Ten years ago, automated inspection systems, otherwise known as machine vision systems, were workable in the lab–but very difficult on the factory floor. Over time, though, advances in silicon chips, optics and lighting and vision processors and the standardization of software and hardware platforms have significantly improved the overall performance of automated inspection systems.
While these advances make systems more attractive, they also make equipment choices overwhelming. And looking for a vendor is just as difficult because more than 200 machine vision suppliers have joined in the quest for quality with a range of inspection products.
More often than not, these suppliers offer niche products that solve one problem, leaving expansion capabilities limited. How do you know what equipment is best for your needs?
Historically, the electronics industry has been one of the largest consumers of automated inspection systems. Due to the small size of components in electronics assembly, human inspection has become almost impossible. There have been thousands of successful applications applied to the assembly process of electronics parts, including circuit board assembly; LCD, LED and vacuum fluorescent display inspections; coplanarity gauging; position verification; and inspection and placement of surface-mount devices (SMDs).
Developing the Specification
The first step in any purchase process is to know and understand your application. Several factors affect the price of an inspection system. Knowing your needs in advance will eliminate disappointments later as well as facilitate discussions with potential suppliers. Address these topics:
(o) Project Scope–What are you trying to accomplish; for example, quality improvement, yield increases or direct labor savings? What type of inspection is required, such as gauging, flaw detection, presence/absence, assembly verification or motion guidance?
(o) Part Definition–What types of parts are you inspecting? What are the allowable tolerances or variations? This will determine the required spatial resolution of the inspection system.
How does the appearance of the parts vary? How may they be expected to change in the future; for example, different materials or surface finish?
(o) System Parameters–How are the parts presented on the production line? What is the rate (parts per minute)? What systems will interface with the vision system?
Facility issues also will play an important role. What kind of lighting will be needed? How will the rejects be handled? Will the parts stop at the inspection station or continue moving?
If the parts are moving, what will be their speed? Where will the inspection take place? Clearance for lighting and cameras could be an issue. Will there be a host computer interface? If so, what kind of data will be transferred?
(o) Vision System Performance–What is the expected part rate on the line? What kind of accuracy is necessary? What kind of expandability do you need?
(o) Run-Off/Acceptance Criteria–What are the minimum acceptable standards for speed, repeatability, accuracy and ease of use? How will these characteristics be assessed?
(o) Services–What type of training, documentation, installation assistance, warranties and customer service are offered?
In addition to answering these questions, prepare a set of good and bad parts for the vendor to evaluate. The good parts should span the range of appearances normally seen for acceptable parts; for instance, from the shiniest to the dullest.
Machine Vision Systems
A typical machine vision system includes these components:
(o) Charge Couple Device (CCD) video camera, lenses and lighting for acquisition of part images.
(o) Frame grabber for acquiring, digitizing, and processing images from cameras.
(o) Hardware and software for calculations and comparisons to nominals.
(o) Communications equipment for digital I/O or other devices.
(o) Software environment for a user interface.
Figure 1 illustrates an automated inspection system setup.
Lighting
Illuminating the part to be processed is the first and most important step taken when constructing an inspection system. With a good image, all the subsequent steps are easier.
The most common lighting techniques are directional, diffuse, structured lighting and strobing Figure 2. Directional light comes primarily from one direction and may be used to create shadows or reflections. It is partially collimated; a beam of light is collimated when all optical rays are parallel, like light from a distant source such as the sun (on a sunny day).
Diffuse lighting is noncollimated light, typically from many sources, or from a nearby large source, and minimizes shadows and harsh reflections. Sunlight on a cloudy day is an example of diffuse light.
Structured light is highly collimated and produces a point or line to establish presence, location, orientation or contour. Lasers are typically used to create structured light.
Strobes allow high-speed processes to be monitored by providing an extremely short burst of light, typically in the microsecond range. Strobing can be combined with some of the other classifications of lighting; for example, by sending the light from a strobe into a fiber-optic light guide or by pulsing an array of light emitting diodes (LEDs).
Cameras
CCD area and linescan cameras are two types of cameras commonly used in vision systems. Each camera uses a different approach for acquiring an image and cannot be substituted for the other in vision applications.
