Machine Vision Bolsters Quality Control

Carl Gerst, Cognex Corp.

Machine vision technology has long played a key role in pharmaceutical manufacturing, particularly at the packaging stage. But now, as the Food and Drug Administration (FDA) bar-code regulation raises packaging standards, drug manufacturers are turning to increasingly sophisticated vision solutions.

Many pharmaceutical companies already use machine vision systems for optical character verification to inspect human-readable data such as the date, lot code and expiration date on drug labels. This type of inspection involves “training” the vision system on the printed character set and telling it which alphanumeric string it should expect to see during production. Through statistical font training, a process that involves building a model for each character by looking at a large sample of characters, sophisticated algorithms have been developed to eliminate false accepts — i.e., passing an incorrect string — by distinguishing between confusing character pairs such as an 8 and a B.

A machine vision system can identify such defects as missing labels (A) and labels positioned too high (B), while passing a properly positioned label (C). Key features, such as the starburst on the bottle (D), also can be checked.

Camera-based machine vision systems also can perform tasks such as reading National Drug Code bar codes — 10-digit strings identifying the manufacturer, product and package size — and inspecting other items on a label, such as logos or directions. More recently, machine vision systems have been used to ensure that safety seals are correctly mounted on bottles, that instruction and warning sheets are placed inside packages and that boxes are properly labeled and sealed.

Another advanced application concerns the FDA’s requirement to place bar codes on small packages, such as individual tablets in blister packs. Because of the space restrictions on tablets, Reduced Space Symbology (RSS) and Composite Symbology (CS) codes have been developed to enable drug manufacturers to encode the National Drug Code number, date, lot code and other information in a space-efficient, machine-readable bar. A typical blister pack, measuring only 3 3 5 inches, may hold as many as 20 tablets, and drug companies are often looking to simultaneously read each RSS/CS code on as many as four blister packs at line speeds of up to 150 packs per minute.

These demands pose a challenge to modern 2-D imaging products (array-based CCD products), which lack the resolution to read all the codes in a single field of view. For example, if an RSS/CS code is printed with a minimum X dimension — the smallest width in a bar code, also known as an “element” — of 6.6 mils, and the reader requires two pixels per element for robust and reliable reading, then a 640 × 480 video graphics array camera can support a field of view of only 2.1 × 1.6 inches. However, reading four blister packs at a time requires an effective field of view of more than 6 × 10 inches. Even readers that use high-resolution superextended graphics array cameras, which support 1280 × 1024 resolution, are limited to a field of view of approximately 4.2 × 3.4 inches.

Cameras can be postitioned at every point on the production line. In this example, a vision system scans bottles of pills, first to ensure that the labels have been properly applied and again to verify that the packages are correctly sealed.

Thus, solutions supported by array-based CCD products often require as many as 10 cameras to meet the desired field-of-view and line-speed requirements. Although this is a feasible option, a more efficient solution might involve a line-scan camera, frame grabber and PC. A single 2k line-scan camera can easily capture the required field of view, with more than enough resolution — three pixels per element with a 6-inch-wide field of view — to enable pharmaceutical companies to read every RSS/CS code, while at the same time delivering adequate resolution for print inspection of human-readable information.

Looking ahead, vision systems can be expected to move upstream on the production line. Some drug companies are already using cameras to count pills, to check for chipped or broken ones, and to inspect their sizes and shapes to ensure that the wrong pills are not being packaged. These types of applications are restricted by the technical limitations of modern cameras.

But as the technology becomes more and more capable, and as prices continue to decline, machine vision is likely to play an ever-larger role, not only at the manufacturing stage, but also, via the use of scanners, throughout the pharmaceutical supply chain and even at the point of care.

Meet the author

Carl Gerst is identification program manager for Cognex Corp. in Natick, Mass.