By Doug Bartholomew, Contributing EditorMost pharmaceutical manufacturers use machine vision in one or more inspection applications, but concerns about validation and 21 CFR Part 11 prevent many plants from using data from these systems to the fullest. Instead, many companies let their machine vision inspection systems perform the simple task of accepting or rejecting capsules, bottles and labels on the production line. We dont typically integrate this data into our line, says Mark Lozen, package engineering manager at Patheon Pharmaceuticals (Cincinnati). Its strictly used on a 'go/no go basis.Unfortunately, drug makers may be missing out on other potential machine vision benefits, experts say. Absolutely, youd want to shoot that data upstream, so that quality service managers could look at it in real time and make decisions, says Reno Suffi, product manager of the sensing group at Omron Electronics (Schaumburg, Ill.), a vendor of vision, safety, and fiber-optic technology. Unfortunately, a lot of companies dont implement the data portion of the technology, they just do the 'go/no go part of it.Regulatory concerns make companies reluctant to integrate production line machine vision data with other upstream information such as manufacturing execution systems (MES) is regulatory. Many believe that, by doing so, the machine vision systems, whicih usually operate in stand-alone mode, would be subject to FDA regulations for electronic signatures, data access and security. You start getting into 21 CFR Part 11 validation and documentation issues, confirms Patheons Lozen.Suffi agrees, although he points out that stand-alone machine vision systems do not, technically, fall under the constraints of 21 CFR Part 11. Nevertheless, pharmaceutical companies are being cautious and taking a wait and see approach, he says.Taking machine vision beyond pass/failDespite their caution, though, more drug manufacturers are extending the use of machine vision not only to cull questionable or defective products and labels, but also to identify production problems early. The pass/fail inspection information from machine vision can be used to troubleshoot a system, says Chris Clarke, president of Clarke Engineering Services, Inc. (Indianapolis), a vision system integrator.Typically, there is a simple pass/fail output signal that goes to a programmable logic controller (PLC) or some other device, which will instruct another component on the line to reject that package, explains Steve Wong, pharmaceutical and medical industry marketing manager at Banner Engineering (Minneapolis), a leading machine vision systems vendor. If a system has, say, 20 fails in an hour, an alarm is triggered and a request is made for maintenance on the line, Wong adds.At Eli Lilly, for instance, Clarke Engineering used machine vision sensors from Cognex Corp. (Natick, Mass.) to develop a fully-integrated tray inspection system that enables operators to isolate and solve root causes of count deviations. Lillys goal with the new system was to improve quality by performing a total visual inspection of vials produced at the companys Indianapolis plant, replacing mechanical counters that were prone to jam and resulted in counting errors.The new quality inspection system, the TIS-3000, combines vision technology with automatic tray handling. In addition to counting vials, it uses machine vision to detect fallen vials, identify missing flip-seals, and verify vial color. The system also differs from most in that it performs the vial count at the front end as well as the back end, thereby establishing a performance benchmark and eliminating much rework.The TIS-3000 establishes the initial count on our packaging line, explains Rob Stapleton, operations team leader at Eli Lilly. If the numbers do not reconcile at the end of the line, we have to open up the entire job and find out why a discrepancy exists. This can be very costly and time consuming, considering packaging order sizes that range up to 360,000 vials.The image is captured by a single Cognex In-Sight vision sensor suspended inside a hood in which the conveyor area is illuminated by high-frequency fluorescent lamps. The system makes a pass-fail decision based on a pre-established count of how many vials should be in each tray. Any tray that fails to contain the proper count is rejected.Trays that are rejected because they contain a cap whose vial is the wrong color, sometimes referred to as a stranger, will cause the system to shut down. If a 'stranger is detected, the system locks up and requires supervisor intervention before the tray proceeds to the reject station, Clarke says. This is an important feature in order to error-proof the process to prevent the wrong medication from being packaged into a lot.Trays rejected because the count is wrong or a vial has fallen down are automatically returned to the load station operator via a reject conveyor. There, the operator can view the cause of the failure on a screen mounted on a pivot arm, fix the problem, and send the tray back through.Chris Clarke, the creator of the tray inspection system with Eli Lilly, says it ensures 100% accountability of all product throughout the process. In case of a process failure, Clarke points out that three consecutive failures will automatically halt the machine. Two of these systems have been installed and validated at the companys Indianapolis facility.Aiding QC with a PCManufacturers that want to further investigate quality issues can elect to save the last 100 rejects as image files. These can be stored on a PC used by the operator or a plant supervisor. The decision point is usually in the operators hands, Clarke says.Some machine vision software packages, such as Cognexs In-Sight system, include traceability capabilities, Clarke says. With this you can track who made changes to the system, what kind of changes