|
1. ABSTRACT |
| The precision, sharpness, and cleanliness of the edges of cut sheet papers has a dramatic effect on the performance of the papers used in copiers and printers. Several manual methods have been used to evaluate the cut quality of such papers. Apogee Systems, in conjunction with Georgia Pacific’s Ashdown Mill, has developed an automated method and instrument for measuring the edges of GP’s cut-sheet papers. Using an 11 X 17 – inch tabletop scanner, a PC, and specialized image analysis software (Spec*Edge), the system provides a variety of quality-related measurements. This presentation describes the features, the measurements generated, and an engineering comparison to human visual evaluations. |
|
|
|
Preventing paper jams in copiers and printers is a critical factor in the quality of cut-sheet papers. Fortunately for the paper industry, few end-users blame the paper that they are using. More frequently, they blame the copy machine or the printer, resulting in a flourishing copy machine service industry. However, Xerox and the other copier and printer manufacturers and the paper-makers know the truth. The primary factor in feeding cut sheet papers at high speed is quality control of the sheet size, shape, and edges. A nick of only a few thousandth of an inch on each sheet can create dramatic jams in a high-speed copier. A few hanging fibers on each sheet not only makes the edge of the ream look rough, but can contribute excessive dust to the feed path of the copier, eventually creating feed problems.
|
|
Traditionally, cut-sheet QC personnel have graded the cut quality of sheets using a modified microfiche reader to magnify a silhouette of the sheet edge by about 42X. A QC supervisor then grades the sheet edge against a template taped to the microfiche reader screen. This method has been used successfully throughout the paper for several years; however, it is dependent on the personnel for repeatability and the interpretations are often the source of considerable debate. Additionally, this technique does not measure the squareness of the sheet, locations of holes, corner quality, or slight curvatures in the sheet sides.
|
Last year, Scott Simmons, a quality control supervisor at Georgia Pacific’s Ashdown mill, conceived an alternative based on a desktop scanner and a personal computer. Because Apogee’s dirt count scanner system has become the tacit standard in several GP finished paper mills and over 150 other locations in 13 countries, Scott approached us with the idea of developing such a system. Along with extensive help from GP, Apogee developed the Spec*Edge system to meet their specifications. The system entered service at the Ashdown mill last December, and more recently at Xerox’s Webster, NY technical center. This paper describes the team relationship between Georgia Pacific and Apogee, the technical requirements, image analysis algorithms, and measurements that result from the method.
Back to Top
|
3. INITIAL TECHNOLOGY DEMONSTRATION
|
| The first step in any technically innovative effort must be to develop or locate the needed technologies, and to demonstrate that the technology is adequate. In our case, the image analysis measurements needed to correlate closely with the human/microfiche readings. Before any contract was written, the team agreed that Ashdown would provide a set of samples covering the full range of manual readings from one to five. Apogee scanned the sheets with a 600dpi Hewlett Packard scanner, and tried several different algorithms for their analysis before we confirmed that the following algorithms matched the microfiche technique readings. These results proved to us and to Georgia Pacific that the technology existed to make the necessary measurements, but this method had to be packaged into a software and hardware system that could be used in day-to-day production. |
| 4. DEVELOPMENT AGREEMENT |
| Most software companies would develop a system like this as a one-of-a-kind installation, requiring the customer to pay all of the development and support costs. However, Apogee has tried to learn a lesson from Bill Gates who started Microsoft by developing a system in close relationship with the ultimate user, and selling it to him at an off-the shelf price, but retaining the license so that it could be sold to other customers as well. This is the relationship we built with Georgia Pacific. In this relationship, both parties carry some of the responsibility for the project success, and both can profit from its success. Apogee's role in the relationship was to: |
|
Georgia-Pacific's Role |
|
Apogee Systems' Role |
| • |
Identify the technical requirements. |
• |
Develop the needed technology. |
| • |
Teach Apogee about the cut-sheet business. |
• |
Identify and purchase the scanners and computers. |
| • |
Provide space and power within the plant. |
• |
Develop the software and perform alpha testing. |
| • |
Perform beta-level testing. |
• |
Install and train GP's operators. |
| • |
Evaluate and correct the user's manual. |
• |
Write and proof a software user's manual. |
| • |
Provide ideas and on-going requirements. |
• |
Provide support for the system for 1 year. |
|
What resulted from this kind of team development was a system that is specifically tailored to the needs of the cut-sheet producers, uses their terminology, and closely matches their operational and technical requirements. Apogee, on the other hand, emerges with a valuable product to market, and finally the industry has a new tool and method for use both by paper producers and consumers.
