1 September 2006
Machine vision finds more uses from automotive to nuclear inspections
By Brian King, Anand Krishnan, and Ellen Fussell Policastro
While the automotive and semiconductor industries have long been the visionaries behind machine vision technology, primarily to achieve their high production output, a new and growing user base is beginning to see the light, realizing the value of this technology on the production floor.
The increasing demand for vision technology in manufacturing is leading to shrinking electronics, better imager technologies, and faster networks, “satisfying today’s mantra of faster-cheaper-smaller-better,” said Himanshu Shah, a senior analyst with ARC in Pittsburgh. “Today’s machine vision technology is critical to supporting manufacturers’ business initiatives. With the expanding requirement for traceability and genealogy solutions, it plays a supporting role in regulatory compliance in many industries. Direct parts-marking identification and reading technology are some fundamental machine vision applications manufacturers are fast adopting,” he said.
When plant personnel shut down and disassemble a nuclear reactor vessel, technicians usually check the inside surface of the head for indications of wear. When they find one, they need to measure it and record its size and position. Today, vision systems are helping manufacturers improve quality and productivity by automating these inspection tasks. Industrial Eye, a vision integrator in Waukegan, Ill., collaborated with PCI Energy Services LLC, a subsidiary of Westinghouse Electric Company in Lake Bluff, Ill., to develop a vision system for the nuclear industry to help technicians locate and measure indications of cracks more efficiently, while viewing them from a monitor and controlling camera head motion via remote console.
“It’s important measurements go quickly, but reliably. But it’s even more important to eliminate the chance of human-sighting errors and reduce strain on technicians,” said Mark Pillard, PCI’s manager of design engineering. The new system reduces radiation dose levels by 75% per repair from 1.6 Rem for manual inspections, to 400 mRem with the remote system, Pillard said.
A camera with an intense UV headlight mounts to a custom robot that scans the interior of the vessel for traces of phosphorescent dye that penetrates cracks and pits in metal surfaces. The dye makes the indications visible. The inspecting robot locks onto the bottom end of vertically suspended tubes (also called penetrations) inside the reactor dome with one arm, and it moves the camera and headlight over the entire surface of each penetration to inspect for fluorescing dye. There are nearly 80 penetrations per vessel. Once it detects the dye, the camera determines its size and shape, storing images of indications on a hard drive for archiving.
Replacing human inspections
Technicians used to perform these types of measurements by hand with a 1/64-inch gradation ruler from underneath the 12-ft diameter hemispherical dome, the reactor vessel head. The manual procedures were tedious, cumbersome, error-prone, and exposed technicians to radiation. Possible defects could include cracks in welds at the intersection of tubes or penetrations with the vessel wall, or pits in the vessel wall.
To inspect these vessels, humans enter the radioactive vessel individually to analyze the walls in heavy protective suits (chemical clothing, encapsulating suits, facemasks, goggles, and gloves, which restrict movement, coordination, manual dexterity, vision, and also hinder the technician’s ability to maneuver). If they fail to find a fault during the inspection, a catastrophic failure could occur in the vessel when it undergoes re-commissioning. The reason could be metal fatigue or degradation that would bend or break a penetration, which in turn could result in a malfunction in the reactor cooling system.
Further complicating the operation is an inverted forest of steel pipes, control rod drive mechanism (CRDM) nozzles. About 80 of these 4-inch diameter by 2-foot-long tubular obstacles, spaced nearly 1 foot apart in a series of concentric honeycomb patterns, hang down from the inside surface of the dome above the technician to obscure line of sight and impede access. All of these challenges slow down the process and increase the time it takes for technicians to do their jobs. “People can be safely exposed to certain dose levels of radiation,” Pillard said. But rotating personnel and using monitoring systems ensures no one is overexposed.
Dual cameras (a standard color camera and a remote-head camera of the vision sensor) display image data simultaneously on a color monitor near the control console. The operator views the indication in real time on the monitor. The operator uses a mouse to align cross hair cursors at key points on the indication to measure it so the operator can measure round, oval, linear, or dog-leg indications of various sizes and shapes. The vision system is linked via Ethernet to a hard drive to allow automatic recording of time, date, location data, and size of each image.
With combined data stored on a hard drive, technicians have instant access via computer to statistical data on each reactor vessel head. “This makes returning to indications to take images later for comparison a lot faster because getting the information we need from a database is more efficient than coping with paperwork,” said PCI’s senior design engineer, Tony Mastopietro.