were made, and when, he says. The systems with a 21 CFR Part 11 package option will track the history of the machine vision inspection. Certain events are logged and that information is stored and can be retrieved.Most higher-end vision systems are able to track and store this information on a database on a PC, he adds. Some companies are tracking and storing that information to enable better lot traceability, Clark says. With vision systems, you can go back and re-create what happened, check the images, find out what happened, and then troubleshoot the problem.Cognex can be used to categorize and delineate pass/fail parts as well as the nature of the defect and can store the related images based on the failure mode, adds Brian King, an engineer with Chicago-based integrator Industrial Eye.On the downside, more sophisticated vision systems can be expensive and difficult to validate. These systems require significant validation procedures, Clarke admits. Validation costs are high as well, says Clarke. As a result, manufacturers are reluctant to change a system that they believe already works.Another pharmaceutical firm, Hi-Tech Pharmacal (Amityville, N.Y.), which manufactures both prescription and over-the-counter products in liquid form, is using a machine vision system to perform label identification and verification. If the vision sensors detect a problem, the system stops the line and the operator is notified, says John Williams, production maintenance manager at Hi-Tech.One problem the pharmaceutical manufacturer faced was ink becoming blurred or illegible on product code labels containing the lot code, product identification and expiration date. With the machine vision system, product information that is stored in the cameras memory enables it to verify the label code on each product.The operator will be prompted by the system to check the product code, to see whether the ink ribbon is running dry or to find out the reason the code is degraded, Williams says. Before we had this system, we were not notified of the problem before it got to the customer.Hi-Tech Pharmacal uses a machine vision system from DVT Corp. (Duluth, Ga.) to upload inspection data and images to the companys Ethernet data network. On a PC in his office, Williams can see whats happening with half a dozen production lines. He can see how a particular label looks on the camera and can make adjustments to the system as needed, says Ed Mullen, outside sales engineer for Axis Industrial Automation (Somerville, N.J), a distributor of DVT products.Similarly, the machine vision system from Banner Engineering provides a live video feed to the operator or quality managers PC. If for some reason the line is off, or is falsely rejecting a product, the operator can see that and make changes, says Wong.Mullen says its typical to have a PC that controls the system in the plant manager or supervisors office. Its not good to have a PC on the line, Mullen says. With this application, you only want people trained on the software to be able to make changes to the system.Clarke, for one, believes pharmaceutical manufacturers have only scratched the surface when it comes to machine vision use and its potential applications in improving production and automating associated data recording, reporting and storage.Identifying new applicationsWhile machine vision is used most commonly for inspection applications, some observers believe other uses, especially identification of materials, products and product labels, offer the most benefit and value to manufacturers going forward.Identification clearly is the most interesting application of machine vision, more so than gauging, inspection or guidance, contends King of Industrial Eye. Reading codes, reading characters, verifying materials, checking that the identifying numbers on a package are correct, and other labeling issues all are important, he says. You want to make sure the right label went on the right product, with the right text printed on it. King works on machine vision integration projects for automotive companies as well as pharmaceutical manufacturers. It used to be that pharmaceutical companies were more stringent in terms of their requirements, but today automotive is almost as strict, he says.To ensure the success of pharmaceutical machine vision projects, King, a mechanical and electrical engineer, keeps certain key elements in mind when he first walks into a plant. Typically, the application is already defined, but there are certain variables I want to control in order to give overwhelming advantage to my optical and mechanical approach, he says. Light is the most important. Generally, this means overwhelming the situation with light, he says, rather than depending on a dim or flickering ceiling light.Other issues critical to the success of the application are distance to the product or package, background, and its positioning relative to the camera. Then there is the design and labeling of the package itself. You want to make sure you are not reading an ambiguous font, King says. Camera location is another factor. So is good mechanical placement of the parts, so that each unit is presented in a repeated fashion.To make it all work, King wants to know that plant management will do whatever it takes to guarantee a successful application of the technology. You have to have a partnership, and they have to understand why certain prerequisites are needed for a successful operation of machine vision, he says. Otherwise, you have to tell them, 'Youre only going to get 90 percent here, and youre going to have to throw away five to 15 percent of your products.' Thats an exaggeration, of course, because for most companies installing machine vision systems, success is measured in the smallest reduction of error. Says King, Its all about the last one-quarter of one percent that makes the system work.