The U.S. Department of the Treasury's Bureau of Engraving and Printing (BEP) is responsible for producing all of the currency for the world's largest economy. From the paper on which the currency is printed, to its output of about 37 million notes each day, most aspects of BEP's design and printing processes are distinctive. Many of its quality assurance needs are unique as well. The agency verifies that its printed materials are clear and accurate and meet all security requirements. To help meet these demanding requirements, the BEP converted several of its inspection stations from human operators spot checking currency to automated visual inspection of each note.
The BEP identified several areas of the process that would benefit from automated visual inspection. To help ensure security, reduce variability and maintain its print standards, the BEP selected the Sherlock machine vision software from ipd, a division of Coreco Imaging Inc. (Billerica, MA). The company developed three machine-vision systems: automated inspection systems for printing plate quality assurance, automation of measuring finished notes for print registration consistency, and on-line measurement of a quality and security feature on newly printed bills.
To reduce or eliminate inconsistencies associated with physical note size and the location of engraved images on the chrome intaglio printing plates on which the currency is printed, the BEP's production engineering group developed a Plate Measurement Device (PMD). The PMD uses positioning technology and machine vision to measure the layout pattern of the security features on these plates. The engraved artwork and registration marks are gaged before production to verify plate accuracy which is essential for precise print registration. Any distortion caused by prolonged intaglio printing is identified.
The team developed the PMD's machine-vision inspection system using a variety of off-the-shelf opto-electronic components. A frame grabber integrated within the host PC captures images from a 1K X 1K monochrome camera. Using a precision lens with a 0.5-inch field of view, the camera achieves a measurement tolerance of ±0.001 inch. In addition to capturing images for analysis, the frame grabber provides the digital I/O that is used to monitor manual switch settings and safety sensors on the PMD. The operator controls the vision system using a customized graphical interface via keyboard and mouse, and a 21-inch color monitor displays a live image of the currency plate being inspected.
Because the polished chrome plates are shaped to follow a four-plate printing cylinder radius, obtaining a high-
resolution image was one of the challenges that the team faced. They solved the problem by using an LED-based, dark field illuminator in conjunction with a megapixel camera, making the engraving appear clear and well-defined for the machine-vision software's "calipers" to do measurement functions with minimal pre-processing of the image.
After the PMD captures all pertinent plate data, analysis is done using machine vision software and an object-oriented program language, and then all data is automatically saved to a network database. The analysis tool can also compare the current plate to a "golden plate," which can either be a mathematical composite of the target plate or a plate that was previously measured. When comparing the current plate to the golden plate, the operator can specify a tolerance. The monitor displays measurements outside the tolerance in red and an acceptable reading in green. The comparison provides the operator with immediate go/no go feedback, while the networked data is used offline to detect process trends so that any recurring positioning problems can be identified and remedied. The machine-vision software is integrated with the object-oriented program interface that enables users of all computer skill levels to successfully analyze images.
The PMD machine vision system has operated for 8 hours a day, 5 days a week for 3 years, and is an integral part of the quality assurance and process controls.
All finished banknotes are targeted to be the same size and are printed 32 per sheet. Although each sheet is electronically inspected on the printing press and again before banknote numbering, the BEP did not have a reliable way of identifying finished banknote feature registration discrepancies that occur as a result of accumulated subtle variations with the many print phases and trim operations associated with currency production. Without this information, the BEP was unable to quantify its process capabilities or to measure improvements in banknote quality. Therefore, the production engineering group developed a vision-based note measurement system that automatically measures and records 27 note registration features (137 data points) on each cut bill. These include substrate size, intaglio print size and position, on both sides, and position of the seals and serial numbers. Measuring bank-notes in reflected light is a challenge, as the intaglio printing process is variable by nature and finding print edges consistently on the fine engraved artwork is difficult.
An image-based inspection system, which had to accommodate multiple banknote designs, proved a success. Each measurement station has a vacuum fixture to hold and position the banknote, three digital cameras, a microprocessor-controlled light source, a continuously diffuse illuminator, a PC equipped with two frame grabbers and two machine-vision software packages that co-exist for image analysis. After each note is positioned, images of the front and back of each note are taken and transferred to the frame grabber. Each image is analyzed by both machine vision applications for optical character reader and caliper measurement. The software's landmarking tools account for trim and print variations, and compensate for incorrectly oriented notes. Using grayscale edge detection, the machine vision software measures pixels and sub-pixels repeatedly for precision within ±0.2 millimeter. This was crucial because many of the fine line features printed on banknotes are smaller than the camera's pixel resolution.
The team developed software to provide an easy-to-use graphical interface, scale pixel measurements to millimeters, enforce data integrity and store the measurements in a spreadsheet-compatible database.
Using this vision system, engineers can quantify the registration variability to within a half-millimeter granularity. As they leverage their experience with this technology, the group anticipates it will build tools to measure even finer reductions.
With ongoing quality considerations and the increasing need to deter counterfeiting, the team developed an automated inspection system to examine a semi-covert, machine-readable feature on each banknote. This feature is created by applying inks in denomination-specific patterns in precise locations throughout each printed sheet.
The BEP had used human inspectors for this application, but, with 8,000 to 10,000 sheets of 32 banknotes coming off the presses each hour, only a small percentage was inspected, and with less reliability than with an online, automated system. The inspection system had to withstand the harsh environment inside a high-speed printing press that applies 80 tons of pressure per sheet. This process creates floor-
shaking vibrations that make capturing an image challenging.
The inspection system starts with an 8-bit monochrome tethered-head camera that provides increased sensitivity and low-light imaging capabilities. In addition to capturing images from the camera, a frame grabber provides digital I/O capabilities that activate a light bar and voice module. These indicators provide operators with feedback about each sheet.
The system incorporates off-the-shelf machine-vision software for image analysis capabilities. Using an object-oriented program language integrated into the software, the team wrote algorithms to enhance the system's ability to capture high-quality images under these conditions.
Over the past four years, the system has operated 24 hours a day for 5 to 7 days per week and performs 100% inspections at a minimum of 2.4 sheets per second.
ipd, a division of Coreco Imaging