|Component in grayscale. |
The Achilles heel of AOI efficiency is arguably the occurrence of false alarms. A solution to delivering more confident results by dramatically reducing these false alarms leaves behind the common approach of grayscale image processing in favor of true color — to literally extend the depth of the analysis process.
Automatic Optical Inspection for populated circuit boards is now essential for manufacturers looking to keep productivity rates high while also coping with miniaturized components and denser circuit patterns. Add the need to operate continuously 24/7, and it's clear that the inspection process is something that rapidly moves beyond the capability of human operators.
Working on the principle that an optical defect detected requires corrective measures, AOI has a wide range of applications. General advantages of AOI include its contactless measuring capability and flexible and fast programming compared to electrical test equipment.
|Component in 24-bit color. |
Limitations include the fact that only visible defects can be detected, and particularly the occurrence of "false calls" caused by observed visual differences that may be of no consequence for the circuit's actual functionality.
Matching Performance Value
An AOI system can be qualified as a good machine if it matches its specified performance indicators: accuracy, sensitivity and specificity values — values that should be as close as possible to 100 percent. These performance indicators are the statistical measures of the performance of an AOI system — determined by its inspection results: true positives, false positives, true negatives and false negatives. This means that while the number of false calls or escapes are a measure of the AOI system quality, but also the results related to the number of inspected test points.
|Missing component in grayscale. |
Another quality factor is the time needed to create a program to achieve the specified performance; this figure should be as low as possible. The first step to reach acceptable performance indicators is to accurately describe the component and circuitry under test to the AOI machine. There are many methods that can be used to do this, with different vendors endorsing a variety of solutions from learning a captured image (or averaging a succession of images) and correlation principles to modeling and programming-in details of each component and board topology. A combination of techniques most often used to refine the AOI's understanding of its inspection subjects.
Acceptable Tolerance Window
An empirical method of defining the acceptable window of tolerance for any given component or camera field of view, employed in most AOI systems in one form or another, is to test a number of "good" boards to define the acceptance envelope. But capturing images to determine acceptability presupposes that you have a discerning imaging system in the first place, and this is not always the case. Among the key differentiators between today's AOI systems are the deployment of digital cameras with all the associated arguments about resolutions and field of view (FOV), of lighting techniques, of image processing technologies, and of color depth.
The great majority of AOI systems capture color digital images from the CCD or CMOS camera sensor but use gray scale processing for data analysis. To be sure, almost all display a live image in color on the equipment monitor to allow the operator to see what's going on — but they only process grayscale, either as 8-bit or 16-bit images.
|Missing component in 24-bit color. |
Grayscale is a range of shades of gray without apparent color. The darkest possible shade is black, which is the total absence of transmitted or reflected light. The lightest possible shade is white — the total transmission or reflection of light at all visible wavelengths. Intermediate shades of gray are represented by equal brightness levels of the three primary colors — red, green and blue (RGB) in camera systems.
The brightness levels of the red (R), green (G) and blue (B) components are each represented as a number from decimal 0 to 255. Every pixel in an RGB grayscale image has the value (R + G + B)/3. The lightness of the gray is directly proportional to the number representing the brightness levels of the primary colors. Because there are 8 bits in the binary representation of the gray level (256 = 28), this imaging method is called 8-bit grayscale.
|Melf diodes show very poor contrast in grayscale that makes for difficult detection, while checking for proper polarity.while the contrast and detection improvement with full 24-bit color is significant. |
In some systems that use the RGB color model, there are 216, or 65,636, possible levels for each primary color. When grayscale is being calculated in this system with (R + G + B)/3, the image is known as 16-bit grayscale 65,536 = 216 which is the binary representation. As with 8-bit grayscale, the lightness of the gray is directly proportional to the number representing the brightness levels of the primary colors. As you'd expect, a 16-bit digital grayscale image consumes more memory or storage than the same image, with the same physical dimensions, rendered in 8-bit digital grayscale.
Going Only Halfway
A 16-bit grayscale image gives the AOI processor more data to work with than an 8-bit image. This has nothing to do with the resolution of the camera, which simply relates to the amount of pixel data to process; instead it is about the depth of data available for processing. With more shades of gray to represent subtle variations in color on the inspection subject, the process can be more discerning — a good start to achieve better performance indicator values. But it's only half a step in the right direction.
