• Font Size:
  • A
  • A
  • A

Feature Articles

Digital Camera System Helps Cerex Advanced Fabrics Raise the Bar for Quality

by Christopher M. Roberts, Process Control Engineer, Advanced Fabrics - AIA

Abstract

Cerex Advanced Fabrics, Inc., an ISO 9000 certified producer of spunbond nonwoven fabrics, recently expanded production capacity and upgraded their automated web inspection system by installing a LASOR/Systronics digital camera system.  This system, coined WebData, meets Cerex's stringent technical specifications and allows them to identify, remove, and automatically react to defects as they occur, while generating data to target future process enhancements.  As a result, manufacturing yields and efficiencies continue to tick upwards as quality soars.

Introduction
Cerex Advanced Fabrics, Inc., a subsidiary of Western Nonwovens, Inc., operates several spunbond nonwoven fabric production lines near Pensacola, Florida.  At Cerex, an ISO 9000 certified manufacturer, we produce both chemically and thermally bonded nonwovens over a wide range of colors and basis weights using proprietary manufacturing processes.

After a capacity expansion in 1998, we saw the opportunity to enhance product quality by integrating the latest advances in computing and detection algorithms with our laser based web inspection systems.  As we looked more closely at the upgrade requirements, we found several significant compatibility challenges between our existing hardware and the newest processors, so we opted to redesign the system from the ground up.

The new system, coined WebData, would:

  • Find and report any defects
  • Provide an objective measurement of product uniformity
  • Have the capability to be used as a tool to automate the removal of defects in the converting process, giving us the capability of feeding defect data to a high-speed slitter rewinder
  • Improve control of the process by providing instantaneous feedback to operators in the control room, identify repeating defects and give us the ability to automatically correct defects
  • Populate a database that can be mined to target areas of opportunity for future capital enhancements

Camera Types
We evaluated the idea of expanding a prototype system engineered in-house, however, we decided that enlisting the expertise of an outside vendor would allow us the opportunity to leverage the learning of others.  In the initial phases of the vendor search, we considered both laser and digital camera inspection systems.  We found that modern laser systems are designed to best find types of defects that are rare to our process, and as such were just not a very good fit for us.  Digital cameras, especially line array, are designed to detect even the most subtle defects in our products.  Line array digital cameras have arrays of 1000 or more pixels arranged in a straight line to capture a cross section of a moving target, such as a web.  If scanned at a high rate, a continuous image of a moving web can be created with very fine resolution and the most minute variances identified.  After a review of our current grading standards, estimates of future customer requirements and capacity enhancements, we decided on a detection standard that required several cameras across our web.

Though we had a good idea of the physical accuracy requirements for a new system, there were key decision points that drove our vendor selection process.  Those factors were: light sourcing, defect processing method, detection algorithm, data strategy and service.

Light Sourcing
In order to determine lighting requirements, we considered the product, conditions at the inspection station, types and sizes of defects and capital budget constraints.  We evaluated several different lighting methods.

Transmissive lighting means that a light is positioned on one side of the web and the camera on the other.  Changes in the amount of light transmission in the product are used to determine the presence of defects.  For example, a polymer drip is typically very dense and acts as a barrier to light transmission.  When compared to the level of light transmission to adjacent regions, the camera will register the drip as a low-light or 'thick' spot.  In contrast, a hole allows more light to be transmitted and will be interpreted as a 'thin spot.'  A common spunbond defect is a blowback, or hang, characterized by a half moon shaped thin area bordered by a thick area.  Transmission detection is very effective in detecting this type of defect.  The largest drawback to transmission is lack of effectiveness with a colored product.  We found that in most cases the opacity of a product was the same regardless of color, thereby making transmission a poor barometer of variability caused by color streaks.

Reflectance lighting focuses the light at an angle to the product; a camera is positioned to catch the reflected light.  Thick or thin spots are relatively simple to detect - the more dense the product, the more light reflected and conversely.  Reflectance detection can be very useful in finding pigment streaks when making a colored product.  Reflectance is also an excellent choice for detecting small areas of discoloration, since they reflect much less light than a brilliant white product.  Transmission rarely catches discoloration because the opacity of these areas is not usually much different from the rest of the product.  Reflection can be a challenging method of detection if the product has been thermally bonded with an engraved calendar.  The pinpoint or patterned bond areas are typically thin films which tend to scatter light much differently than the unbonded areas and generate a large amount of signal noise.

Several of the vendors we evaluated recommended using both methods, reflectance and transmission.  Unfortunately, using both methods increases the cost dramatically, often requiring two light sources and twice as many cameras.  We decided to employ transmission detection because it most effectively met our evaluation criteria.

Defect Processing Method
The two types of line-array cameras available are static (with raw data) and smart (with processed data).  A static camera receives a light intensity value from each of its pixels once every scan and passes this information to a computer or specially designed input card.  For each pixel or block of pixels, this data is processed and compared to the previous and adjacent pixel values by the card or computer.  The static camera has some advantages as compared to the smart camera.  It has fewer electronic components in the camera, therefore it can be used in more severe conditions.  Since the data processing is done outside the camera, an upgrade of the input card or server allows you to handle large volumes of data generated by the cameras.  However, high data rates require the use of traditional server operating systems such as Unix and can be a challenge to integrate into Windows NT or Novell-based networks.

