How Eli Lilly Does In-Line Medical Device Inspection

Quality management of Eli Lilly medical devices got faster and more accurate with the use of an advanced vision system incorporating a 5 Megapixel camera and telecentric lens that inspects at the point of assembly.

For many industries and applications, implementing inspection at the point of assembly improves quality of the supply chain while reducing the impact of downstream detection of defects. Eliminating non-compliant material from entering final assembly increases yield, decreases factory losses, and has the potential to increase revenue. But doing accurate inspection at production speeds can be a challenge.

A medical device produced by Eli Lilly and Company presented a major inspection challenge for an automated testing system: Alignment of two cylinders needed to be measured within tolerances too close to be accomplished with conventional vision systems. During system development, inspections were made manually. The process took about 1 minute per part, which was much too slow for production operations. When tests showed that conventional vision systems would have produced too many inaccurate readings, Lilly chose a 5 megapixel In-Sight 5605 vision system from Cognex coupled with 5 megapixel telecentric lens from Edmund Optics. This provided accurate readings while reducing inspection time to about 200 milliseconds.

To create the devices, two cylindrical components are stacked on top of each other with a common centerline and glued together to form a subassembly. The components need to be aligned to the same axis within a very tight tolerance. During the startup of production, parts were manually loaded into a set of fixtures on a coordinate measuring machine (CMM). The operator then triggered the CMM to touch the part in several locations in order to determine the centerline of each component. The CMM program calculated the difference in the angle between the centerline of the two cylinders. The problem with this method is that it was too slow to support routine production requirements. The Cognex In-Sight 5605 vision system addressed this challenge by providing 5.0 megapixel resolution, more than twice the resolution available with its previous systems.

“Machine vision is a must on this application to meet our accuracy and production rate requirements,” said Aubrey Hawkins, consultant engineer for vision inspection for Lilly. “But we tried standard 640 by 480 pixel and 1200 by 1600 pixel cameras and discovered that they could not achieve the required level of accuracy due to resolution limits and perspective distortion. The only way to ensure accurate readings was to use the high resolution of the 5 megapixel camera.”

Resolution is a measurement of the vision system’s ability to reproduce object detail. For example, if two tiny objects are imaged onto adjacent pixels, it becomes impossible to resolve the distance between them. Increasing the resolution of the camera so they are separated by pixels makes it possible to make accurate measurements. The In-Sight 5605 also offers support for Gigabit Ethernet communication, IP67 rating to withstand dust and washdown, and a library of Cognex vision tools.

As more powerful sensors are employed, it’s important to ensure that the optics are able to reproduce the details that the vision system is capable of resolving. Distortion is an optical error or aberration that results in a difference in magnification at different points within the image. Perspective distortion, caused by the fact that the further an object is from the camera, the smaller it appears through a lens, is particularly important for gauging or other high-precision applications.

A telecentric lens corrects for perspective distortion so objects can be measured accurately. With a telecentric lens, the image size is constant regardless of object displacement, providing the object is within the range known as the telecentric range. These properties are achieved by capturing only the ray cones whose principle ray is parallel to the opto-mechanical main axis. This behavior is achieved by positioning the stop aperture exactly on the focal plane of the front optical group so that the incoming rays aim at the entry pupil, which appears as being virtually placed at infinity.

Lilly chose the Edmund Optics lens because it is designed for the small pixels associated with 5 megapixel cameras.

Tilt angle calculations

Craig Overton, senior engineer at Eli Lilly and Company’s Equipment Development Group, used the Cognex In-Sight Explorer spreadsheet-programming environment to configure the vision application. First he used the Find Pattern function to locate the subassembly. He used the Left Edge and Right Edge functions to find the edges of the lower cylinder. He used the Centerline function to identify the centerline of the bottom cylinder and determine the angle of the centerline relative to the field of view. Overton used exactly the same functions to determine the angle of the centerline of the top cylinder relative to the field of view, and then he performed a subtraction operation to determine the difference in the angles of the two centerlines.

After system development, Eli Lilly decided add the capability of measuring tilt angle to automated assembly lines globally.

The subassemblies are placed in trays that are introduced to the inspection machine. A six-axis Mitsubishi robot picks six subassemblies at a time from the tray. The robot moves the first subassembly in front of the camera. The Ethernet ports on the Cognex 5605 vision system enable it to be directly connected to any switch or hub on a factory network and, in turn, communicate with all other devices on the network.

The vision system also includes tools that make it easy to interface directly to common factory automation hardware such as programmable logic controllers (PLCs), robot controllers, human machine interfaces (HMIs) and personal computers (PCs). These include drivers, templates and sample code for open standard industrial Ethernet communications protocols such as MC Protocol, EtherNet/IP and Profinet for connection to automation devices from Mitsubishi, Rockwell, Siemens and other manufacturers.

Configuring the application

In this application, the vision system is connected via EtherNet/IP to the Rockwell Automation CompactLogix PLC using a Rockwell Add-on Profile (AOP). The first step was to load an electronic data sheet (EDS) file supplied with the vision system into Rockwell’s EDS wizard software, RSLogix. EDS files are simple text files used by network configuration tools to help identify products and commission them on a network. The vision system or ID reader AOP was then loaded using Rockwell’s AOP setup utility. The third step was to set up a new module on the PLC, add a unique name and set the IP address of the Cognex camera itself. The final step was to define the input and output nodes. A Lilly programmer then defined the associated logic in the familiar Rockwell environment.

Control signals are sent from the PLC to the vision system to indicate that a part is in position. The camera then acquires an image and calculates the angle between the centerlines of the two components. The precision and repeatability of the vision system generates very few false readings, says Overton.

Lilly uses similar vision inspection systems in a wide range of applications. In one example, a color In-Sight vision system is used to detect fallen vials, identify missing flip-seals and verify cap color on each vial. The elimination of count-related deviations and resultant rework saves money, time and the resources required for rework. The root cause of count deviations are also much easier to identify, correct and explain. Other inspection applications at Lilly handled by vision systems include printed packaging material inspection (label material, lot-code and expiration-date verification) and product identification via color inspection.

“The vision inspection accomplishes all of our objectives and performs flawlessly,” Hawkins says. “The 5 megapixel camera and telecentric lens has demonstrated the ability to inspect to a level of precision that would be impossible with a conventional vision system. The system delivers accurate readings at low levels of tilt with the throughput needed for success.

More in Data