Back to Top
|
| 5. IDENTIFYING A SCANNER |
| GP manufactures paper in a wide variety of sheet sizes and hole placements, including 81/2x11", 81/2x14", and 11x17", and larger folio sheets. The first problem was to find a scanner with sufficient resolution and bed size to measure sheets up to at least 11x17." Most commercial scanners are designed to image a standard-sized sheet, but not an area large enough to include the sheet edges. Additionally, the sheet needs to be scanned in transparency mode so that the edge of the sheet is view in silhouette, and finally the early testing revealed that the sheet needs to be scanned in at least 600dpi and preferably higher resolution. Scanner manufacturers make several classes of scanners, from those intended for scanning photographs to put on the internet, to professional scanners used in prepress applications, to large drum scanners for digitizing blue prints and large line drawings. Our initial survey found five scanners that could meet these specifications, ranging in price from $3000 to over $30,000. Detailed testing of four of the scanners revealed that only two had the size and image clarity that we required and the team settled on the less expensive of the two, a UMAX Mirage IIse, manufactured in Taiwan, but widely used in the US prepress marketplace. |
 |
|
UMAX Mirage IIse Specifications |
| • |
Single pass color, flatbed |
| • |
11.4 x 17" image area |
| • |
700dpi optical resolution |
| • |
Closed (dust free) housing |
| • |
12/36-bit grayscale/color |
| • |
Density up to 3. |
| • |
SCSI-2 Interfa |
| • |
29"x21"x6", 57lb |
| • |
Retail Price: $2995 |
|
| Back to Top |
6. MEASUREMENT REQUIREMENTS
|
In addition to the sheet size, Georgia Pacific specified that the system must measure squareness of the sheet, corner sharpness, hole location, "cut quality" of the four sides, and a count of hanging fibers per inch. The dimensional measurements needed to be accurate within about 0.01 inches. All of the measurements are tested against user-input warn and fail limits. The system needs to track several administrative factors, including date, time, operator, roll set, machine number, etc., and save both this data and the measurement results to a Microsoft Access“ database, an Excel“ spreadsheet, and/or a text file. In determining how to measure cut quality, we found that there were two kinds of variation along the sheet edge. In addition to the small nicks and hanging fibers along the edge, there are long variations in the sides of a sheet. We termed this variation as curvature. So far, we have not correlated the sources of this apparent curvature with physical attributes of the slitter, but it seems conceivable that slight wobble in the slitter or flutter in the sheet might cause variations like this. For a quality control point of view, the cut quality measurement is by far the most important, and considerable effort has been spent correlating our measurements with those obtained visually from the microfiche technique.
Back to Top
|
| 7. Acquiring a digital image of the sample |
| In our method, a desktop scanner is used to acquire an image of the sample, and a desktop PC then applies a series of image analysis algorithms to produce quantitative measures of the sheet quality. The digitized image is simply a two dimensional matrix of picture elements or pixels. The scanner generates a numerical measurement of the amount of light which passes through the scanner glass for each pixel. Where the sheet blocks the light, the scanner generates a gray value close to zero. Where the sheet is not blocking the light, the light level approaches 255. This measure of light level is referred to as gray scale values (gsv). |
| The size of the pixels is defined by the resolution of the image, 700 dpi (dots per inch) implies 700 pixels per inch in each direction. The gray scale value of each pixel, measuring the light level in that area, is stored as a byte in the computer, thus an 11x14-inch sheet, digitized at 700dpi becomes a 95MB file on the computer. |
| 8. IMAGE ANALYSIS AND MEASUREMENTS |
| The best way to describe the actual image analysis process is though the flowchart in Figure 1. The process flow is actually fairly simple, starting with placement of a sheet on the scanner and scanning the image. Unfortunately the current scanner technology and the scanner interface limit the speed of the system. Consider an 11x14" sheet, scanned at 700dpi. The resulting image is about 95megabytes and requires about seven minutes to scan. Once the image is transferred to the PC through the SCSI interface, the PC software identifies the edges, holes and corners and creates a sub-image for each; a 3-hole sheet, becomes 11 images, and each image is analyzed separately by type. |
 |
| Sheet Edges |
| Analyses of sheet edges are by far the most complex of the three types, primarily because of the several effects that that must be measured. First, the image of a sheet edge is limited to the area 0.4 inches away from the corners to eliminate the possibility of folded corners, etc. from corrupting the measurements. Then at each pixel along the image of the edge, the actual edge location is determined by thresholding the gray value and a resolution enhancement technique described in the next section. Each edge is processed separately. First, a straight line is fit through the edge data, so that we have a reference to where the ideal sheet edge should be. Then at each pixel location along the edge, we measure how far the actual edge is from the ideal edge, the straight line. Figure 2 depicts this measurement. |
 |
| In our initial testing, we found that there are two kinds of error in the sheet edge. There are narrow variations caused by nicks in the knife, tearing of the fibers along the edge, etc. Separate from those fine variations are long waves in the edge. Because the short variations correlated very closely with the judgements using the microfiche technique, we call this variation the "cut quality." The longer waves along the edge are called curvature. These two effects are measured separately, by passing the data though a high pass filter for the narrow (high-frequency) variations, and through a low pass filter to measure the (low frequency) curvature variations. |
| To create a numerical measurement of the cut quality, we simply create a histogram of the distances from the straight line. That is, we count the number of places along the edge where the distance from the straight line to the sheet edge is 0.001 inches, the number that are .002 inches… A statistical measure of the width of the histogram is the standard deviation. That number captures both the number of places that the edge is inside the line as well as the number of places that the edge is outside the line. |
| Curvature is measured by plotting the data resulting from the low-pass filter against warn and fail bands around the straight line. If the sheet edge exceeds the warn or fail levels, the side is flagged as bad. Finally, the requirements specified that we count the hanging fibers. Because fibers hanging from the edge of the sheet are normally curved, the method does not do an adequate job of measuring their length; however, the sharp spikes that they create in the high frequency cut quality data can be counted whenever they exceed the fail limit. Hanging fibers are counted simply as the average number of spikes per inch along the edge. |
| Corners |
| Measuring the corner quality is actually fairly simple once the sheet's edges have been located. Remember that in analyzing the edges, we fit a straight line through each of the sheet edges. If we compute the intersection of each pair of these edge lines, we can find the ideal corner location. By simply counting the number of sheet pixels missing inside the corner, and the number of extra sheet pixels outside the corner, we have a measurement of the area of the sheet which differs from the ideal corner. By allowing a tolerance around zero, we can flag a corner as passing, warning, or failing. |
| Holes |
Holes are measured both in their diameter and the location of their center. Because the edges of the holes are often serrated, locating the exact center and diameter can be difficult. The method simply measures the area of any hole in the sheet by counting pixels and averages their locations to determine the centroid of the hole. The diameter with an area equal to that measured and the centroid are then compared to the user-specified values to generate a pass, warn or fail grade for each hole.
