; for very large, heavy parts (aircraft, large castings).
  • Portable articulated arms and laser trackers: taken to the part on the shop floor, at lower accuracy.
  • Granite tables and air bearings give thermal stability and low friction.

Probes

  • Touch-trigger probe: a kinematic seat (three rods on six balls) carrying the stylus. On contact the seat opens, the circuit breaks, and the scale readings are latched. Simple and robust; collects discrete points.
  • Scanning (analogue) probe: measures stylus deflection continuously, so it can collect thousands of points while sliding along a surface; best for form (roundness, profile).
  • Non-contact probes: laser line scanners, vision cameras, white-light and confocal sensors; for soft, thin or delicate parts and fast surface capture.
  • Probe tip compensation: the machine records the centre of the stylus ball, so software offsets each point by the ball radius along the surface normal. Stylus qualification on a reference sphere determines the effective ball radius before measurement.

Measuring workflow

  1. Qualify the stylus on the reference sphere.
  2. Establish the part coordinate system (alignment), usually 3-2-1: a plane from at least 3 points fixes the primary datum, a line from 2 points on the secondary face fixes rotation, and 1 point fixes the origin on the tertiary face.
  3. Measure features with enough points: line 2, plane 3, circle 3, sphere 4, cylinder 5, cone 6 at the mathematical minimum. Use more points in practice, because with the minimum the fit has no redundancy and form error cannot be assessed.
  4. Fit features by least squares (Gaussian) or by minimum-zone, maximum-inscribed or minimum-circumscribed criteria, and evaluate dimensions and GD&T.
  5. Report against tolerances, often by comparison with the CAD model.

Error sources Scale and geometry errors of the machine (21 parametric errors of a three-axis machine, mapped and compensated in software), probe lobing, stylus bending, temperature (reference 20 °C), dirt, part clamping distortion, and too few points or a poor alignment. CMM performance is stated as a maximum permissible error, commonly in the form E = A + L/K μm, verified to ISO 10360.

Machine vision inspection Components, in signal order:

  • Lighting: the most important design decision. Backlighting gives a sharp silhouette for dimensions; diffuse dome lighting suppresses glare on shiny parts; low-angle dark-field lighting highlights scratches and embossed text; coaxial lighting for flat reflective surfaces; structured light (projected lines) for 3D height.
  • Optics: lens focal length sets the field of view (FOV) and working distance. Telecentric lenses keep magnification constant with depth, so they are used for gauging.
  • Sensor (camera): CCD or CMOS area or line-scan sensor; resolution in pixels, frame rate, exposure.
  • Processing: acquisition, pre-processing (filtering, contrast), segmentation (thresholding, edge detection), feature extraction (blob area, centroid, edge positions, pattern match), then decision (pass/fail, measurement, classification, increasingly by trained neural networks).
  • Communication: results go to a PLC or robot over digital I/O or industrial Ethernet to reject parts or guide a robot.
  • Applications: presence/absence checks, dimensional gauging, surface defect detection, OCR and barcode reading, and robot guidance (pick and place, bin picking).
  • Calibration: pixel size is found by imaging a calibration target; lens distortion is corrected with a grid. Sub-pixel edge detection can resolve edges to roughly a tenth of a pixel under good lighting.
  • Inspection errors: false reject (good part rejected) and false accept (defective part passed). Lighting and threshold settings trade one against the other.

CMM versus vision CMMs are more accurate and handle 3D features and GD&T, but are slower and usually off-line in an inspection room. Vision is fast, non-contact and in-line, but mostly 2D or 2.5D and limited by pixel resolution and lighting.

Formulas

L = √((x₂ − x₁)² + (y₂ − y₁)² + (z₂ − z₁)²)

  • Distance between two measured points or feature centres; coordinates and L in mm.

Circle through three points: centre (a, b) satisfies (xᵢ − a)² + (yᵢ − b)² = r² for i = 1, 2, 3

  • r = radius (mm). With more than three points, a least-squares fit is used.

E = A + L / K

  • E = maximum permissible length-measuring error (μm), A = constant term (μm), L = measured length (mm), K = dimensionless constant, both from the CMM specification.

