Machine Vision Lighting: A Study in Contrast

In its most basic sense, the lighting strategy for almost any machine vision inspection will use fiber-optic-based, light-emitting-diode, fluorescent or structured laser light sources. The question that ultimately determines lighting success is also fairly basic: Does the light source produce enough contrast between the background and the features to be inspected? If not, even the highest-resolution machine vision system will fail at its task.

Selecting the right lighting for a machine vision application is anything but basic, though. Optimized illumination will depend on several issues, including the following:

• The work material. Is it translucent, transparent, opaque, shiny?

• The location of defects or features. Are they inside an opaque, transparent or translucent component or on the surface? With what background must they contrast?

• The environment. Not all machine vision inspection tasks can be set up in a black-box environment, where stray lighting is easily controlled. Something as simple as a change in ambient lighting of the work environment from daytime to evening could affect the lighting strategy.

• The camera technology. Area-scan vision systems based on smart camera technology, available from companies such as DVT and Keyence, are gaining in popularity in applications from color to gray-scale inspection because of relatively low system cost and camera flexibility. The lighting demands in such cases can differ dramatically from those required with high-speed line-scan applications, such as optical character recognition or web inspection. For example, long-working-distance limitations common with LEDs may not be as much of an issue because the inspection cameras and the light source are small enough that they can often be located on the manufacturing or assembly line fairly close to the component under test.

• The imaging sensor. Is it CMOS or CCD? With CCD cameras, settings such as gain and shutter speed will affect system sensitivity and required illumination. Still, these sensors tend to be more light-forgiving than CMOS sensors, which, despite this, are gaining popularity in the machine vision market.

• The optics. In general, optimization of illumination becomes more critical with higher-power lenses.

To further complicate issues, Joe DiRuzza, director of sales and product management at StockerYale, with headquarters in Salem, N.H., states that most machine vision projects have several goals. It isn’t uncommon for automotive or semiconductor applications, for example, to require multiple inspections on the same component throughout the manufacturing and assembly process. Each inspection might check for something different, such as dimensions, surface defects, or the presence or absence of features, and could thus require different lighting techniques.

Consider the inspection demands found with a foil pill pack. A major concern here, according to DiRuzza, often is the specular nature of the surface. In other words, light reflects easily, so shiny work material tends to require a diffuse lighting source. Depending on the inspection requirements, simple fluorescent lighting might work because it is naturally diffuse.

The inspection task might also require some type of on-axis or polarized illumination scheme using fiber optic or LED lighting. Here, the selection decision could be based on something as simple as device lifetime. LEDs can sometimes last for 100,000 hours, whereas halogen light sources used with fiber optics might require changing after 2000 hours.

StockerYale engineers report that, in some cases, LED brightness limitations may dictate the need for fiber-optic-based strategies when direct, intense lighting is required on the specimen/object to be inspected. Research into LED luminous intensities is continuing at a fairly furious pace; however, it is powered in part by the technology’s potential in bioterrorism and national security applications. Scientists at Sandia National Laboratories in Albuquerque, N.M., say that, although the deep-UV LEDs they are focusing on for such tasks are unlikely to be used for lighting, basic advances in wide-bandgap semiconductor materials science could boost the luminous efficiencies of the UV and visible LEDs that are used for illumination.

Perhaps the one conclusion that can be drawn, especially after reviewing the following series of articles, is that because lighting for machine vision is anything but simple, development of the illumination design should take place as early in the machine vision project as possible. The goal should be to maximize contrast of the object to be inspected with the lowest-cost lighting scheme. End users might be surprised at how something as simple as the addition of low-cost color filters can boost the contrast of defects, even in a gray-scale environment.