Image analysis – Applications – Manufacturing or product inspection
Reexamination Certificate
2002-04-22
2003-09-02
Patel, Jayanti K. (Department: 2625)
Image analysis
Applications
Manufacturing or product inspection
C382S147000, C382S155000, C382S171000, C382S190000, C382S225000, C382S276000, C382S305000, C716S030000
Reexamination Certificate
active
06614925
ABSTRACT:
TECHNICAL FIELD OF THE INVENTION
The present invention relates generally to the fields of image processing, pattern recognition, and machine vision, and more particularly, to a system and method, including software programming, for automated dynamic image recognition for the detection and classification of defects on printed wiring boards using visual imaging and analyses of the visual imaging.
BACKGROUND OF THE INVENTION
Most manufactured products go through quality control inspection cycles in an attempt to reduce the number of defective products that leave the manufacturing facility. Especially in industries incorporating electronic printed wiring assemblies (PWAs), visual inspections are often performed at several points in the manufacturing cycle of the PWA. For PWA inspections, component level defects that are examined for by visual inspection include missing, misoriented, misaligned, damaged components, and foreign material.
Automated inspection of objects, including PWAs, has been tried in many manufacturing industries to attempt to 1) reduce the number of defects that escape detection and false defects reported in human visual inspections, 2) reduce the time required to perform inspections and therefore reduce the overall manufacturing cycle time, and 3) to reduce the labor costs associated with manual visual inspections. Machine vision systems have become more widespread in industries requiring rapid inspection of objects to determine whether the object is spacially correct. Several automated inspection methodologies have been commercialized, with varying degrees of success, for PWA inspection to detect component level defects. These include golden board correlation, template correlation, pattern recognition with fixed feature sets, and statistical feature models. Very often, specialized hardware and software techniques are used in conjunction with these methods to increase the proficiency of these methodologies.
However, each of these previously developed methodologies proves inadequate in a real world inspection scenario because each are predisposed to the visual characteristics of their “good board” models (e.g., geometric patterns or statistical models). Thus, in order to maintain effective defect detection, the PWAs inspected must remain fairly consistent. However, the image of a PWA and particular areas of a “good” PWA can change from run to run and even within a single run of boards (e.g., if a different vendor component is used). In reality, there are countless variations of PWAs that are correctly assembled. These conventional systems do not adequately compensate for the real world variations that will still result in a good board.
Furthermore, each of these previously developed systems and methods require an extensive amount of programming expertise to set up and maintain the inspection systems. The programming must typically include elaborate image processing algorithm configurations for each new part, which, depending on the algorithm used, can include as many as 150 distinct parameters. Typically, these complex problems cannot automatically switch between part types, the part libraries cannot be ported to new PWAs or to other inspection systems, and each new board can require in excess of 80 hours of initial set-up time.
As mentioned, specialized hardware and/or software programming is often used to reduce image variability for these previously developed inspection systems to improve the system performance. However, special hardware and/or software programming typically 1) increase the complexity for the user and 2) limit the scope of applicability for the inspection system. For example, the modifications may make one class of defect easier to detect, while reducing the effectivity or even eliminating the ability to detect other classes of defects. Thus, in order to inspect for all relevant component defects, several modifications must be made and the inspection system must image the same PWA multiple times to get full defect coverage, thus increasing both the complexity and the cost of the inspection system.
For example, some previous systems are model-based approaches that use software programming to model the object under inspection (such as a PWA with components). Model-based approaches will often build a very exact model of the components by outlining the edges of components on the PWA, perhaps including an edge-enhancement algorithm, and performing a connectivity step to form a line segment. However, as the lines (edges) become more obscure (due to gaps or less distinct gray-scale differentiation between pixels), the software programs and algorithms become significantly less effective and therefore increasingly more complex and expensive.
As shown, these previously developed inspection systems tend to be non-robust and inflexible. Accordingly, these systems consistently produce high numbers of false alarms (i.e., indicating defects when none exists) which cannot be reduced to acceptable levels without causing high numbers of false passes (i.e., indicating a PWA is good when it has component defects). In addition, these systems tend to have relatively high operating costs, maintenance costs, and operator training costs.
Additionally, previously developed machine vision inspection systems, most (if not all) simply determine whether a part is present or absent and some make a determination of polarity. However, these systems do not perform an additional analysis of the presence/absence determinations to sub-classify defects to determine not only whether the component is present, but also whether the component is misoriented, misaligned, damaged, or surrounded by foreign material. While these prior art systems can often identify a defect, they cannot classify the defect to aid in the troubleshooting and improvement of the PWA process.
SUMMARY OF THE INVENTION
The present invention provides an image recognition system and method that substantially eliminates or reduces disadvantages and problems associated with previously developed inspection systems and methods that utilize imaging technology, including those systems and methods used for inspection of printed wiring boards.
The present invention provides an image recognition program stored on a tangible medium for classifying and subclassifying regions of interest on an image. The image recognition program can be used in an image inspection system to determine defects on objects such as printed wiring boards.
More specifically, the image recognition program is executable to collect raw image data, segment out rectangular regions of interest that can be component sites defined by CAD data, preprocess each region of interest by scaling, gain and offset correction, and gamma correction, generate a set of image spaces for each region of interest using a set of spatial image transforms, generating feature scores on the image spaces, scoring the features, comparing the features to a knowledge base of feature scores to make a class determination for the features, generating a presence/absence decision confidence for the features, calculating a class determination and decision confidence for each region of interest. This will determine if the component within the region of interest is correct or defective including whether a part is present or absent and whether it has the correct polarity.
Another aspect of the image recognition software includes the ability to perform a relative confidence determination to flag defects and extended relative confidence to subclassify the defects. In another aspect of the present invention, class determination, decision confidences, relative decision confidences and extended relative decision confidences can be added to the existing knowledge base to “train” the knowledge base to accept or reject additional defect classes and subclasses.
The present invention provides an important technical advantage by providing an automated object inspection system that, in addition to determining whether a component is present and has the correct polarity, can also determine various subcla
DeYong Mark R.
Eskridge Thomas C.
Grace John W.
Newberry Jeff E.
Chawan Sheela
Gray Cary Ware & Freidenrich LLP
Intelligent Reasoning Systems, Inc.
Patel Jayanti K.
LandOfFree
System and method for performing basic training does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for performing basic training, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for performing basic training will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3091071