Back to Top |
| 9. RESOLUTION ENHANCEMENT |
| The actual resolution of the system and how it is achieved has been the subject of numerous questions and discussion. Normally, we would assume that the resolution of an image measurement system is limited to the size of an individual pixel from the scanner or camera. If the scanner, however, generates a grayscale or color image, the grayscale value can be used to achieve resolutions finer than an individual pixel. Image analysis firms generally threat these techniques as proprietary; however, they are generally referred to as resolution enhancement or "pixel splitting." Basically, the technique uses the grayscale value of each pixel to interpolate the percentage of a pixel that is covered by the target. Consider the diagram in Figure 2. If we determine that the scanner sensors measure the grayscale value to be 0 where the sheet is entirely masking the sensor, and the grayscale value where the sensor is entirely unmasked to be 255, then a grayscale value of 128 might indicate that the sensor is half covered. Of course, this assumes that the sensitivity of the sensor is uniform across its viewing window, that no shadowing occurs, and there is no pixel spreading. None of these are true, but with similar techniques, these factors can be accounted for as well. Theoretically, pixel splitting can generate resolution enhancements equal to one over the grayscale difference between the masked and unmasked pixels, i.e. about 1/200 pixel. In actuality, resolution enhancement can achieve measurements of about 1/10 of a pixel. In our case, with 700dpi resolution, this means our measurement resolution is about 0.0002 inches. |
|
|
To actually determine the resolution of the system, we have had several calibration targets manufactured with known or measurable dimensions, holes, and notches. Figure 3 illustrates the target that we are currently using. It consists of a 2x3-inch target on a 5-inch square background. One edge of the target has a 0.002" deep notch. A second edge has a 0.005" notch. The third side has five narrow nicks, varying from .0005" deep to .004", and the fourth is smooth. All four corners are square, and a 5/16" hole is properly space from one of the corners. All measurements are accurate to 1/2 micron or about 0.00002 inches, except that corners are rounded with a radius of about 2 microns or 0.00008". The target is made of etched chrome on optical-grade glass, and is used by the microchip industry as a photographic mask. For our purposes, we expect to make photographic contact prints on fine grained photographic paper of film. The target and the photographic prints will be measured using an NIST/ISO-certified measuring microscope, so the target accuracy will be traceable. Once we have a target, an ISO test procedure and certificate will be prepared for periodic testing of the units. The photographic prints should be cheap enough that they can be used on the production line and replaced periodically without undue expense.
Back to Top |
| CONCLUSION |
| The technique described here has been in use at Georgia Pacific's Ashdown, Arkansas mill for about nine months and more recently at Xerox. Both have found a high degree of correlation between the cut quality measurement described above and the human/microfiche technique, and their results will be presented separately. Generally, cut quality of 0.001 is about a microfiche grade of 1, 0.002 is about a 2, etc. Testing of actual resolution is incomplete at this point as is a complete understanding of the curvature measurements. |
| Basically, this technique is generating highly accurate results that are repeatable and well correlated with human judgement. Several additions and changes are anticipated in the software, including: |
| • |
Slight modification to the cut quality/hanging fiber calculations |
| • |
Improved scanning and analysis speed |
| • |
Analysis of a single edge (for speed and problem analysis) |
| • |
Improved software reliability (disk/memory constraints) |
| • |
Up to 50 holes/sheet rather than 10 |
| • |
Square/Rectangular holes |
| • |
Tab location and size measurements |
Perforation measurements, including holes, location, and lands |
The team-based development model was highly successful and has resulted in "wins" for all involved.
Back to Top |
|
© Copyright 1999 - All Rights Reserved. Apogee Systems, Inc.
|
|
