A properly designed and tested machine vision system provides robust results and needs little tweaking. If that doesn't sound like your typical machine vision system, then perhaps some critical items were overlooked. If the application doesn’t give consistent results and needs constant adjustments by a select group of programmers or maintenance personnel, then perhaps your vision is failing.
Often with troubled machine vision applications, the vision system, lens or illumination may be pushed to do inspections the hardware or software algorithm is not capable of. The bottom line is that machine vision systems do not perform well in every application, so when starting a project be sure to test the application, using vendor or in-house resources. Feasibility studies, lens and lighting specifications and system repeatability and reliability testing go a long way in ensuring a robust machine vision application.
Vision system options—hardware and software
A broad range of hardware and software is available for automatic-inspection and part-identification machine vision applications. With machine vision you often get what you pay for, so don't be too cheap. The vision hardware starts with specific-use devices such as barcode readers, 2D ID readers and other auto ID products. These are on the lower end of the cost scale.
As the application requirements expand, vision sensors can be used. These small all-in-one devices have more inspection tools and work well for edge-find and pixel-counting applications. If it’s a difficult photo-eye application or the photo eye almost works for the application, the vision sensor probably will work.
Also read: Add vision to robots — see the difference
As the application becomes more difficult and many inspection tools are needed, the higher-performance vision system is used. These all-in-one devices or packaged vision hardware have much greater capabilities—more powerful processors, better inspection algorithms, higher-resolution image processors and a wider variety of inspection tools—than vision sensors. These vision systems are often the sweet spot for machine vision and are capable of supporting 90% of all machine vision applications. And they’re much easier to learn to program than many software-based machine vision systems.
The high-end vision systems are often software based and can run on a variety of PCs and use a variety of camera hardware. Although these software-based machine vision systems can handle any application, their powerful capabilities and high-end programming languages often can make development difficult. These high-end vision software applications are best used by experienced programmers.
Machine vision fills many product-improvement, error-proofing and quality system requirements. Many part-inspection checks that are done manually can be done with machine vision. For example, the barcode reader, 2D ID reader, and optical character recognition and verification (OCR/OCV) functionality help to ensure the proper product number and lot number are being made with little operator involvement.
Often a manual off-line inspection can become an in-process inspection. Assembly confirmation and error-proofing are big parts of machine vision and some of the more robust and appropriate applications. This includes part-presence detection and confirmation that the correct type of part is present. The assembly inspections can also include color check and printed graphic checks. Many vision systems have a defect tool, which can be difficult but powerful. But, as the machine vision system moves into debris and foreign material detection, the application's reliability and repeatability can be difficult and show poor performance, so it must be tested carefully. Or just run for your life, and don't look back.
Part-presence detection using pattern-matching and measurement tools can help to determine part orientation and simplify mechanical fixturing. The part measurements available using machine vision are often a final inspection quality check of the product assembly. These pattern matching and measurement tools and related inspection results also provide vision-guided motion and vision-guided robotics capabilities—just pass the coordinates and rotation angle to the motion system or robot. Testing will show x, y and theta repeatability and accuracy of the chosen machine vision system, which will affect the final manufactured product. While machine tending may allow a wide positional tolerance, precision vision guided robotic assembly will need highly accurate machine vision.
Testing, specifications and inspections
Careful testing is a required first step in the machine-vision design process. Borrow a camera, camera stand, lens and lights and build a test system. Write a simple test application and collect the data. The repeatability and accuracy data will often show the need to change camera resolution or lens field of view. A variety of lighting methods must also be tested to determine the proper illumination for best inspection results. Then it all will need to be repeated when you discover the focal distance must be changed due to mechanical constraints of the machine it will be mounted to.
With all automated assembly machine vision applications, it’s important to carefully define the requirements at the start and then test the camera, lens and illumination chosen.
The field of view, depth of field and focal distance all feed in to camera resolution requirements and final system capability, and zooming in helps with accuracy. Also consider speed of the camera. High throughput will be needed for fast production lines. This includes both inspection rate and frame rate—images per second.
With machine vision, there is a huge amount of variables affecting the outcome. Ambient light, shadows, color changes, surface reflections, surface finish and many other material characteristics affect the inspection. Inspection tools such as edge detect, pixel count, light count, dark count, pattern match and caliper measurement are all affected by the machine vision installation. It's not just about the program. Lens, light, environment conditions and choice of camera all lead to success or failure in machine vision.