Nemak had been monitoring sand cores for its aluminum cylinder head and engine block casting operation using a two-dimensional, grayscale inspection system, because this is an effective way to track the interior formation of its castings.
Cores are made of compressed and bonded sand, and then carefully placed within a sand mold. Once molten aluminum is poured into the mold, the cores hold the spaces where cavities form in the finished casting after the aluminum solidifies. Inspecting cores is a more productive effort than inspecting the casting interiors, but after eight years of use Nemak was looking for an update for its 2D system.
Vision-system integrator Vista Solutions (www.vistasolutions.ca) suggested that a three-dimensional imaging system using laser triangulation could provide a more accurate inspection system at a fraction of the cost of the two-dimensional view, and it promised to ease timing constraints later in the production process by inspecting cores “on the fly,” rather than having a robot stop to present the part to the vision system.
3D development
It’s widely believed that 2D inspection is cheaper than 3D approaches, but that overlooks the capabilities of 2-GHz microprocessors and low-cost field programmable gate arrays (FPGA), which put the power of a mainframe computer into handheld devices.
“Nemak contacted us about a little over a year ago and said it was time to update their core inspection system,” explains Dean Scarlett, vice president at Vista Solutions. “There were a handful of other groups all specifying the latest cameras in another four-camera configuration. We contacted the purchasing manager at Nemak and said that we were declining to quote because we believed that grayscale was not the right approach to solve this problem.
“But,” Scarlett recalls telling the Nemak officials, “if you’re open to suggestions, let’s look at different solutions.”
Vista Solutions is “vendor and technology agnostic,” but the Windsor, ON, group had recently achieved authorized integrator status with Sick AG (www.sick.com), a well-known maker of the IVC-3D integrated vision camera with laser triangulation.
According to Vista v.p. Mike Sirizzotti, “we told Nemak, ‘If you really want to inspect this core, you need volumetric defect information, and not just area.’ So, we obtained some sample cores, and we knew as soon as we’d seen a screenshot (from the IVC-3D) that this was the right way to approach the problem. We set it up on our own robot, worked through the problems and algorithms associated with inspection, and they virtually committed to giving us the order on the spot.”
Pass/fail inspection
For about 10 years Nemak had been inspecting cores using four analog cameras to capture four separate images that revealed 12 breather tabs on one side of a core. In this system, a robot would extract the core from the machine, position it in front of the cameras, stop momentarily while the images were captured and collected, and then pass the core along. Depending upon whether a core failed or passed its inspection, it would be trafficked either to a reject conveyor or on to a conveyor leading to the aluminum casting line.
Because the six-axis ABB robot had to stop and hold the core in front of the cameras, the inspection routine imposed a few extra seconds into the coremaking process while the handling device decelerated, stopped, and accelerated again.
Each breather tab measures approximately 1 cm in diameter and less than 2 cm tall. In order to produce an acceptable casting, Nemak’s manufacturing process requires only that a majority of the tab be present to achieve a good aluminum part. The older system, and all of the other proposals Nemak fielded, suggested upgrading the camera technology but staying with grayscale imaging and geometric pattern searches, to identify key features that indicate the presence of a breather tab.
The Vista Solutions proposal also called for a static inspection station served by the ABB robot. A Sick IVC-3D 200 camera was mounted just to the side of the robot path. The IVC-3D 200 holds a CMOS sensor offering a maximum profile width of 1,024 pixels and a red diode laser-line generator in a single housing. A laser line is directed straight at the object under inspection while the CMOS camera is placed slightly off the optical axis.
In this configuration, displacement of the laser line in the acquired image indicates differences in height. A bandpass filter placed under the IVC-3D 200’s protective window filters out wavelengths other than those emitted by the laser-line generator. This reduces the need for additional vision-specific lighting, as well as the system’s footprint and the vision system’s component checklist.
The camera also houses an 800-MHz microprocessor with FPGA acceleration. According to the system developers, this makes it possible to process images entirely at the camera head, using Sick’s IVC Studio objectbased image-processing library, or to send the image to a separate processing station.
The new approach
Once the robot retrieves the first core from the mold, the controller sends a query signal across a relay to one of six programmable I/O ports on the IVC-3D 200. If the system is ready for inspection, a signal is sent to the robot controller for the process to begin.
Then, the robot takes the core from the machine and moves it across the camera’s field of view (200600-mm), at a distance that allows the full width of the core to be captured from a single image.
Keith DeGroot, Vista Solutions manager of application engineering, reports that in designing the system a important consideration was whether or not to use an encoder to track the position of the core as it moved in front of the camera. “We determined that the ABB robot could pass the part at a constant speed in front the camera, so we could using timing to create the 3D surface map, rather than physical position using an encoder,” DeGroot recalls.
The camera builds a 3D surface map by combining “profiles,” individual images that reveal the position of the laser line. The camera collects up to 5,000 profiles per second, processing them in real time on the FPGA and microprocessor, and compositing them into a continuous 3D surface map of the core’s surface. On the map, height is displayed as intensity — in the case of grayscale display — or as color, in the case of false-color surface map display.
In addition to the six discrete, programmable I/O ports, the IVC- 3D 200 offers an RS-422 interface for encoders, and a RS-485 output for PLC communication. The latter wasn’t needed for Nemak’s application (the ABB robot is in the workcell), but the camera there includes a 10/100 Ethernet port and internal web server. DeGroot says he connected the IVC-3D 200 to the plant’s local area network and set up the camera to store result data and images on the network’s FTP server, for remote viewing by plant personnel, and debugging by Vista Solutions, if necessary.
“We can save images on the plant network, but we typically save text files,” DeGroot explains. “We can dial up the IP of the camera and view the latest images, and results for each task, whether it passed or failed, etc. The system itself doesn’t have a monitor or keyboard. We program the camera by taking a laptop to the camera and uploading the program developed in IVC Studio. You can leave the laptop there, if you want to display the camera in real time, but it’s not necessary.”
Moving to a scanning mode called for some robot reprogramming, because the ABB device no longer had to present the part at the inspection station. However, this allowed the robot to move more quickly, and directly from the press to the output conveyor.
Sirizzotti says the time savings was not significant for coremaking productivity, but helped to ease some production restraints later in the casting process.
Although the combination of optics, camera field of view, and size of the core, established a minimum spatial resolution of 0.2 mm for inspecting the breather tabs, Nemak ‘tweaked’ the threshold to catch larger defects in the tab structure, which improved the system’s throughput.
In tests, the Vista Solutions/Sick system showed a total effectiveness of 98.17% on 10 parts: six were good parts, four had known faults. None of the errors allowed failed parts to pass. Similar grayscale systems generated 85% effectiveness, and passed parts that were known to be defective.
“During our trials with the 3D system, the false trials were attributed to multiple handling of the parts that caused them to fail on their third trial,” explains says Nemak’s lean manufacturing coordinator, Mandy Karnib. “In short, the 3D laser system has been extremely beneficial in allowing us to significantly reduce our breather-core scrap, both internally and externally.”
According to the system integrators, the installation has been so well accepted that it may lead to further applications. “We’re in talks with Nemak to look at deploying more ‘black box’ solutions based on this architecture at other Nemak facilities,” says Dean Scarlett.