Area cameras are most commonly used for vision applications and there are many companies that manufacture them. The images are formed by exposing a 2-D sensor array to form an image like that of an ordinary video camera. The analog signal from the camera is converted into a digital image by the frame grabber and each pixel, or picture element, in the image is assigned a gray-scale value that indicates its brightness.
Most image processing and analysis packages have 256 gray-scale levels. Lighting is a critical component for accurate image acquisition for area cameras, and special techniques may be necessary to evenly distribute light across the part.
Linescan cameras use a single line of light sensors instead of a rectangular array. The camera is generally focused on a narrow line as a part moves past, and produces 10,000 or more lines of video per second. The lines can be processed individually or, as is more common, they can be stitched together into a 2-D image.
This camera is especially useful when inspecting rounded or cylindrical parts because, as the part is spun, its surface can be “unwrapped” into one long image, which then allows for easier inspection. In addition, lighting techniques are not as complex because you only have to light the one scan line of the part and not the whole part.
Inspection Systems
If there is a need for network or instrumentation connectivity, it is imperative that you purchase an inspection system that can support standard computer platforms. Due to the wave of platform standardization, vision systems now are available to support any computer architecture you prefer, such as 80486, 68040, PowerPC and Pentium.
These systems include standard off-the-shelf computer platforms like DOS and MAC or VME, and PC-based front-end systems that interface to programmable logic controllers, like Siemens and Modicon. Also available are proprietary architectures that support standard platforms by plugging into your PC or communicating to networks via Modbus Plus or other types of communications packages.
Standard off-the-shelf equipment supports a range of form factors. The three main platforms are DOS, UNIX and Macintosh. In addition, some vision suppliers have already ported their systems to support the new RISC technology on the PowerPC platform.
The UNIX-based platforms support Berkeley standards, and are based on the VME bus which runs in real-time UNIX. UNIX is typically used for nonstandard applications, and often requires customization due to high-speed inspection and resolution specifications.
Proprietary architectures which support standard platforms can take advantage of high-speed inspection by using array processors and digital signal processors to achieve guaranteed cycle times. These systems plug directly into a PC and have speeds of up to 60 parts per second guaranteed, regardless of how many inspections are occurring. For example, your application may require four computations per inspection or 4,000 computations, and each computation takes 16.7 ms.
Specialized pipelined processing allows for multiple computations to take place simultaneously, achieving guaranteed cycle times. These types of systems are popular in the electronics industry due to the simple nature of inspection and guaranteed inspection rates. Finally, PC-based front-end computers with a vision board to a PLC provide alternative means for companies who already use PLC technology.
All of these hardware options allow for connectivity to a networks like LANs and Ethernet via TCP/IP and Modbus Plus, as well as data communication via RS-232 serial ports. If a vendor you are considering cannot interconnect to your network or equipment, look for another vendor.
All automated inspection systems must also include processing board for image acquisition. Other boards, such as a servo control board, are required if integrated motion is a part of the application.
Many vision-guided inspection systems are used for SMD component placement. For example, a camera is attached to a robot and used to locate the printed wiring board while another camera is used to orient the component held by the robot gripper. Through precise calibration, the robot places the component on the board.
Software
The software of an inspection system is the brains behind the muscle. An excellent program will have a graphical user interface (GUI) that supports the standard click-and-drag format using dialog boxes and pull down menus for setting up inspection applications.
Vision software is complex, but it is no longer necessary to purchase a system which requires programming. The software must also support available hardware standards to ensure communication with other equipment and desktop software packages, such as spreadsheets. The GUI should support one of the three main windowing environments: Windows, Macintosh or X Windows/Motif.
A variety of image preprocessing and analysis operations are also required, depending on the specifics of the application at hand. To be useful, a general-purpose machine vision system must offer several options and computational capabilities such as:
(o) Area counting.
(o) Gray-scale analysis (pixel value computations).
(o) Connectivity/blob analysis (counting connected pixels).
(o) Edge detection (finding change in gray-scale values to determine an edge).
(o) Normalized correlation (pattern matching).
(o) Morphology (dilation and erosion of desired or unwanted features in images).
(o) Line and circle finding.
Figure 3 illustrates sample image processing and analysis operations.
