Process Photonics (Ottawa, ON, Canada) builds processing systems for the PCB, electronics assembly, and medical device markets. Their ProVisionTM series is a solution that uses vision to inspect panel and sheet-based circuit features. The system concept was devised after a customer expressed interest in replacing multiple existing inspection machines. The customer wanted a unit that was capable of significantly higher resolution and throughput, but also wanted to plan for a system that could perform defect recognition tasks beyond the original system’s capabilities.
ProVision systems combine vision with integrated material handling automation for high-volume production. With multiple cameras operating in parallel, ProVision acquires images at a resolution better than 5 microns, and implements algorithms to make precise measurements. The machine architecture allows for real-time image acquisition, processing, and analysis of multiple parameters per object, at over 60 parts per second. The system is scalable to higher resolution and throughput.
ProVision’s components include PPI Linear XY motion stages, five DALSA Piranha HS-40-04k40 cameras, three Matrox Odyssey XCL vision processor cards, Schneider lenses, Gardasoft Vision PP861 LED Lighting Controller controls (5 CCS UV LEDs), USAF 1951 calibration targets, automatic pick and place sheet loader that puts sheets onto a vacuum platen, WAGO Ethernet IO controller, Datalogic barcode scanner, and an Aerotech motion controller.
Screen-printed medical sensors are arranged on the sheet in 20 columns and 40 rows. The five bidirectional TDI line-scan cameras are spaced four columns apart. The sheet is scanned in four passes, so the first camera sees columns 1, 2, 3, and 4; the second camera sees 5, 6, 7, and 8; and so on. The odd-numbered columns are scanned from top to bottom, and the even-numbered columns are scanned from bot- tom to top. The three Matrox vision processors are connected to cameras 1 and 2, 3 and 4, and camera 5, respectively. At the sheet’s Y-axis, an encoder provides a quadrature signal through a custom-designed circuit that triggers each Matrox card to acquire a line every 5.1 mm.
At the start of each scan, the boards begin grabbing small frames into a circular buffer. Every time one of these buffers is filled, a call-back function copies a portion of the image into another buffer — one that is large enough to store the entire image of the sheet and wide enough to contain the ROIs (Regions of Interest). Processing threads wait until enough image data is acquired before processing the next sensor location.
At the start of the first scan, each camera locates a fiducial and measures the grayscale color of several swatches on the sheet. The fiducial location is used to adjust the sensor ROI locations, and the grayscale measurements are used for determining binarization thresholds. The image processing performs several binarization and blob analysis operations to prepare the images for the measurement operations that are performed on the sensor geometry. In order to ensure measurement accuracy, images of USAF 1951 calibration targets are acquired and measured at the start of every automated inspection task. A com- bination of blob analysis and marker measurement functions was used to determine the vertical and horizontal pixel scaling factors of each camera.