Medical products use PC-based machine vision to speed inspection
R. Winn Hardin
Although mostly associated with the food industry, product grading plays a more important role for the pharmaceutical industry, where contaminants could cause tragic consequences from injected fluids or ingested pills. Building on years of knowledge from working with industrial
partners, such as Allen-Bradley Co. (Richland Center, WI) and Seidenader Maschinenbau (Kirchheim/Munich, Germany), SVResearch (Harrisburg, PA) has developed a machine-vision product-grading module based on Intel Corp. (Santa Clara, CA) dual Pentium II processors working in a Microsoft Windows NT environment. By incorporating robust microprocessor technology and limiting frame-grabber operational dependence to image acquisition, this module achieves fast imaging analysis and offers reconfigurable upgrading with minimum US Food and Drug Administration (FDA) and Conformite Europeene (CE) Mark revalidation--a time-consuming and costly responsibility imposed on the pharmaceutical industry.
"The pharmaceutical industry has strict requirements for machine processor and software validation with strict system change requirements and procedures," explains SVResearch president Ron Lawson. "If you have a system that is currently functioning and need to add one more cameras or frame grabbers, you can`t do that without modifying the software. It would then require [system] revalidation and testing. By using a dedicated four-channel asynchronous [frame grabber] card, we get a dedicated upgradeable module that doesn`t have a CD or floppy-disk drive for software upgrades. And every time the Pentium processor is improved, we get its system improvements without changing our application software, which has already been tested and validated. All we have to do is revalidate the hardware," he adds.
The key component of the machine-vision product-grading system is the Seidenader vision inspection module (SVIM) from SVResearch. This processing module needs no additional software or programming and is configured via a point-and-click interface. All field-wiring connections are available at the module`s front panel via a digital input/output (I/O) board plugged into the SVIM backplane. Optically isolated I/O connectors allow access to camera trigger, strobe controls, eight 5- to 30-Vdc input lines, and sixteen 5- to 30-Vdc output lines. Other front-panel connectors are available for the cameras, RS-232 serial ports, SVGA monitor, power, keyboard and mouse, Ethernet port, universal serial bus port, and a parallel port. "In this way, the signals do not have to come through D-shell-type connectors, which are not [industry] CE Mark-qualified for these voltages," Lawson says.
As many as eight SVIM modules can be networked to a single host PC, which controls the operator-interface display monitor and a processor video switch (see Fig. 2). The embedded Windows NT operating system in the module provides symmetrical processing for two 350-MHz Pentium II vision processors and a compatible environment for a frame grabber from Coreco Inc. (St. Laurent, Quebec, Canada), software tools from Matrox Electronic Systems Ltd. (Dorval, Quebec, Canada), and SVObserver front-end software from SVResearch.
According to Lawson, the company`s front-end software is a fully configurable operator interface that draws upon the strengths of other software libraries without the need for C-language programming knowledge. System configuration is accomplished by merely "pointing and clicking," providing the system operator with fast access to a range of stored procedures and checks. These Windows-based operations include image and blob analysis, edge detection, and feature measurements; logic and math functions; and image-manipulation techniques such as Laplace filters, rotation, image subtraction, scaling, masking, rotation, thresholding, and 3 x 3 or 5 x 5 kernel definition.
A typical machine-vision platform uses a PC host connected to one or more SVIM modules through a customer-specified (10-Mbit/s) Ethernet port, or, in many cases, via the Allen-Bradley Data Highway Network. Each SVIM module connects the frame grabber in the vision-processor module to the color or monochrome cameras at the inspection points.
Shielded-pair wires and coaxial cables interconnect the trigger signals for lighting, cameras, and the programmable logic controller (PLC). This controller accumulates pass-fail product counts and communicates back to the system PC via the SVIM module for archiving and reporting.
Using this architecture, Seidenader Maschinenbau has installed several SVIM modules in machine-vision systems to check multiple pharmaceutical-inspection parameters, such as cosmetic type and content, and multiple product lines, such as for examining capsules, vials, syringes, tablets, and ampoules. A recent installation involved a machine-vision system for inspecting pharmaceutical vials (see Fig. 2). In this application, the SVIM-based system checks for cosmetic faults and fluid contaminants. If the seal is broken or loose, the defective vial is removed before it has a chance to leak on acceptable vials or on the production and inspection equipment. This particular check calls for visualizing the metal cap and the crimp at the bottom of the cap over a glass lip (see Fig. 3).
During product-inspection setup, the system operator uses the SVFocus software interface to select the measurement parameters to be performed, including type of cap, type of seal, proper height of cap, and other miscellaneous factors. Depending on the product line, each inspection station might use a single camera and rotate the product for multiple checks or use multiple cameras for a fixed-position product.
According to Lawson, a standard cap inspection requires the acquisition of three to six images obtained around the circumference of the rotating product. Typically, the SVIM module is used with an RS-170 camera and requisite cable to convey an image of the vial to the four-channel, asynchronous, frame grabber. No signal preprocessing is needed in the frame grabber.
