High definition imaging for web inspectionIntroduction Web inspection has become commonplace in the paper making and flat metal sheet production industry. Inspection systems are today overwhelmingly based on CCD linear or matrix array cameras. The capabilities of present day technology have made it possible to image the web in real-time. The growing quality requirements of reprocessing mills and end product users are constantly pushing web inspection system developers to increase their systems’ spatial and grey scale resolutions. However, the utilization of more sensitive detection settings for finding ever smaller and more subtle defects results in increasing the information flow to a level where only an intelligent solution can cope with all of it. In addition, defect types vary greatly in size and nature. Normally, a single solution is not capable of finding all these defects. Therefore, special cameras with dedicated software have been used instead of standard hole/spot cameras if continuous streaks must be detected. The subtle defects, such as faint wrinkles, were insurmountable obstacles for many inspection systems. ABB has recently released Web Imaging Solution HDI 8. One of the most salient features of this new system is the camera that tackles the above-mentioned challenges with a single-sensor approach. Web inspection systems Simple system architecture, high definition imaging, and high quality classification have been the main goals of the system design. The system architecture is shown in Figure 1. All systems contain at least one inspection frame and a computer system. The inspection frame consists of a camera beam with several electronic cameras and a light source beam for providing the necessary illumination on the web (Figure 2). The computer performs the high level data processing, reporting and other human/machine interface tasks. The differences between various solutions can be found in the front end and intermediate data processing. Many system integrators utilize simple cameras that, for example, only transform the optical image into an analog signal without any data processing. The image capturing is done by so-called frame grabbers, and the following data processing is done by dedicated hardware in the form of a standard PC or special hardware. However, the PC can handle only two cameras due to the huge processing power needed by the image analysis. If some ten or more cameras are necessary for the required resolution, which is quite normal, several PCs are needed for this task only. Another solution is to use generic image processing boards, but the systems still require a lot of hardware, e.g. an electronic cabinet containing these boards can support only four cameras. It is obvious that adding hardware makes the systems complicated and adds risks for a device failure. Instead, the HDI 8 system is relying on hardware that has been specially designed for the web inspection (not for a generic image processing) task. As a result, more has been gained with less: more processing power, more reliability, and more simplicity with less hardware. The new simple but extremely powerful cameras provide refined information to a single intelligent communication module which, after certain processing tasks, forwards the results to the host computer. The data processing is thus divided on three hierarchy levels focusing the capacity where it is most efficiently used to guarantee the maximal total throughput.   New high definition camera The digital camera of HDI 8 is shown in Figure 3. When comparing it to the previous generation models (see e.g. [1]) the most prominent visual feature is the mechanical compactness and simplicity. In spite of that, the camera is intelligent, which means that it not only captures images but also detects defects and performs various image analysis and classification tasks. The new architectural solutions have made it possible to squeeze the camera size and increase its speed, processing power and memory capacity at the same time. The cameras utilize Ethernet technology for messaging under the control of a common camera communication module. To ensure interference-free operation, the connection between the inspection frame and the camera communication module is an optical fiber link.  The processing power of HDI 8 camera is based on combining the hardwired logic speed and reprogrammability of the latest FPGA technology with the flexibility of a Power PC – type general processor. The capacity of modern FPGAs makes it possible to run various different algorithms in parallel, so that common discrete defects, subtle defects and streaks can be detected at the same time by a single camera. The camera architecture also provides the possibility to load completely new camera programs for the FPGAs and the Power PC at an installation site – even remotely – if necessary. This guarantees an easy upgrade path without the need to make any changes to the hardware, e.g. acquiring new boards, changing memory modules etc. This feature may be useful if the mill is making changes to the process or is starting the production of a new grade with requirements which were not anticipated when the decision to purchase the web inspection system was made. Obviously this feature makes also smaller modifications a snap. There are several factors that have to be considered when striving for the best possible performance when detecting the most subtle defects. The weakest link of the whole imaging sequence determines the total capacity, therefore not a single item can be neglected. Selecting linear arrays of the latest CCD technology provides the web inspection system with high speed, good sensitivity, large dynamic range, anti-blooming, low noise, uniform PRNU, etc. For example, the pixel clock frequency of the HDI 8 camera is 50 MHz, which provides a scan time of 20 microseconds at its best. However, the advantages of a high quality CCD array can be lost if the electronics and optical design is not comparable with it. It is also obvious that high quality lenses are needed to minimize various optical aberrations Utilizing camera capabilities ABB has taken a lot of care to design a camera that fully utilizes the above-mentioned features. For example, all CCD arrays have an unavoidable dark signal that is highly dependant on the temperature and proportional to the integration time. It is obvious that this signal must be compensated for. A standard solution is to use a static value which corresponds to the normal operating conditions. However, in the process industry, a constant ambient or cooling air temperature is hard to guarantee. Therefore, the HDI 8 camera utilizes a more advanced method: the dark signal level is measured and compensated for on every scan. The curvature of the cross web signal and the need for its linearization is very well known. The imaging geometry and imperfections in optics are the reasons for it. In addition, the unevenness of illumination, dust build-up, differences in pixel sensitivities, etc. are other causes of non-linearities. All these problems can be corrected by a well-designed AGC. However, if the difference between the edge and center pixel signal levels is large, the real gray-scale resolution of the edge pixels remains poor because the amplification cannot make the initial quantization any better, i.e. to create new information. Therefore the average raw signal level must be brought to the middle of the camera’s sensitivity range and efforts have to made to get it initially as straight as possible. Here, proper lighting is the key. Because the web to be inspected is moving fast and the goal is high spatial imaging resolution, the exposure time should be as short as possible. However, the full speed of the camera cannot be utilized if the brightness of the light is not adequate. This is not only a question of how strong the radiation is, but also how well its spectral sensitivity curve complies with that of the CCD array. Fluorescent tubes and LED light sources are commonly used when low to medium amount of light is enough, typically with reflection measurement, but for example when high-opacity grades are to be measured in transmission mode, the halogen lamps show their strength due to the both of the above mentioned reasons. For the best results in varying circumstances and with different web grades, automatic control of the light level is necessary A large contrast range (i.e. the ratio of the brightest and darkest gray scale values) and a high gray scale resolution are also desirable features. The CCD arrays available today can manage the contrast range of 1:6000 or more. If, in addition, the A/D converter has a high resolution such as 12 bits (4096 gray levels), differentiating between the defects whose gray levels are close to each other or even saturating, becomes easier. For example, separating holes from spots caused by various liquid drips becomes more reliable. More efficient tools for web producers When imaging precision gets better, it provides new possi-bilities to classify the defects more accurately than before. Neural network classifiers are the tools which give the web material producers easy-to-use and intuitive tools to name and classify the defects the way the users actually know them [2, 3]. To fully utilize the potential provided by the technology described above, a web inspection system user interface has to be such that it serves the mill personnel with the essential information which helps them to meet the customers’ requirements and the mill’s productivity tar-gets. Therefore, a system has to give an easy-to-grasp gen-eral view of the process and correct detailed information on the essential exceptions. References 1. J. Rauhamaa. “Paper web inspection with intelligent line scan cameras”. Machine Vision News, Vol. 6, 2001.
|