Subpixel accuracy is another important feature to evaluate in software. Subpixel processing is a technique that results in a measurement with a resolution less than one pixel. Electronics applications often require subpixel accuracy because of minute tolerance levels and small regions of interest.
A region of interest is the area of the part where the inspection takes place. In controlled environments, you can expect accuracies up to 1/64 of a pixel in a standard 256-level gray-scale system. In practice, due to changing operational conditions such as sunlight and part variations (inconsistent part finish), expectations of around 1/10 pixel are more realistic.
If you have a very complex application, you will require a customizable inspection system. Many systems come in multilevel environments: a GUI for factory floor personnel and a programming environment for the process engineer.
For example, you may need a customized user interface, special I/O devices or special hardware. Such flexibility can be accomplished by creating customized routines with an interpretive language.
Services
The final step in the process of selecting a machine vision supplier is to ensure that the one you choose has the services you expect. One offering that is most valuable is a feasibility study. Suppliers can evaluate your application early in the process and provide written documentation on the best approach for your company. These studies are often free of charge.
Standard product warranties are available and range from 30 days or more. Many suppliers now offer additional services, such as full training classes as well as technical seminars on the latest technology. Some vendors really go the extra mile and provide regular newsletters as well as expanded upgrade programs to keep you posted on their latest developments.
Overall, the purchase of an automated inspection system is quite rigorous and requires some homework. First, fully evaluate your supplier and ensure that the company has proven experience in the industry. Secondly, the system you evaluate must be flexible and contain standard hardware and software to meet your connectivity requirements.
Finally, the software interface you choose will be the heart of your application. It must be easy to use and contain robust algorithms that have been developed over years of experience.
An automated inspection system chosen by these criteria will improve your productivity, reduce training time and, most importantly, improve the quality of your products.
About the Author
Scott Cole, a Senior Applications Engineer, has been with Acuity Imaging, formerly Automatix, since 1984. Previously, he was employed by Texas Instruments. Mr. Cole received B.S. and M.S. degrees in physics from the University of Florida, and completed engineering courses at the University of Texas. Acuity Imaging, Inc., 9 Townsend West, Nashua, NH 03063, (603) 598-8400.
Sidebar
Key Terms for Vision
Area Analysis–process of determining the area of a given view falling within a specified gray level.
Blob–a connected region in an image in which all pixels have the same gray-level value.
Calibration–translation of image processing data to real-world measurements.
Collimated–a beam of light in which all optical rays are parallel.
Connectivity Analysis–a technique for segmenting binary images based on pixel adjacency; for example, isolating pixel blobs.
Continuous Image–image not broken into its discrete levels of image fragments. A picture is a continuous image.
Edge Detection–finding change in pixel values to determine an edge.
Feature–any characteristic descriptive of an image or a region in an image.
Feature Extraction–process of generating a set of descriptors or characteristic attributes from a binary image.
Field of View–the area of the object imaged at the focal plane of a camera.
Flaw Detection–process of examining an object for unwanted features of an unknown shape at an unknown position.
Gray-Scale Processing–image enhancement operations that involve altering the gray-scale value of a pixel; must have a gray-scale image as its output.
Hough Transform–a global parallel method for finding straight or curved lines by mapping all points on a particular curve into a single location.
Image Analysis–process of generating a set of descriptors or features on which a decision about objects in an image is based.
Image Processing–transformation of an input image into an output image with desired properties.
Line Scan–a one-dimensional image sensor.
Mathematical Morphology–mathematics of shape analysis. An algebra whose variables, shapes and operations transform (usually filter) those shapes.
Normalized Correlation–a process whereby two image segments are compared to determine the similarity, or to find the position at which optimal similarity exists.
Pixel–an acronym for picture element. The individual elements in a digitized image array.
Region of Interest–the area inside defined boundaries that the user wants to analyze.
RS-170–Electronics Industries Association standard governing monochrome television studio electrical signals. Specifies maximum amplitude of 1.4 Vp-p, including synchronization pulses.
Subpixel Resolution–Any technique resulting in a measurement with a resolution less than one pixel.
Excerpted from the Automated Imaging Association’s Machine Vision and Imaging Glossary.
Copyright 1995 Nelson Publishing Inc.
March 1995