A ring light placed above the inspection station illuminates the fast-moving vials. The cameras are placed orthogonal to the vials to obtain proper side views. Varying the viewing angles of the cameras to the bottle cap obtains a sharp line around the crimp, which verifies that the crimp surface continues beyond the halfway mark (45° in relation to the light) and indicates a tight seal. A threshold procedure, which isolates the bright reflection to check a measurement line for length and consistency, ensures that the crimp is unbroken around the cap`s circumference.
The lighting design also highlights any tears, rips, or surface defects on the side of the metal cap. However, because the light varies along the side of the cap, the SVIM module uses a nine-pixel kernel for edge detection across the vertical surface of the cap. This cap inspection continues with a dimensional measurement to check the height of the cap, ensuring that a rubber stopper has been placed underneath the cap and that the proper amount of compression was used during the capping process.
Glass and content checks
The tests made at the first inspection station conclude with a check of the glass vial itself. If the glass is made from tubing, the obtained pattern behind the glass could be distorted by cracks or other defects. If edge-detection techniques show that the distortion exceeds set parameters (or if any of the other checks fails the preset thresholds), the SVIM module sends a signal across the optically isolated lines to the PLC. This controller then increases its defect counter by one and conveys the vial to the defect bin.
Molded glass presents a different set of imaging problems because its walls vary in thickness more than tube glass and, therefore, some degree of distortion is expected. In this application, a light is placed under the vial that illuminates upward, using the vial as a light guide. Edge-detection techniques are then implemented to look for sharp reflections or shadows in the glass that would indicate defects.
Bottom lighting is also used for contaminant detection of the liquid contained in the vials. For this application, a lens system directs light upward through the vial. This setup illuminates such defects as fibers from inspector gowns, dust, or glass fragments that have infected the liquid. To avoid identifying surface blemishes on the outer glass surface of the vial as internal contaminants, the vial is spun at a velocity based on the type of contained liquid, viscosity, vial height, and friction characteristics. The velocity value is predetermined by SVResearch so that the system operator needs only to select the type of product under test and the associated inspection checks using the monitor`s configuration setup screen.
Depending on the type and color of the contained liquid, the vial is either front- or backlighted or is otherwise illuminated in conjunction with polarized transmission and receiving lenses. The liquid is then inspected by taking a series of images to determine the changes in reflection, refraction, and other light variations. If defects are detected, the light source is tracked over the series of images to make sure that it moves when the vial is stable, indicating that the contaminant is inside the vial and is not a surface defect.
For the pharmaceutical industry, the FDA requires that machine-vision inspection systems be able to detect all liquid contaminants that exceed 50 μm in width, says Lawson. This requirement establishes the necessary camera, lens, and field-of-view parameters. "The imaging speeds are based on line speeds, product, container size, and fluid dynamics," explains Lawson. "We`ve used as few as two camera images and as many as ten. These images are taken several times throughout the inspection system."
In addition to the use of threshold techniques, some liquids require blob analysis to distinguish bubbles from contaminants. According to Lawson, some pharmaceutical products effervesce like champagne when they are disturbed, forming uniform bubbles. By performing blob analysis, any object over a certain size is considered a contaminant, such as a glass filing or a foreign substance.
An SVIM module can communicate with the operator interface on the host PC or on another plant network via its installed ports. In this way, different product lines can be downloaded to the PC from a central core and then passed on to the SVIM module. Lawson says that the PLCs keep track of the overall inspection details, such as pass/fail numbers and defect types, and these data are passed through the network to the PC for archiving and reporting. Because the processing is done at the SVIM module, the PC itself does not require a large amount of computing strength.
Modular upgradeable machine-vision system detects contaminants for quality-grading pharmaceutical vials.
FIGURE 1. As many as eight Seidenader vision inspection modules (SVIMs) can be networked to a single host PC, which controls the operator-interface display monitor and a processor video switch. Each SVIM module can accommodate as many as 18 cameras; these cameras can be split among eight inspection stations containing one to three color or monochrome cameras. An external programmable logic controller accumulates pass-fail product counts and communicates back to the host PC via the SVIM module for archiving and reporting.
FIGURE 2. Seidenader vision inspection modules are installed in machine-vision systems to inspect pharmaceutical vials for cosmetic faults and fluid contaminants. The vials are fed single file from mechanical trays into a "star-shaped" wheel, where each vial is held by a spoke.
These vials enter into the first inspection area at line speeds of 100 to 600 vials per minute. The first vision check verifies that each liquid vial has been sealed properly.
FIGURE 3. During the inspection of pharmaceutical vials for cosmetic faults and fluid contaminants, the first inspection check verifies that the liquid vial has been sealed properly. Varying the viewing angles of the inspection cameras to the bottle cap obtains a sharp line around the crimp, which verifies that the crimp surface continues beyond the halfway mark.
Natick, MA 01760
St. Laurent, Quebec, H4T 1V8 Canada
Fax: (514) 333-1388
Matrox Electronic Systems Ltd.
Dorval, Quebec, H9P 2T4 Canada
Fax: (514) 685-2853
Seidenader Equipment Inc.
Morristown, NJ 07960
(201) 267 8730
Fax: (201) 267 2932
Fax: (011) 49-89-9090634
Harrisburg, PA 17112
(717) 540 0370
FAX: (717) 540-0380