To really get the maximum depth of data from a captured digital image — and therefore to have the most data with which to discern subtle characteristics of the imaged subject — full color processing is required. Full RGB color requires that the intensities of three color components be specified for each and every pixel. It is common for the intensity of each component (R, G and B) to be stored as an 8-bit integer, and so each pixel requires 24 bits to completely and accurately specify its color. Better systems are also transforming this raw 24 bits RGB data to different color models with the ability to switch between the different models. By doing so, new parameters are being created such as Hue (the pure color) and Saturation (the color strength) along with the Brightness (grayscale).
|Detecting color of colored LEDs in grayscale is virtually impossible, but color shows clearly in 24-bit color. |
Applying tolerance values to these parameters make the system work more reliably, since the colors are spread logically in the model and are very close to reality. Image formats that store a full 24 bits to describe the color of each and every pixel are known as 24-bit color images.
Capturing 24-Bit Color
Using 24 bits to encode color information allows 224 or 16,777,216 (16.8 million approx.) different colors to be represented. An AOI imaging system that captures and processes 24-bit color is at least as capable as an experienced human operator in discerning colors. And since it an effective automation solution, the system can do it a lot faster and continuously.
|Looking at melf resistor values, grayscale imaging presents a difficult subject to discern, but color clearly shows the difference. |
24-bit color demands more processing power and greater data storage capacity. Today however, cost-effective memory and hard drive storage options are abundant. And Moore's Law still holds true in the exponential growth of computer processing power over time. All AOI systems from Marantz Business Electronics use 24-bit color technology with several color transformation models, and deploy every graphic designer's favorite tool to cope with the 24-bit image processing workload — an Apple Mac at the heart of every machine. Macs are renowned for their exceptional graphics capability. An obvious advantage of 24-bit color AOI is the ability to discern colors for certain components as well as printed legends. Combined with synthetic modeling principles that start with a "real" picture, very accurate yet flexible models can be quickly produced that provide optimum tolerance. This eliminates the possibility of an operator "adding" an incorrect component to the acceptability range — thereby making the inspection process meaningless.
Detecting color on components is useful for melf resistor banding and orientation of melf diodes. It's also very useful for checking LED colors. Grayscale struggles with the difference between red and green — the most common colors for LEDs. But component color checking is only the tip of the iceberg, and like an iceberg, it's the depth that matters.
Among the common causes of false alarms are simple factors like board color, board surface finish and the contrast between boards and components. Here, 24-bit color can make a huge difference. Introducing additional technologies like coaxial lighting and data processing techniques such as Synthetic Imaging indicates just how much detection improvement can be gained when starting with 24-bit color data.
|Introducing additional technologies like coaxial lighting and data processing techniques such as Synthetic Imaging indicates just how much detection improvement can be gained when starting with 24-bit color data. |
Traditionally, problems with AOI have involved the level of expertise and/or the amount of programming time that must be invested before the machine can even begin its inspection task. This means that truly useful new-product inspection programs often require too much programming time for today's high mix, quick-turn manufacturing environments. "Useful" inspection programs optimize the balance between failing to detect true defects (known as escapes) while at the same time maintaining a low false defect rate. Combining sophisticated device libraries for component identification with the import of pick-and-place program data for positional location is one example of how AOI machine programming can be accelerated. Achieving an optimal balance of true defects and false alarms quickly is the ultimate measure of fast programming.
A system that is fast to program but which overlooks real defects (i.e. permits escapes) is just as useless as one that finds all defects and bombards operators with false alarms. In the latter case, operators will tend to quickly click through the many false alarms they are seeing and inevitably begin passing through real defects (operator escapes). Today's AOI systems typically offer ongoing learning that can gradually provide best performance indicators, and indeed, 24-bit processing can dramatically accelerate that artificial intelligence process. But total reliance upon such an approach not only runs fundamentally against lean manufacturing principles, it is volume-dependent and thus totally unsuitable for NPI and other high-mix manufacturing situations where effective AOI has the potential to deliver the most impact.
Contact: Marantz Business Electronics Europe, Beemdstraat 11, 5653 MA Eindhoven, Netherlands +31 40 2507870 fax: +31 40 2507840 E-mail: firstname.lastname@example.org Web: http://www.escapethegrayworld.com