Smart cameras are essentially static cameras with an on-board industrial grade computer.  The defect detection algorithm is processed inside the camera, and defects are reported back to the server.  The server is used to capture and store defect data, display maps and lists of defects, send parameters for detection to the cameras and serve data to workstations on the network.  The greatest advantage of a smart camera system is simplicity; with no special cards, everything is processed in the camera or PC.  Thus, it doesn't take a lot of hardware and an engineer to operate the system.

Detection Algorithm
Each pixel on a line-array digital camera reports a value relative to the intensity of light it registers.  Most cameras utilize a 256-color gray scale, where no light registers as 0 and full light is 255.  The camera is set-up so that the product absorbs enough light to give an average intensity of 127.  All of the digital line-array camera vendors we looked at offered systems that use thresholds to identify defects; the user defines low and high thresholds.  If the video level of any consecutive minimum number of pixels passes the threshold, a defect is reported.  A drip that registers at an intensity of 99 (where the low threshold is 100) will be reported as a defect.

Thresholds would be a perfect solution if nonwovens were perfectly uniform.  Typical nonwoven products are made of interlaced fibers where voids are arranged randomly.  In a given sample, there may be tiny clumps and voids that are larger than a single pixel.  Thus in addition to thresholds, most camera vendors allow the user to also specify a size requirement of length and width in addition to an intensity threshold for each defect type.

Size requirements coupled with thresholds would be a consistent solution again, if nonwoven defects were consistently uniform. However, nonwoven defects are normally a random aggregate of small defects.  For example, consider a camera system set up to report an area of 0.26 inches long by 0.26 wide that surpasses a threshold limit of 150 as a defect.  A void that is 0.26 by 0.26 inches passes by the camera and registers as an intensity of 255.  In that region there are a few filaments that cross the void with an intensity of 120.  hose few filaments fragment the defect into several smaller defects, none of which meet the minimum size requirements.  To the camera, there is no defect.

To overcome the limitations of thresholding and size requirements, a simple solution is to 'defocus' the camera.  Defocusing makes pixels overlap, thereby softening the effect of small anomalies like single filaments in a void.  But why pay a premium for superior resolution only to negate it by purposefully sabotaging accuracy?  The LASOR/Systronics company developed a much more eloquent approach to detecting small aggregates and eliminating false defects in nonwovens: two-dimensional filtering.  To detect aggregate defects, it is best to average the intensity for a certain length and width and then apply the thresholding and size requirements.  In the example above, if the signal was filtered so that the intensity of each pixel in a 10 x 10 area was summed and then divided by 100, and repeated for each pixel across the web, the void would probably have an intensity of about 230 and an area of about 0.3 by 0.3 inches, making it a defect.  We found two-dimensional filtering crucial in detecting voids in our thermally bonded products, where a good portion of the web is made up of small regions of thin film that register an average intensity well above the rest of the product.  Without filtering, each bond point would have been reported as a defect.

With two-dimensional filtering, we were also able to ignore regular uniformity variations while capitalizing on the high pixel resolution.  We simultaneously use filtering to negate bond points and no filtering to detect tiny specks of dirt.  This would be virtually impossible if we practiced filtering by defocusing.

Data Strategy
Once the camera system is in place and reliably reports defects, requests for access to the data will invariably be made.  If the method of data storage is incompatible with other networked systems in the plant, it can be very frustrating to meet these requests.
We evaluated proprietary databases, the standard two decades ago, and open databases like Oracle or SQL, all third-party systems maintained and supported by companies whose primary focus is software.  However, we determined that 'flat files' were optimal for our process.

Flat files are simple text files.  They contain a header that defines the data fields followed by the data corresponding to the header fields.  Most flat file data sets are delimited or separated by commas or tabs and can be easily imported into a spreadsheet, even by computing novices.  Usually, each product lot has a unique flat file that contains pertinent production information.  This is a simple and direct solution for storing and retrieving information about single lots, but it can make analysis of long-term trends more difficult.  On the other hand, integration of defect data into other information systems is relatively straightforward with flat files.  In proprietary business systems, it is often easier to import information from flat files than open databases.  The LSAOR/Systronics system utilizes flat files.  To overcome the long-term trending challenge, we closely integrated WebData with our continuous process historian so that time-based defect information is continuously archived and available for instant retrieval by operators and management.

Service
After wading through all of the technical complexities, learning the digital camera jargon and understanding what we needed to specify as system requirements, we closely examined vendor service.  Cerex is a lean manufacturer and looks for vendors with the resources, depth and internal talent to continue support of their products well into the future.  We found such a company I LASOR/Systronics.  They were not the largest vendor we evaluated; however, the more we interfaced with them and learned of their support systems and strategies, the more confidence we gained in their technical and service capabilities.

Success
We have achieved our objectives of building a WebData system that consistently finds and reports defects, provides an objective measurement of product uniformity, acts as a tool for the removal of defects and gives us the ability to automatically identify and correct defects.  As a result, our technicians are now targeting process optimization efforts, the engineering staff is examining ways to integrate alternative processes, and our manufacturing yields and efficiencies continue to tick upwards while our quality soars.


 

 

 

 

 

Comments:

There are currently no comments for this article.


Leave a Comment:

All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.

First Name: *
Last Name: *
Your Email: *
Your Comment:
Please check the box below and respond as instructed.

Search AIA:


Browse by Products:


Browse by Company Type: