Proprietary Measurement Engine
How MeasureMaster Turns Pixels Into Metrology
A layered pipeline that goes from raw image to a defensible geometric statement — with a confidence figure attached to every number.
Architecture
The Measurement Pipeline
Six stages. Each one is designed to fail loudly rather than quietly return a wrong number.
Acquisition & Calibration
The image is normalised for exposure and contrast, and a scale factor in microns-per-pixel is established from a calibration artefact or a known reference dimension. Without a scale, the system reports geometry in pixels and says so — it never invents a physical unit.
Feature Isolation
The circular feature of interest is separated from background, tooling, glare and shadow. Adaptive thresholding combined with morphological cleanup handles the mixed lighting typical of a shop-floor bench rather than a metrology lab.
Sub-Pixel Contour Recovery
The boundary is refined below pixel resolution using intensity-gradient interpolation across the edge transition. This is what lifts accuracy beyond the naive limit of one pixel per measurement and makes small ovality percentages detectable at all.
360° Geometric Analysis
From the recovered centroid, diameters are sampled continuously around the profile. The engine computes best-fit circle and best-fit ellipse, then derives Dmax, Dmin, mean diameter, standard deviation, ovality percentage, circularity, radial deviation per angle, and the precise angular positions of the extremes.
Confidence Scoring
Edge-point count, contour continuity, fit residual and contrast quality are combined into a confidence figure and an estimated measurement error. A low-confidence result is surfaced as a warning, not buried — a measurement you cannot trust is worse than no measurement.
AI Interpretation
The metric set is passed to a language model layer that translates geometry into diagnosis: which wear pattern this signature resembles, what process cause is most likely, and what corrective action to take. The raw numbers remain visible alongside, always.
Outputs
What You Get Back
Every analysis produces a complete, exportable geometric record.
Dimensional Set
- Mean diameter
- Maximum & minimum diameter
- Standard deviation
- Angular position of extremes
Form Set
- Ovality percentage
- Circularity index
- Radial deviation per angle
- Best-fit ellipse parameters
Trust Set
- Confidence score
- Estimated accuracy
- Measurement error estimate
- Edge points used
Deliverables
- Annotated overlay image
- Diameter-vs-angle plot
- Full CSV export
- AI diagnostic summary
Reading the Ovality Signature
Plotting diameter against angle turns a die's condition into a shape you can recognise at a glance. Different wear mechanisms leave different fingerprints, and that pattern — not the single ovality number — is what tells you the root cause.
Two-lobe sine
Classic ovality. Usually misalignment or asymmetric lubrication.
Single sharp spike
Localised damage — a chip, score line or hard inclusion strike.
Three or more regular lobes
Chatter or clamping distortion rather than simple wear.
Rising baseline over time
Uniform bore growth — the die is drifting oversize and is due for reconditioning.
Deployment
Three Ways to Run It
Browser / Offline Bench
Analysis executes locally in the browser. Images never leave the machine — important for plants with proprietary tooling geometry.
- No installation
- Works air-gapped
- Ideal for the die crib
Inline Station
A fixed camera and controlled light source at the line, running continuous checks with tolerance-based alerting.
- Continuous monitoring
- Threshold alarms
- Machine-agnostic mounting
Server / Fleet View
Central analysis with per-die history, so ovality growth is tracked as a trend across the whole tooling inventory.
- Per-die trend lines
- Condition-based scheduling
- Multi-plant rollout
On our intellectual property
The specific edge-recovery, fitting and confidence-estimation methods that make this pipeline work on low-cost optics are proprietary to MEASUREMASTER LLP and are maintained as protected material. This page describes what the system does and why, not the implementation details. See the IP & Legal Notice for the full position.
Honest limits
We would rather tell you where this does not work than have you discover it on your own line:
- Accuracy is bounded by your optics and your calibration. A blurry image with no scale reference produces a confident-looking number that means nothing. The confidence score exists precisely to catch this.
- Heavily occluded or partially visible profiles will be rejected rather than extrapolated.
- Highly reflective polished bores need controlled lighting; specular glare is the single most common cause of degraded results.
- This is a geometry tool, not a materials tool. It measures shape. It does not assess hardness, coating integrity or sub-surface cracking.
Run the Engine Yourself
The live demo executes the real analysis pipeline in your browser.