Dimensional inspection is the measurement of a part’s size, shape and feature locations to confirm they are within the tolerances on the drawing. It answers questions such as: is this diameter within ±0.05 mm, is this hole in the right place, is this surface flat enough? Automated dimensional inspection uses cameras, laser scanners, robots and software to make those measurements on every part, without calipers, fixtures or an operator’s judgement.
Most automated dimensional inspection is a branch of machine vision. The same cameras and lighting that find defects can also measure, provided the optics, calibration and part presentation are designed for metrology. This article covers the main automated methods, how much accuracy each can reach, how to size a camera for a tolerance, and how to choose between them.
Manual vs automated dimensional inspection
Traditional dimensional inspection relies on hand gauges, go/no-go gauges, height gauges and coordinate measuring machines (CMMs) in a quality lab. These are accurate but slow, so most plants sample a few parts per lot. Automated inspection moves the measurement to the line or the bench and checks 100% of parts, which catches drift and outliers that sampling misses.
| Manual and lab methods | Automated vision and 3D methods | |
|---|---|---|
| Coverage | Sampled parts per lot | Every part |
| Speed | Seconds to minutes per feature | Many features in one image or scan |
| Contact | Calipers and CMM probes touch the part | Non-contact; no wear or marks on the part |
| Data | Recorded by hand or per lab report | Logged automatically for SPC and traceability |
| Best for | First article, very tight tolerances, low volume | Production checks, high volume, soft or delicate parts |
The two are complementary. A CMM remains the reference for first-article inspection and calibration, while an automated system checks production parts against the same tolerances.
Automated dimensional inspection methods
| Method | How it measures | Typical use |
|---|---|---|
| 2D vision with telecentric optics | Measures edges in a backlit or front-lit image; telecentric lenses keep magnification constant with depth | Diameters, lengths, hole positions, profiles of flat or turned parts |
| Laser triangulation (laser profiling) | A laser line is projected onto the part; a camera at an angle reads its shape to build a 3D profile | Heights, step changes, warp, flatness, extruded and long parts |
| Structured light | Projected patterns are decoded into a full 3D point cloud | Complex shapes, castings, comparison with CAD |
| Rotary 360° imaging | The part turns in front of cameras so every side is measured | Round and tall parts: diameters, concentricity, wall features |
| Robot-presented multi-station | A robot moves each part through cameras, scanners and a scale | Small high-tolerance parts with several kinds of measurement |
Laser triangulation is covered in depth in what a laser profilometer is and how it works, and the trade-offs between flat images and depth data in when to use 2D vs 3D machine vision.
How accurate does the measurement need to be?
A common rule of thumb is that the measuring system should be about ten times more precise than the tolerance it checks. For a feature with a tolerance of ±0.05 mm, the total tolerance band is 0.10 mm, so the measurement uncertainty should be about 0.01 mm (10 µm) or better. Measurement systems analysis, such as a gauge repeatability and reproducibility (gauge R&R) study, is used to confirm the system meets that target on real parts.
Worked example: sizing a camera for a tolerance
Suppose the part needs a 50 mm field of view and the camera sensor is 2,448 pixels wide. Each pixel then covers 50 ÷ 2,448 ≈ 0.020 mm, or about 20 µm. Good edge-detection software locates an edge to a fraction of a pixel, typically between a quarter and a tenth of a pixel, so the edge position is known to roughly 2 to 5 µm in ideal conditions.
That looks comfortably inside a 10 µm budget, but lens distortion, lighting, vibration, temperature and how consistently the part sits in the fixture all add error. In practice the budget is shared across these sources, which is why optics, fixturing and calibration matter as much as the camera.
| Input | Value | Result |
|---|---|---|
| Field of view | 50 mm | |
| Sensor width | 2,448 pixels | ≈ 20 µm per pixel |
| Sub-pixel edge location | 1/4 to 1/10 pixel | ≈ 2–5 µm |
| Tolerance ±0.05 mm, 10:1 rule | 0.10 mm band | Target ≈ 10 µm uncertainty |
For features measured in microns, high-magnification cameras with telecentric optics are used. For example, the RobotVIEW robotic vision inspection cell reaches micron-level accuracy in suitable setups. For tight work at very small scale, see vision systems for very high tolerance part inspection.
Common sources of measurement error
- Perspective error: with a standard lens, a feature closer to the camera looks bigger. Telecentric lenses remove most of this.
- Lighting: front lighting can shift apparent edges on rounded or shiny parts; a backlight gives the sharpest silhouette edges.
- Part presentation: a part that tilts or shifts in the fixture changes the measurement. Robots and precision bases present parts the same way every time.
- Temperature: metal parts and fixtures expand as they warm; measure at a stable temperature or compensate.
- Calibration drift: check against a certified reference part on a schedule.
How to choose an automated dimensional inspection system
| If your parts are… | Consider | Example Sciotex system |
|---|---|---|
| Round or tall, with features on every side | Rotary 360° imaging | Benchtop vertical 360° inspection |
| Long, extruded or need height and profile data | Laser triangulation scanning | Benchtop 3D laser scanning system |
| Small, high tolerance, needing several measurements | Robot-presented multi-station cell | RobotVIEW robotic inspection |
| Moving on a production line | Inline 2D and 3D stations | ConveyorVIEW conveyor inspection |
- List each dimension with its nominal value and tolerance.
- Decide whether you need 100% inspection or sampling, and at what rate.
- Pick the method for each feature (2D silhouette, 3D profile, rotary view).
- Set the accuracy budget with the 10:1 rule and plan a gauge R&R study.
- Define what happens to the result: pass/fail, sorting, SPC data, alarms.
- Test real good and bad parts before choosing hardware; Sciotex offers free testing of sample parts.
Need to measure every part, not just a sample?
Sciotex builds benchtop and inline vision inspection machines that measure dimensions and find defects in one pass, designed, built and supported in the US since 1997.
Frequently asked questions
Dimensional inspection is the measurement of a part’s size, shape and feature locations to confirm they are within the drawing tolerances.
It uses cameras, laser scanners, robots and software to measure parts automatically, usually on every part, without hand gauges or an operator’s judgement.
It depends on the optics and field of view. A 50 mm field on a 2,448-pixel sensor gives about 20 µm per pixel, and sub-pixel edge detection can locate edges to a few microns. High-magnification telecentric setups reach micron-level accuracy.
It is a rule of thumb that the measuring system should be about ten times more precise than the tolerance it checks, so a 0.10 mm tolerance band calls for about 0.01 mm measurement uncertainty.
Usually not. A CMM stays the reference for first-article and calibration work; automated systems check production parts against the same tolerances at much higher speed.
Related articles
- Machine Vision: The Complete Guide to Machine Vision Systems
- What Is a Laser Profilometer? How Laser Triangulation Measures Parts
- Robotic Inspection: How It Works, Types and Where It Fits
- When to Use 2D vs. 3D Machine Vision Systems
- Considerations for Vision Systems in Very High Tolerance Part Inspection
