Gerald Budd 0000-00-00 00:00:00
MACHINE VISION SYSTEMS DESIGNED FOR THE PURPOSE OF SURFACE INSPECTION HAVE PROGRESSED CONSIDERABLY IN THE LAST FIVE YEARS. Human inspection procedures—though still more commonly used—cannot compare to the machine vision approach. A machine vision system in its simplest form is a computer, a camera (CCD or CMOS sensor), front end optics (lens), an illumination source and software (operating system and machine vision algorithms). The method of moving the images from the sensor to the computer has a number of options, but lately the trend has been toward the use of less expensive cable formats such as GigE Vision or USB 3.0. Although there are a number of machine vision suppliers to choose from, the hardware performance of each is pretty similar. The selection of the image processing software is much more important. The combination of the two is usually selected because someone in a facility is familiar with a particular brand or has used another product of a vendor in the past. However, certain applications are more dependent on the design and implementation of the machine vision algorithms for their success than the hardware used. Surface inspection requires good design and implementation of both hardware and software for a successful application. Inherent in the complexity of the surface inspection requirements as defined by the engineering specifications, the lower cost systems are usually excluded because of resolution and/or software limitations. Let’s not get the wrong impression, surface inspection applications are generally more difficult and require a more knowledgeable integrator to increase the likelihood of a successful implementation. A lot of machine vision applications can be tackled using smart camera or a compact vision system, but they are usually less difficult. Unless the engineer has a fair amount of free time available, surface inspection applications generally do not fall into the category of do-it-yourself projects. ADVANCES Machine vision systems designed for the purpose of surface inspection have progressed considerably in the last five years and the resolutions of the sensors used have improved dramatically. It is now common to find area scan sensors with resolutions of 16 to 29 megapixels, providing the ability to isolate imperfections as small as 50 square microns (ìm) when viewing a 600 millimeter by 480 millimeter component with a single image. Line scan sensors have also increased in resolution to as large as 16,000 pixels, creating images that can be more than 320 megapixels in size with the capability of resolving defects as small as 3.6 ìm. Most customers with seal integrity applications do not require this level of resolution. However, in order to pass a gage repeatability and reproducibility (GR&R) study with 10% or less it is better to have more resolution than not enough. The general rule of thumb of having 10 times the resolution of your maximum allowable defect size is applicable here. If your minimum size imperfection is 500 ìm then you should plan to have a sensor capable of resolving a 50 ìm object in the field of view. The two-megapixel and fivemegapixel sensors are useful for inspection of smaller surface areas or when mounted to the end of a robot. It is usually more cost effective to move a single sensor to various positions around a large component than it is to mount multiple sensors in fixed positions. This is especially true when multiple components are processed on the same production line. It is unlikely that the features that are to be inspected on the various components will have the same alignments. The robotic approach will also allow the user to modify the inspection positions should an engineering change require a different inspection. As with any machine vision application, the front-end optics are critical to the success of the project. The front-end is normally defined as the sensor, lens, filters and lighting. In order to achieve the maximum resolution from a large format sensor we must also consider the lens used to acquire the image. It is expected that the lens selected for the application provides a flat field with minimum distortion in all regions of the image. This may be difficult to achieve using larger pixel format sensors. In general, longer focal length lenses have less distortion than short focal length lenses. However, this requires that the inspection sensor be placed further from the inspection surface. The ideal inspection setup is for the sensor to observe the same number of pixels for a given size defect regardless of its position within the field of view of the optical system. This isn’t always possible, so either a calibration algorithm must be employed or an elaborate masking routine that compensates for defect sizes based on position must be employed. It is better to have a calibration routine that corrects for all systematic errors in a complete front-end optical system. LIGHTING When attempting to inspect large machined components such as a transmission housing, engine heads or deck faces, the lighting must be carefully engineered. When using an area sensor the inspection of a large area requires that the lighting conditions are consistent over the entire surface. This is best performed using diffuse light, which will minimize the effects generated by the cutting tools and material surface texture. Directional lighting is difficult to implement over a large surface area and can result in “hot spots” caused by machining operations. Since the inspection will be performed in the plant environment it is recommended that a high frequency fluorescent system with feedback intensity control and operation above 25 Khz should be used so that a short integration time can be used with the electronic shutter of the sensor. This will minimize most of the effects caused by mechanical vibration during image acquisition. As an alternative LED lighting can be used; however, it should be energized prior to image acquisition so that the output intensity is stabilized and is consistent from image to image. It should be noted that fluorescent lamp replacement is much less expensive than LED lamp replacement but LEDs have a longer useful life. The ideal lighting setup is described in U.S. Patent 7,394,530 B2 entitled “Surface Inspection Technology for the Detection of Porosity and Surface