Manual inspection of a 4-inch wide, 8-inch circumference elastomeric band is difficult because there is no objective way for a human inspector to determine what is or is not a defect. In general, defects are areas that are darker or lighter than the surrounding material. Invotec Engineering, Miamisburg, Ohio, USA, was asked by a company that produces the bands if machine vision could be used to define an objective specification and produce repeatable inspection results. Working with the customer, Invotec engineers developed a specification that considers both the size of a light or dark area and the degree to which it contrasts with the surrounding material.
The method was automated with two Cognex In-Sight® 5000 series vision systems that completely inspect a band in 16 seconds. The systems capture several images of each section of the band. The first few images are used to adjust the gain to compensate for possible thickness variations caused by stretching of the band as it is rotated in front of the cameras. “We developed an objective specification for defects based on machine vision parameters and demonstrated the ability to repeatably inspect bands based on that specification fast enough to enable 100% inspection,” said Daryl Greywitt, Vice President and Technical Director for Invotec.
Challenge of ensuring quality of critical component
The elastomeric band is used in a medical device, making it critically important to reject bands with pin holes, thin spots, inclusions, and embedded particles. Previously, these parts were manually inspected by backlighting the band and rotating it on two spindles while an operator looked for defects. The nature of human vision means that there are certain substantial flaws that every inspector agrees is a defect. But there are also many minor flaws or variations that some inspectors would consider a defect and others would not. Many inspectors might even pass a part with a certain minor flaw on one day and reject it on the next day.
The producer of the bands asked Invotec to get involved because of the company’s extensive experience in machine vision applications. Invotec’s 60 engineers, technicians, and toolmakers provide custom designed assembly, inspection, and test systems as well as fixtures, tools, and gages to manufacturing companies in many different industries.
Invotec engineers chose Cognex’s In-Sight vision system as the platform upon which to develop the application. “In-Sight vision systems provide everything that is needed to develop a powerful vision application in a compact package,” Greywitt said. In-Sight provides access to the extensive Cognex vision tool library that handles nearly any vision application. These tools have the intelligence to ignore non-critical variations in appearance while focusing on the critical features that determine a product’s acceptability. They don’t require extensive image pre-processing to achieve the final solution, adding faster application development for the vision engineer and lower lifecycle costs for the company. Drag-and-drop linking between tools enables quick communication of values, results, and images.
Developing the vision application
During the prototype phase, Invotec engineers worked with customer engineers to mathematically define each type of defect and develop the lighting and vision tools needed to consistently identify defects. It was determined that multiple inspection processes were needed to characterize the product. “When looking for field defects such as inclusions or thin spots, the system applies two histograms to look for areas of high contrast,” Invotec Controls Engineer, Mike Updike said. “One histogram looks for dark areas and one histogram looks for light areas. If no such areas are found then we assume that the region is free of defects. If an area of high contrast is found—either light or dark—a regional blob tool is applied to find the potential defect. The blob tool uses information from the histogram to better classify the size of the defect. Once the suspect area has been localized with the regional blob tool another histogram tool and a defect specific blob tool are applied directly to the potential defect. The output of these last two tools is a contrast level, defect size, and defect location. These three values are compared to the preset thresholds that define a reject.”
When looking for edge defects the camera applies ten edge tools perpendicular to the edge of the membrane in a small region. These tools each output a numerical reading of the position of the edge. As the membrane passes under the camera, the camera takes the average of these readings to find the edge. To find edge defects, each of the individual readings are compared to the average and if they vary by more than the preset limit the area is flagged as a defect. The system’s variables for defects are: contrast limit; light, size limit; light, contrast limit; dark, size limit; dark, edge variance limit positive; and edge variance limit negative.
Invotec used an Advanced Illumination red backlight to illuminate the membrane for imaging. This light generates 0.550mw/cm2 in the 660nm wavelength and performed well in testing by creating good contrast with the pigment in the band. Two vision systems provide the necessary inspection resolution while imaging the full width of the band. The first vision system incorporated into the set-up is an In-Sight 5403 with a resolution of 1600 pixels by 1200 pixels. This resolution makes it possible to accurately inspect roughly 60% of the width of the membrane. Invotec added a Cognex In-Sight 5603 vision system to cover the full width of the membrane in combination with the In-Sight 5403. The In-Sight 5603 has the same resolution as the 5403, but its processor is roughly twice as fast. Two In-Sight 5603 cameras have been incorporated into a production system to minimize cycle time.
Moving to the production environment
Invotec developed a manually loaded and actuated prototype fixture as a first step toward moving the application into the production environment. This fixture holds the band with a pair of one-inch diameter rods. These rods are spaced to place the part under a small amount of tension. The front rod is manually rotated to present the membrane to the camera. The rear roller has a rather high rotating friction. This applies some additional stretch to the membrane as it stretches slightly before the rear roller begins to rotate.
A difficult challenge that arose in making the application production-ready is the way that the band stretches on the spindles used to move it in front of the vision system. As the band stretches, it alters the thickness of the portion of the band in the field of view of the vision sensor. Invotec engineers recognized the importance of compensating for such alterations in thickness in order to provide repeatable inspection results. They accomplished this goal by dividing the field of view of the camera into three sub-regions that are parallel to the direction of travel of the band.
When an area of the band moves into the first sub-region, the vision system performs an inspection of a section of the band and also measures the average brightness of the background of the band. When this section of the band moves into the next sub-region, the gain of the vision system is adjusted to compensate for the brightness of the background and another image is taken. Average brightness is measured again in this image and the gain is further adjusted when the section of the band moves into the third sub-region. The result is that the vision system provides consistent inspection results regardless of how the band may be stretched as it moves around the spindles.
“This machine vision application has demonstrated the ability to, for the first time, provide an objective specification and repeatable inspection of the elastomeric bands,” Greywitt concluded. “It allows users to set any desired size limit for defects and delivers repeatable pass/fail results even when the size limit is set as low as 0.2 square millimeters. The vision system has demonstrated the ability to improve inspection consistency while reducing inspection costs.”