Vision-Based Measurement and Dimensional Inspection

The chain that determines measurement accuracy is lighting, optics, resolution and calibration. Backlighting gives a sharp silhouette of the part and is the most reliable method for external dimensions; front lighting is used for surface features. The camera and lens are selected so that the whole part, or the region to be measured, is seen at the required mm/pixel resolution; where perspective error is critical, a telecentric lens is preferred.

Calibration converts pixels in the image into millimetres in the real world: a calibration plate of known dimensions is imaged in different positions, and lens distortion and scale are calculated. Measurement is performed by locating edges with sub-pixel precision, and is verified at set intervals with a reference part to account for temperature or mechanical changes.

The role of AI is not to perform the measurement, but to reliably find where to measure: it locates where the part lies in the image and in which orientation, as well as the hole, corner or cross-section region to be measured, even under variable lighting and surface conditions. The measurement result is compared with the tolerance limits; the result goes to the PLC and is logged for every part, and by monitoring the trend a warning can be given before the process drifts out of tolerance.

Dimensions perpendicular to the camera, such as height, cannot be read from a single flat image; laser line triangulation is used instead. A laser line is projected onto the product and the camera views it at an angle; the shift of the line in the image is converted into height in millimetres through calibration. On concrete paving block lines, the line sweeps across each row of blocks as the pallet leaves the press and the height of every block is measured. The aim is to produce blocks consistently close to the minimum height the standard allows and to see a drift the moment the press setting changes; measurements are logged against the pallet number and shown to the operator in real time.

The camera image reaches the edge AI computer over GigE. The part and feature location model runs on the GPU, and edge detection and measurement are performed with OpenCV-based algorithms. Results and measured values go to the PLC via Modbus TCP or OPC UA, and the OK/NOK signal can go directly to a digital output if required.

Thanks to its high resolution, the Crevis MG-A121L-9 in stock (4024 × 3036, 1.85 µm pixels) is suited to viewing a wide area at a fine scale; because it has a rolling shutter, measurement is carried out while the part is stationary or with a strobe in Global Reset mode. The V1628-MPY 12 MP lens in stock matches this resolution; for measurements where perspective is critical, telecentric optics are selected separately.

What decides the success of an AI project is your product, your lighting and your line speed, not a catalogue figure. That is why every project starts with your own samples: first, images are collected on the bench with the right camera, lens and lighting, an initial model is trained, and the false reject and miss rates are measured on your parts. If the result meets the acceptance criterion, the system runs on the line as a pilot, initially in a mode that only logs and makes no decisions.

In measurement projects, the feasibility study centres on a repeatability test: your reference parts with known dimensions are measured repeatedly with the same system, and the spread of results is compared with your tolerance. During the pilot phase, camera measurements are logged side by side with your existing measurement method (caliper, micrometer, CMM).

Vision-based measurement is usually added at the end of an existing line, replacing or supplementing manual sample measurement. For example, on cut-to-length and flying shear lines, the length of the cut part can be verified with a camera, and any drift trend can be fed back to the cutting control.