Case studies / Computer vision inspection

Computer vision inspection for PCBAs and LED lighting

One universal inspection software that we configure for each product. It compares what the camera sees with a good reference, marks every wrong, missing or dead part in red, and passes the result on to the line.

Talk to the founder
Universalone software for PCBAs, LED lighting and more
24 MPcamera for the finest details
60lamp models in one application
2+cameras on one ordinary PC
A panel of three circuit boards checked by the software: the top image has nothing marked, in the bottom image a missing capacitor is marked in red

The challenge

A wrong, missing or turned component, or a dead LED, can hide in plain sight. A person checking by eye gets tired, and every product looks different, so a fixed inspection machine needs new work for each new part.

Manufacturers need a check that is fast, the same every time, and quick to set up for a new product.

Our solution

We built universal inspection software in LabVIEW with NI Vision. The user configures it for a particular part, whether a PCBA, a lamp or an electro-mechanical product, so a new product needs a new configuration and not new programming.

It is used by partners in PCBA manufacturing, LED lighting manufacturing and the automotive sector.

How it works

  • 1. Teach a good part

    The user shows the software a good part and sets up the regions to check and the tolerances. This is the reference, and it is saved with the configuration for the product.

  • 2. Capture

    A camera takes an image of every part, with controlled lighting and an external trigger when the line needs it.

  • 3. Compare

    The software compares the live image with the reference and finds the differences: a missing component, a wrong one, a shifted or turned one, or an LED that does not light.

  • 4. Mark and decide

    Every difference is marked in red on the image, so the operator sees at once what is wrong, and the part is passed or failed.

  • 5. Report

    The result is shown, saved and exported, for example to Excel, and sent to the server. It can also print a label for the part.

live image camera sees the board reference good part, taught once Component missingmarked in red, result sent to the line

Finding the wrong component

This is the application during a demonstration on a panel of three identical boards. The software compares the live image with the reference and marks every difference in red.

Live image of the three boards with a missing capacitor marked in red
The live image, with the missing capacitor on the middle board marked in red.
The reference image of a good panel, in grey
The reference image of a good panel, which the live image is compared with.
All three boards complete, nothing marked
All components are in place, and nothing is marked.
A capacitor is missing on the middle board and is marked in red
A capacitor is missing on the middle board, and the software marks the spot.
Two boards carry the wrong capacitors and both are marked in red
The wrong capacitors are fitted on two boards, and both are marked.

Case: testing LED street lights

For an LED lighting manufacturer we built a complete test cell, first delivered in 2014. The camera looks at the lit luminaire and the software counts the working LEDs, then a spectrometer measures the light and the software compares it with the limits.

The same application handles three families of luminaires and about 60 models. It also talks to a label printer and to the server, and saves every result.

Camera image of a lit LED luminaire, a grid of bright LEDs on a dark background
What the camera sees: every working LED is a bright spot, and a dark spot is a dead LED.
Part of the processed image, with every LED found marked by a red shape
Part of the processed image: the software marks every LED it finds in red and counts them.

One test, from switch-on to label

  • Switch on

    The operator switches the luminaire on, and the software notices by itself.

  • Warm up

    It waits for the luminaire to reach its working temperature, with a countdown on the screen.

  • Count and measure

    The camera counts the working LEDs. Then the spectrometer measures the light, such as colour and luminosity, and everything is checked against the limits.

  • Confirm and print

    The operator confirms the result, and a label with the serial number is printed. A failed test prints a failure label with the measured values.

  • Switch off

    When the luminaire is switched off, the software is ready for the next one.

The test cell on the production line with a luminaire under the camera
The test cell on the production line.

A reference luminaire sits beside the cell. The software measures it at any time to prove that the camera, the spectrometer and the settings are still correct, and it reports a calibration problem, for example when something is blocking the camera.

Later we extended the same platform with checks for the lit surface: bright and dark spots, the uniformity of the light, and masks that skip logos and labels.

Light on hardware

The software uses industrial GigE cameras with lenses of adjustable focus and aperture. A 5 megapixel camera at about 10 frames per second is enough for many parts, and a 24 megapixel camera reads the finest details.

It runs on an ordinary Windows PC. In the LED project, two cameras and the whole application used only about 15 to 25 percent of a basic PC, and many more cameras can be connected over the network.

Windows or real-time

The software is developed in LabVIEW and delivered as an executable for Windows. It can also be built for real-time targets, so it can run inside a controller on a real-time operating system.

It combines naturally with our testers, so one station can do electrical tests and vision checks together.

What the customer gets

  • One software, many products

    A new part needs a new configuration, not new programming.

  • Same result every time

    No tired eyes: every part is checked in the same way.

  • Errors you can see

    Differences are marked in red on the image, so a repair starts in the right place.

  • Integrated with the line

    Results go to the screen, to Excel, to the server and to the label printer.

  • Proven in production

    Used for PCBAs, LED lighting and automotive parts.

  • Scalable

    From one camera on a bench to many cameras on a line.

Computer visionLabVIEWNI VisionPCBALED lightingAutomotiveGigE cameras

Tell us about your project

Send us a few lines about what you want to build or improve. I will reply personally.

Email us