A small remote head vision sensor is a great choice for use in hazardous environments because the CPU containing the sensitive electronics can mount a safe distance away from the camera. “In order to navigate through the forest of nozzles and access all the locations requiring inspections inside the dome,” developers had to look for “the smallest camera we could find,” Pillard said. A compact head assembly can help navigate between the CRDM nozzles, allowing easier access to the recesses of the reactor head. You should seal the compact head assembly against moisture, and it should protect the two cameras and optics from mechanical damage to ensure reliable operation. This protection from hazards and moisture is important because in order to inspect the reactor vessel head, you should first treat the inside surface with an ultra-violet sensitive penetrant die. The result is easily detectable indications; the die soaks in and stays after the excess rinses off the rest of the surface.
It’s important to deliver fluids remotely when cleaning, applying the dye chemicals, and rinsing the surface after an inspection. A debris enclosure and collection system can retrieve fluids used in this process for disposal. When exposed to UV light, the die fluoresces to help technicians quickly locate and accurately measure the indications.
Spark plug inspection
The hallmark of a successful measurement application is its reliability and repeatability, an even more difficult goal in a vision-based application. Individual hardware components and the software that ties them together need to be reliable and repeatable. Companies such as Ford, GM, Daimler Chrysler, and Boeing take advantage of developments from machine vision technology company, Soliton Technologies, in Bangalore, India. The job of a vision system engineer is gradually evolving from writing and debugging thousands of lines of code to architecting creative ideas for solving image processing problems. One design for an automated optical inspection system for automobile spark plugs, allowed developers to meet Six-Sigma standards of reproducibility, repeatability, and accuracy while also saving floor space.
Proper sparking distance is critical in the production of an automobile spark plug. Two key dimensional parameters affecting the functional performance of the spark plug are eccentricity between the outer housing profile and the inner electrode and the offset between the earth electrode and the center of the inner electrode.
Regardless of form factor, basic components are common to most vision systems, including a camera, processor, image analysis software, lighting, and networking capability.
In a typical gauging application, a camera mounted above or to the side of a part captures an image of the part to be measured as it enters the field of view. You can then analyze the image using gauging software tools, which calculate distances between various points in the image. Based on these calculations, the vision system determines if the part dimensions are within tolerance. If dimensions fall outside of tolerance, the vision system sends a fail signal to a controller (such as a PLC), which in turn triggers a reject mechanism to eject the product from the line.
If users have a manual process for measuring eccentricity and offset, they would need to find the position of three points on each circular part and fit the circle equation using this data. A measurement error on even one of these points could significantly skew the calculated eccentricity and offset. The lesser accuracy of this method could force a user to narrow their tolerance band, thus decreasing yield.
To ensure reliable quality control, faster inspection, and good return of investment, the vision-based dimensioning system for the spark plug production line can measure eccentricity and offset to an accuracy of 0.01mm with a six-sigma repeatability of better than 10% of the tolerance value. In the initial design, developers wanted to ensure system functionality was independent of the inherent variations in plug texture from piece to piece and model to model. It was also important the product minimize floor space with a resistance to harsh production environments. The system needs to also interface with the line PLC to remove human intervention, thus making the measurement process quick, objective, and repeatable. And it should store records of results for further analysis and improvement in process and yield.
The vision manufacturer selected a 1280 x 960 pixel digital camera with programmable features required for this application. Writing the application with real-time software gives users the flexibility to configure the system via TCP/IP, periodically calibrate the system using a calibration target, and transfer results data through FTP. A compact vision system served as the platform on which this software would run, keeping in mind its real-time performance, ruggedness, and small size. True to developers’ expectations, the result was a zero-manual intervention system with a gage repeatability and reproducibility of less than 10% of the tolerance band.
Because the application required repeatability of two pixels, it was important to design algorithms so the tuning parameters were in the middle of a wide tolerance band. This ensured changes in the image due to camera and imaging repeatability did not affect algorithm results. During the algorithm prototyping, a vision assistant tool helped cyclically inspect stored plug images and tabulate intermediate results using a batch processing feature. So parameter tuning became systematic and goal-oriented as opposed to a trial and error approach.
About the Authors
Brian King is an engineering manager at Industrial Eye, a vision integrator in Waukegan, Ill. Anand Krishnan is a product development team leader at Soliton Technologies, a machine vision service provider in Bangalore, India. Ellen Fussell Policastro is assistant editor at InTech magazine. Her e-mail is email@example.com.
Tips for successful machine vision application
Sight System Mimes Housefly www.isa.org/link/Housefly
How to Apply Machine Vision (SP30P) www.isa.org/link/SP30P
Vision Precision www.isa.org/link/VisPrecision