ΔL = L·α·(T − 20)

  • ΔL = thermal expansion (mm), α = coefficient of linear expansion (1/°C), T = part temperature (°C).

Pixel size (object space) = FOV / number of pixels across that FOV

  • In mm per pixel. The smallest reliably detected defect is commonly taken as about 3 pixels; measurement resolution with sub-pixel methods is roughly 0.1 pixel.

Worked examples

Example 1 (standard): hole diameter and hole-to-hole distance. Given: three points probed on the edge of a bore (after tip compensation, in the part coordinate plane): P1 (0, 0), P2 (40, 0), P3 (0, 30) mm. A second bore centre is at (60, 45) mm.

  1. P1-P2 and P1-P3 are perpendicular, so P2-P3 is a diameter (angle in a semicircle).
  2. Centre = midpoint of P2 and P3 = ((40 + 0)/2, (0 + 30)/2) = (20, 15) mm.
  3. Radius = √((20 − 0)² + (15 − 0)²) = √(400 + 225) = 25 mm, so diameter = 50 mm.
  4. Centre distance = √((60 − 20)² + (45 − 15)²) = √(1600 + 900) = 50 mm.

Example 2 (GATE level): vision resolution and thermal correction. Given: a camera with 1600 × 1200 pixels views a 120 mm × 90 mm field. Separately, an aluminium part (α = 23 × 10⁻⁶ /°C) of 200 mm nominal length is measured at 26 °C, and the CMM has E = 1.8 + L/300 μm.

  1. Pixel size = 120/1600 = 0.075 mm per pixel (and 90/1200 = 0.075 mm, so pixels are square).
  2. Smallest detectable defect ≈ 3 pixels = 3 × 0.075 = 0.225 mm. Sub-pixel edge resolution ≈ 0.1 × 0.075 = 0.0075 mm.
  3. Thermal growth: ΔL = 200 × 23 × 10⁻⁶ × (26 − 20) = 0.0276 mm (27.6 μm). The measured length must be corrected by −27.6 μm to report the size at 20 °C.
  4. CMM MPE at L = 450 mm (its longest axis): E = 1.8 + 450/300 = 3.3 μm, far smaller than the thermal effect, which shows why temperature control matters more than machine accuracy here.

Common mistakes

  • Forgetting stylus tip radius compensation, which makes holes read small and shafts read large by one ball diameter.
  • Measuring a circle with exactly three points and then reporting roundness; with no redundancy, the form error appears to be zero.
  • Building the alignment on a rough or non-functional surface instead of the drawing datums.
  • Ignoring part temperature, which often contributes more error than the machine itself.
  • Calculating vision pixel size from sensor dimensions instead of the field of view in object space.
  • Assuming a defect one pixel wide will be detected reliably.

For GATE ME

CMM and vision appear mostly as conceptual MCQs: CMM configurations, touch-trigger versus scanning probes, minimum points for features, alignment, and components of a vision system. Numericals are simple coordinate geometry (distances, circle centre and radius from points), thermal correction, and pixel resolution. Practise fitting a circle through three points quickly.

Quick check

  1. What is the minimum number of points to define a cylinder on a CMM?
  2. Which CMM configuration suits a large aircraft wing spar?
  3. A 2048-pixel line-scan camera views a 512 mm wide strip. What is the pixel size?
  4. Which lighting gives the sharpest silhouette for edge gauging?
  5. Why does the CMM software compensate for the stylus ball radius?

Answers: 1. Five. 2. Gantry. 3. 0.25 mm. 4. Backlighting. 5. Because the machine records the ball centre, not the contact point on the surface.

Interview questions

All Metrology, CIM and Industrial Engineering interview questions

Try answering each one aloud before you open it.

  1. 1.What is a Coordinate Measuring Machine (CMM) and what are its primary components?Concept

    A CMM measures the (x, y, z) coordinates of points on a part's surface with a probe moved along three orthogonal axes, and software fits features such as planes, circles and cylinders to those points to evaluate sizes, positions and GD&T tolerances. Its main parts are the structure (bridge, cantilever, gantry or horizontal arm, usually on a granite table with air bearings), linear scales on each axis, the probe system (touch-trigger, scanning or non-contact), the controller and drives, and the metrology software that does alignment, fitting and reporting against CAD.

  2. 2.Explain the principle of operation of a machine vision inspection system.Concept

    A machine vision inspection system uses cameras and image processing software to capture and analyze images of objects. The system compares the captured images against predefined criteria to detect defects, measure dimensions, or verify the presence of components. It operates by illuminating the object, capturing images, processing the images using algorithms, and making decisions based on the analysis.

  3. 3.How does a CMM differ from a traditional measuring tool like a caliper?Concept

    A caliper measures one two-point size at a time, its reading depends on the operator's feel and alignment, and its resolution is typically 0.01–0.02 mm. A CMM measures many points in a defined part coordinate system, so it can evaluate positions, angles, form and GD&T tolerances that a caliper cannot, with micrometre-level accuracy and programmable, repeatable routines. The trade-off is cost, a controlled environment and slower cycle time, so calipers remain the right tool for quick shop-floor size checks.

  4. 4.Why is machine vision inspection preferred in high-speed production lines?Application

    Machine vision inspection is preferred in high-speed production lines because it can quickly and accurately inspect products without human intervention. It reduces the risk of human error, increases throughput, and ensures consistent quality by providing real-time feedback and data for process control.

  5. 5.What happens if a CMM is not calibrated regularly?Application

    Scale errors, squareness and straightness errors of the axes and probe errors drift over time with wear, crashes, foundation settling and temperature changes, so the error map stored in the software no longer matches the machine. Measurements then carry systematic errors that nobody sees, so good parts may be rejected and bad parts accepted, and traceability to the metre is lost. Regular verification to ISO 10360 with gauge blocks or step gauges, and daily stylus qualification on the reference sphere, catch this drift.

  6. 6.In what scenarios would you choose a non-contact probe over a contact probe in a CMM?Application

    A non-contact probe is chosen over a contact probe when measuring delicate, soft, or easily deformable materials where contact could alter the measurement. It is also used for high-speed measurements and when measuring complex surfaces that are difficult to access with a contact probe.

  7. 7.Estimate the measurement uncertainty contribution of a CMM that has a resolution of 0.001 mm and a repeatability (standard deviation) of 0.002 mm.Numerical

    Resolution is treated as a rectangular distribution of full width 0.001 mm, so its standard uncertainty is 0.001/√12 ≈ 0.00029 mm. Combining independent contributions by root-sum-square: u = √(0.00029² + 0.002²) ≈ 0.0020 mm, so repeatability dominates. An expanded uncertainty at about 95% confidence uses a coverage factor k = 2, giving about ±0.004 mm, before adding temperature, probe and machine-geometry terms.

  8. 8.What are the advantages of using a gantry-type CMM over a bridge-type CMM?Application

    In a gantry CMM the bridge runs on elevated rails supported on columns, not on the table, so the part sits on the floor or a separate foundation. This gives very large measuring volumes (several metres) and lets heavy parts such as aircraft structures, large castings and car bodies be loaded by crane without loading the machine's guideways. A moving-bridge CMM is cheaper and usually more accurate for small and medium parts, so the gantry is chosen for size and weight, not for better accuracy.

  9. 9.Explain how lighting conditions affect machine vision inspection accuracy.Application

    Lighting conditions significantly affect the accuracy of machine vision inspection. Proper lighting ensures that the features of the object are clearly visible and distinguishable from the background. Inadequate or inconsistent lighting can lead to poor image quality, making it difficult for the system to accurately detect defects or measure dimensions.

  10. 10.A machine vision system inspects 1000 parts per hour. 5% of parts are defective and the system detects 98% of defective parts. How many defective parts are missed per hour?Numerical

    Defective parts per hour = 0.05 × 1000 = 50. The system misses 2% of defective parts, so missed defects = 0.02 × 50 = 1 part per hour (a false accept). Note that the 98% detection rate applies to defective parts only, not to all 1000 parts; false rejects of good parts are a separate rate.

Finished this topic? Mark it so your progress, study plan and readiness keep up.

Stuck on something here?

PreviousSurface finish, interferometry and alignment testingNext NC and CNC part programming