Machine vision system for inspection

01

Size the camera by pixels across the smallest feature

1stVision, a machine vision camera distributor, states that finding an edge needs at least 2 pixels given excellent contrast, and that a robust system ideally has 3–4 pixels across an edge or feature. Its method divides the smallest feature by the pixels wanted across it to get the spatial resolution, then divides the field of view by that to get the sensor pixels needed: a 0.25 mm pin hole on a 20 mm part, with a 40 mm × 30 mm field of view to allow for misplacement and 4 pixels across the hole, needs 0.25 / 4 = 0.0625 mm per pixel and 40 / 0.0625 = 640 pixels in X. [1]

1stVision notes that the method assumes optimum conditions and a lens that does not limit resolution, and that the pixels actually required depend on lighting, optics and the processing algorithms. [1]

Editorial calculation: for a casting inspected in a 200 mm × 150 mm field of view, finding a 0.5 mm defect with 4 pixels across it needs 0.5 / 4 = 0.125 mm per pixel, so 200 / 0.125 = 1,600 pixels in X and 150 / 0.125 = 1,200 pixels in Y — about a 1,600 × 1,200 (1.9 megapixel) sensor before any margin. The part, field and defect sizes are assumed examples; use your own. [1]

Give the supplier sample good and defective parts, the smallest defect or tolerance to detect, the field of view including part placement variation, line speed and parts per minute, part surface (machined, cast, reflective, dark), working distance and space for camera and lights, reject or PLC interface, and data and traceability requirements.

02

Freeze the motion with exposure or strobe

Texas Instruments' technical note on LED lighting for machine vision states that holding motion blur to one pixel or less is the best approach, by shortening exposure time through the camera shutter, the strobe duration or both: tE = FOV / (v × NP), with field of view and pixel count both taken in the direction of motion. Its example — 12.8 mm field of view, 1,280 pixels, conveyor at 3 m/s — gives 3.3 µs. [2]

TI's LED lighting control reference design described in the note drives a string of five high-power LEDs at 200 mA to 2.4 A, with up to 40 W pulse power, strobe durations as short as 200 ns and frame rates up to 10 kHz, from an 8–36 V input with a 15 W input power limit and an isolated trigger interface. [2]

Editorial calculation: for the casting example above at 0.125 mm per pixel on a conveyor moving at 0.5 m/s (500 mm/s), one pixel of blur is reached in 0.125 / 500 s = 250 µs, so exposure or strobe should be at or below 250 µs. The conveyor speed is an assumed example. [2] [1]

03

Ask for EMVA 1288 camera data

EMVA 1288, the European Machine Vision Association's standard for characterising image sensors and cameras, took effect as Release 4.0 in June 2021, split into a Linear document for cameras with a linear response and a General document for wider classes of cameras. [4] [3]

Release 4.0 Linear lists sensitivity, temporal noise and linearity; nonuniformity; defect pixel characterisation; and dark current as mandatory measurements, with temperature dependence of dark current and spectral quantum efficiency as optional. It defines the absolute sensitivity threshold as the mean number of photons giving a signal-to-noise ratio of 1 and the dynamic range as saturation exposure divided by that threshold, and notes that saturation capacity “is normally lower than the full-well capacity”. The Linear method cannot be used where preprocessing such as debayering, denoising or edge sharpening breaks its noise assumptions. [3]

Editorial note: comparing two cameras' EMVA 1288 datasheets is more meaningful than comparing pixel counts alone, because the same resolution can come with very different noise and dynamic range.

04

Compare what suppliers offer

Suppliers linked to this page list machine vision inspection machines or components for them; each card links to the page reviewed.

Listings include complete CCD inspection and sorting machines as well as SWIR cameras and machine vision lenses; ask a component supplier who integrates the lighting, software and reject handling.

05

Trace the system components

This is a functional reference built from the sources cited; it is not one maker's specification.

06

Fill the specification fields with values for your parts

A machine vision quotation needs values for your parts and line. The table explains each field, gives published values where a reviewed source states them, and says what to obtain from the supplier.

07

Send the part, line and inspection data together

Paste it into your enquiry below or into an email to a maker, then replace every bracket.

Parts: [samples good and bad; material; surface]
Inspection: [defects or dimensions; smallest feature; tolerance]
Field: [field of view; placement variation; working distance]
Line: [speed; parts per minute; trigger]
Camera: [resolution; frame rate; EMVA 1288 data]
Lighting: [type; strobe; pulse length]
Outputs: [reject; PLC interface; data and traceability]
Environment: [temperature; dust; washdown]

See manufacturer records Send an enquiry through CNFX