Imperfections on Machined Metal Surfaces.” This details that the illumination surface (what the front-end optic system sees) should be at least twice as large as the object being inspected in all dimensions. This dimensional measurement does not include the area required for the illumination source itself, typically 80 to 200 millimeters in width and 2450 millimeters in length if fluorescent lamps are used. Using the lighting technique mentioned earlier, Figure 1 shows the typical grayscale variation that can be achieved. Although variations in the surface uniformity can be observed (these are a result of the machining operations), the surface to be inspected is well defined and defective regions are easily identified. In the case that the components can be translated by an inspection point with consistent velocity or an encoder can be implemented to track the motion, one should consider using a line scan sensor for image acquisition. It is important that the displacement between each line scan is consistent to minimize image distortion and so that the size of a defect can be calculated correctly. If the dimensional measurement of features is important then an encoder circuit must be employed. The orientation of the line scan sensor and the illumination source will be fixed and therefore the image of the surface will appear uniform. This fixed orientation between the sensor and the illumination source will dictate the use of either a bright field or a dark field vision algorithm. If the surface imperfections dictate more than one type of illumination technique be used it may be necessary to utilize more than one inspection station. The bright field technique is useful for the isolation of porosity and non-cleanup on machined surfaces whereas the dark field technique is better at the isolation of dings, scratches and uneven surfaces. THE IMPORTANCE OF SOFTWARE Probably the most important component in a machine vision system is the software. There are two major parts to the software: the machine vision algorithms and the operator interface or human machine interface (HMI). The machine vision algorithms consist of a compilation of low level function structured to extract features that should or should not be present in the image. A good surface inspection algorithm will automatically adjust for slight variation in the surface texture and material color. The customer may request multiple inspection zones with different tolerance levels in each zone. A typical seal integrity inspection application will have a tight tolerance in the seal bead region and a more forgiving tolerance away from the seal region. Sophisticated engineering specifications will require that a defect not be larger than a percentage of the seal width. The width of the seal region may change in different areas of the component, making a simple defect size measurement an invalid testing procedure. The better vision algorithms will automatically calculate the percentage of seal width based on the defect position on the surface. The best machine vision system is really only as good as the operator interface. If the user cannot understand how the system functions or cannot find a method to change the sensitivity of a tool the system will not find optimal usage. The operator interface is the dashboard of the inspection system and it should be organized so that it is logical and easy to use. The machine vision system that is used for surface inspection applications is usually more complicated than a system that checks for part presence or presence of a feature. This is because it usually requires more inspection tools and a longer sequence of operations in order to perform the complete inspection process. A good operator interface allows the user to make the necessary program changes without actually having to edit the vision algorithms. The programs are designed to pass parameter variables between the operator and the vision algorithm in such a way that the user feels comfortable with making the change and the confidence that the correct parameter was modified. REPEATABLE RESULTS Generally speaking one the most important selling features of an automated machine vision inspection system is the fact that it can produce consistent inspection results over an extended period of time. Unlike the human inspector, a machine vision system will yield a non-subjective inspection, 24 hours a day, seven days a week, provided that the equipment is maintained and functions at the same level as when the vision algorithms were originally tuned. This is an important factor in the successful implementation of the automated inspection system, the system must be maintained! The inspection system is a complex compilation of electrical and mechanical components engineered to work together. All of the pieces must function as designed or the system will not properly function and produce an undesirable result. Plant personnel want a system that requires as little maintenance as possible or is completely maintenancefree. In most cases such an inspection system does not exist except those for those found in the cleanest environments. The typical facility that is machining metal components has dirt, dust, oil, cutting fluids or a combination of all of these. Equipment gets dirty and requires routine cleaning and maintenance in order to remain in top operational form. If designed correctly and maintained properly the machine vision system can provide consistent inspection results even if there are slight variations in the machined surface that is to be inspected. For more on this topic, see the author’s sidebars on Human Inspection and Lighting at www.visionsensorsmag.com. Gerald Budd is president of Phoenix Imaging (Livonia, MI). For more information, call (248) 476-4200, e-mail firstname.lastname@example.org or visit www.phoeniximaging.com. tech tips Surface inspection applications are generally more difficult and require a more knowledgeable integrator. A lot of machine vision applications can be tackled using smart camera or a compact vision system, but they are usually less difficult. Surface inspection applications generally do not fall into the category of do-it-yourself projects.
Published by QualityMagazine. View All Articles.
This page can be found at http://digital.bnpmedia.com/article/Quality+Inspection+On+The+Surface/1147819/122831/article.html.