Machine vision isn’t as challenging as it used to be. Setup, testing, and start-up can be easier and faster than prior technologies. Mark Hoske of Control Engineering talks about how two recent automotive case studies show value in recent, high-powered machine vision applications, as discussed at NIWeek 2016. A high-speed, clear, plastic bottle inspection also showed how processing power and smart software helped speed integration of the machine vision applications. The case studies were part of the Vision Summit, looking at high-speed, high-resolution continuous motion inspection.
Fast-moving web blurs defects
In production of automotive 20-ft-wide rolls of ceiling foam, human-eye detection in a 600-ft/minute web process was impossible, especially with defects in the foam or attached fabric as small as 0.020/in., explained Craig Borsack, P.E., president G2 Technologies. Human inspectors were missing tears in the foam longer than a foot.
The vision system integrator used a fourth generation automated web inspection platform, which evolved from reflective memory, local storage, to field programmable gate array (FPGA)-based, real-time image processing. The integrated technologies provided automatic defect detection, marking of defect areas, thickness detection, and other capabilities with off-the-shelf components, Borsack said. The FPGA handled image processing, acquiring, and the 20-ft-wide image, at 70 MB/sec bandwidth.
Multi-threaded process was used, with information going to the human-machine interface to apply a binary mask and edge detection, which is used for continuous width measurement; a second thread was used for image defect processing and another for image storage to a database for data analysis and recording of defect location and images of the defect, he explained.
The installation used nine cameras in 3 sets of three to provide unique orientations required to highlight the defects. Instead of defective rolls causing customer anguish, defects could be detailed and offered at a discount. Information allowed customers to easily work around the defects, Borsack said, at the right price.
Edge detection, pattern matching
FCC Indiana rebuilds clutches for Honda. Dirt and scratches are not part of the piece being inspected, and the prior vision system couldn’t always make that distinction, said Tanner Blair, embedded systems engineer with National Instruments.
Issues with the application included inconsistent lighting, metallic surfaces, dirt, grease, and scratches, Blair said. Also in the field of vision were varied rivet types. The customer wasn’t interested in changing the lighting and focused on changing the algorithms for image analysis. Within the field of interest, the inspection software focused on circular edge detection from the inside out, dark to light, ignoring a center hole, and providing an accurate inspection, he said.
Bottle thread inspection
There can be challenges with inspecting thread quality in molded, clear, plastic bottles, explained Vivian Le, application engineering with Graftek Imaging. Sometimes hanging flash can be mistaken for multiple threads.
The area of interest was an “S” measurement, which is an industry standard measurement for the distance between the top of the bottle to the top edge of the top-most thread. The bottle was backlit and imaged the full 360 degrees, with an image acquired every 10 degrees of rotation. Within the area of interest, five-measurement pattern matching was used. The challenge was that the thread shape changed with angle, and the top of the bottle looked similar to threading, she said.
Pattern matching was used so the software could learn the threads as the template, since view varied depending on rotation. Three pattern matches were used, and the software selected the top score of all matches. Rakes were used to detect the edge detection, looking at the top edge of the threads and the top of bottle, calculating the distance between two points. Continuous acquisition identifies features and dimensions, variations as small as 0.05 in., Le said.
>>Originally posted by Mark T. Hoske, Control Engineering, October 14, 2016