The Fundamentals: Getting Started with Machine Vision
AIA Posted 03/13/2000
The following tips on how to get started with machine vision come from Valerie Bolhouse, an Automation Systems Specialist with Ford Motor Company.
- Learn about the Technology
Before you even get started, you need a fundamental understanding of vision technology, what the basic functional clocks are, how they work, and what the limitations are. This information will help guide you down the right path for the following steps.
There are many good sources of information: textbooks, professional societies (Machine Vision Association of the Society of Manufacturing Engineers), trade associations (Automated Imaging Association), trade magazines and journals, and the Internet, to name a few. If you are able, attend a training seminar, conference, and/or exhibition. The cost will be a wise investment before you proceed to select and purchase equipment. - Identify Application
Get a handle on what the requirements really are for the application. Your job as a vision engineer is as much to identify where vision should not be used, as where it should be used. If there are processing problems causing a high fallout and management is asking for a vision system to sort defects, the processing problem will be solved long before you are able to install a vision system for inspection. Sometimes a mechanical solution for location is better than vision. Think of vision as only one tool in your tool box for manufacturing solutions. Once you are convinced vision is the right solution, analyze the application. You will need to know affordable cost for justification, technical requirements, precision or accuracy requirements, cycle time, etc. Draft a rough specification you can use to approach vendors, and accumulate a number of sample parts to take with you for explanation of task and evaluation of equipment. - Match the Technology to the Application
'Machine vision' encompasses a wide range of sensor and imaging technologies. You could employ color sensors, range sensors, infrared detectors, x-ray, or standard video cameras. The system could be a low-cost ($5,000 or less) smart camera, where the sensor and camera are in the same box , or a high-cost multi-processor computer. The trick is to employ just enough of the right technology to solve the problem. - Identify Vendors Who Can Provide Solutions
If the application has already been solved by a vision vendor, look there first. Also look in trade magazines for your industry to see what others are doing. Network within your industry for information. Use the Internet to search for key words that describe your technology. Look for commercial sites as well as searching the Patent Database (www.uspto.gov is one of several free database search sites) for suppliers who have developed solutions. Take your sample parts to a vision, sensor, or automation trade show and talk to vendors. - Feasibility Demonstration
Take parts to key selected suppliers to see how well they can image the part and solve your problem. The solution should be straightforward and obvious. Look for high-contrast images, tight distribution of measurement data, and good signal-to-noise ratios. The system needs to be repeatable - you can calibrate accuracy only if you have repeatability. A system that provides different answers on the same parts is useless. - Write the Specification
Make your expectations clear to the vendor. Explain what the system must do, not how it will do it. Be open to different ways to solve the same problem. What performance does the system really have to have in order to be acceptable to production? What level of false rejects will be tolerated? It is not practical to say that the machine will make no incorrect decisions. What would you do if it makes an error during runoff? Require the use of statistics for acceptance testing. - Request for Quotation
With a detailed specification in hand, go to three to five suppliers for a formal proposal. Be sure to ask for costs to support installation and launch. - Evaluate Responses
Use a decision analysis matrix to analyze responses. Price should be one of several factors in the decision. Confidence in the vendor meeting the requirements, ability to provide the support your operation requires, and your assessment of long-term maintainability by production are just as important as price. - Issue Purchase Order
- Project Management
If the system requires custom engineering, check on progress throughout the design and build phase. Keep the vendor supplied with a variety of sample parts for verification that the system can indeed handle the variance expected in production. - Machine Acceptance
Run the acceptance according to the specification. Do not accept the machine if it does not work on the vendor's floor. - Launch
Provide training for both engineering and production. If the production operators are involved have equity in the machine, they will keep it working. Otherwise, it goes into bypass. A well-designed system that solves a real manufacturing problem will be used and valued by production for the life of the product.
