I want to use machine vision to measure color variation and perform real-time control on our paint nozzles. I’m not an expert, so I need a machine vision system that will be relatively easy to design, install and maintain. Can I do this myself with assistance from the right vendor, or should I look for a system integrator that specializes in machine vision?
—from November ’07 Control Design
Learn From the Integrator
[This comment was posted on Machine Builder Forum at ControlDesign.com/MBF]
This is a tough question, because it depends on your experience. If this is your first vision install, then use an integrator, ask for a guarantee of performance, and learn all you can.
In general, here are some things to watch for when “measuring” color.
- Lighting is more important than any other component, so make sure the lighting is consistent in your operating environment.
- Define how much color shift is allowed, how much is a fail condition and how you will test this vision system to see if it can detect your defined color differences. Can you produce “bad” samples at the edge of good and bad?
- Get your quality people to sign off on a standard. This is tough, but it will save countless days of testing during startup.
- Finally, you’ll want to do some homework on whether or not you need a relative color measurement or an absolute color measurement. There are ways to accomplish each, but the hardware and software are very different and not usually interchangeable.
GLENN ARCHER, director of business development,
Epic Vision Solutions
Not an Expert? Go With the SI
Cognex recommends that “non-experts” attempting to use machine vision for real-time process control, rather than simple pass/fail inspection, employ an integrator that has a successful track record of integrating similar systems.
JOHN LEWIS, public relations manager,
Seeing is believing
CVS Monitors RGB
A number of variables must be considered. It’s important to know whether problems occur when the colors change from one product to the next, or if this application requires continuous color data with multiple inspections of one signal unit. Since the options vary in complexity and overall ability to distinguish differences in color, it’s crucial to know the reason for the additional inspection.
Since the customer has asked for a simple solution that could be installed and maintained by people in-house, a color vision sensor with an analog or serial output connected to a PLC would be ideal for this application. The CVS monitors the RGB values and triggers an output if the reading varies too far from the desired color. Plus, the CVS has its own output based on a taught color. If the color fades or changes too much, the CVS can directly signal a paint controller to stop spraying.
JIM ANDERSON, machine vision product manager,
One Step at a Time
The application you describe can be done with assistance from the right vendor—one that offers machine vision testing lab facilities and can provide automation integration services. You should expect several things from a competent supplier.
First, to ensure the success of your machine vision application, an initial site survey is needed. This helps document all aspects of how your inspected object arrives at the inspection, how it’s presented to the camera, space limitations of the inspection area and speed of the line. It provides a reference to the critical parameters involved. Most vision suppliers ask 50 or 60 questions to simulate your operation and prepare an accurate assessment for proof of concept prior to recommending a system to test on your line.
Second, along with the site survey, you can help your supplier by providing a video and/or still photos of the process. Provide samples of good and bad pieces, identified with the problem in cases of bad pieces, to evaluate at proof-of-concept. During proof of concept, your supplier verifies the required elements necessary in the inspection process, identifies the illumination requirements and confirms the system’s capability to meet the specific application demands. After that, engineering identifies associated elements necessary for the inspection to generate a bill of materials and quotation for the solution.
Third, your application is highly dependent on proper and consistent illumination. Expect to be asked to identify QA/QC standards for color, how you implement them, and how you adhere to stipulated standards such as the color temperature of lighting, acceptable tolerance ranges and the frequency of standards testing to ensure lighting and paint composition are within spec. They also should ask about consistency of object presentation relative to the cameras. For example, the various shapes of car doors and body panels may require different areas of inspection and lighting to ensure accurate inspection of paint color.
Fourth, to integrate the inspection system in your control system, your supplier should ask about system dynamics and process lag, the plant’s environment, compliance with required protection or safety standards and adherence to preventive maintenance guidelines.
Finally, some new vision systems integrate a touchscreen terminal into the controller to provide real-time images from the camera with displayed inspection results. They use interactive menus to provide assistance for lighting, filtering and automatic setting of parameters. The result is simplified setup, monitoring and changeover, so operators without extensive machine vision expertise can maintain operations.
TOM KAHN, vision/auto ID product marketing manager,
Don’t Forget the Nozzle Adjustments
Whether you choose to purchase full turnkey vision integration services or take the time to learn and apply a solution yourself, you’ll need to define these parameters.
- What degree of change of shade or tint level do you want to trigger a signal to adjust the nozzle? Pictures might be a simpler way to derive and communicate this, rather than color data.
- How many nozzle adjustment levels will you need to make to keep the color consistent? This helps define whether several discrete color sensor outputs will be efficient or if Ethernet or serial communications will better handle continual, minute nozzle adjustments.
- The physical space needed on the machine to mount the color vision sensor and controlled white light(s), so the paint spray is at a constant distance from the sensor.
Assuming the above application specifications are defined and within the scope of a color vision sensor, here’s an estimate of performance and costs to expect with our color vision sensor and lighting—including a day of setup and training service. The sensor will position its average color tool on the paint spray and sense the hue, saturation and intensity, or the red, green and blue values, roughly 10 times per second. The sensor can communicate these values through its Ethernet or its RS232 serial outputs, or be configured to have each of its four discrete outputs change state at specified color values. The sensor can store and recall more than 100 recipes to sense different paint colors. Expect to spend more than $5,000 for the sensor and software, lens, light(s) and cables, and around $1,000 for a day of on-site assistance configuring the sensor and training for maintenance and troubleshooting. You’ll need to budget your own time to design, mount and connect the sensor and light(s) to 10-30 Vdc power, as well as design and write any software for controlling the nozzle. You’ll also need to have a Windows O/S computer (with Ethernet card) for setting up and troubleshooting the sensor.
JEFF SCHMITZ, business manager-vision sensors,
Three Steps to Measuring Color Variation
First, measure the paint color variations. If we assume the paint is a consistent color and texture at the point of measurement, this can be done with a color camera and some relatively easy measurements. If the paint is still being mixed, giving it multiple colors at once and possibly different textures, this task is less straightforward. If you are interested in what the paint is going to look like to a human eye, you may want to use some advanced measurements using the CIELab color space.
Next is the control, which also has many variables. How exact does this control need to be? Are you mixing paint for a fence at a hardware store, or is this for an application where slight differences in color are going to make a big difference? How fast does this control need to be? Low-speed control can be done with simple on/off signals sent via one digital I/O line. For faster applications with multiple nozzles, you might require a real-time system backed by an FPGA for high-speed, deterministic control.
Lastly, what are you going to do with the data you collect with the system? If your only purpose is to control the system and you have no desire to have feedback to other systems for, perhaps, some SPC down the road, this part can be disregarded. Most people will want to store process variables to a host PC or a server. Also, you might need to make changes on the fly or need a simple pause button for when your paint source is empty and needs to be changed.
Just how comfortable are you with each of the above pieces of the puzzle? If the inspection simply detecting changes in contrast between different shades of paint and the control is as simple as sending pulses on a single digital line, a color camera and a simple vision system might be the only hardware you need and the application can be configured in an interface like NI’s Vision Builder for Automation Inspection quite quickly and easily. If it’s a much more advanced application with many variations in paint conditions and high-speed control of the paint nozzles requiring multiple pieces of automation and control hardware (and potentially many software APIs, as well), in this case, you might want to call your local integrator.
product marketing engineer,
What is the best way for our customers’ techs to interact with our machines via handheld devices? Should we be using touchscreens, self-contained keyboards, pointing devices or a combination of all three? What are the pluses and minuses of each approach?
SEND US YOUR COMMENTS, SUGGESTIONS OR SOLUTIONS FOR THIS PROBLEM. We’ll include it in the April ’08 issue, and post it on ControlDesign.com. Send visuals if you’d like—a sketch is fine. E-mail us at RealAnswers@putman.net. Please include your company, location and title in the response.
HAVE A PROBLEM YOU’D LIKE TO POSE to the readers? Send it along, too.