Data processing: generic control systems or specific application – Specific application – apparatus or process – Product assembly or manufacturing
Reexamination Certificate
1999-09-20
2003-03-18
Picard, Leo (Department: 2125)
Data processing: generic control systems or specific application
Specific application, apparatus or process
Product assembly or manufacturing
C700S109000, C700S121000, C382S149000
Reexamination Certificate
active
06535776
ABSTRACT:
CROSS REFERENCE TO RELATED APPLICATIONS
(Not Applicable)
BACKGROUND OF THE INVENTION
1. Technical Field
This invention relates to the field of process step characterization and more particularly to a method and system for automatically localizing and isolating an errant process step using a content-based image retrieval engine and a historical defect characterization database.
2. Description of the Related Art
Recently, automation has played an increasingly more important role in manufacturing process monitoring, characterization, and control. Particularly, the availability of inexpensive, high-performance computing platforms in combination with digital signal processing technology has spurred the broad-based adoption of automatic defect inspection and classification systems for on-line, in-line and off-line product inspection and review. Automatic defect classification system typically are used to assess product quality at various points in the manufacturing process. The intent is to measure product quality on-line or in-line to facilitate the rapid discovery of errant process steps that are detracting from product quality. Consequently, automated inspection systems are required to reduce the need for a human operator to monitor the output of the inspection process.
Automated defect classification systems typically monitor a single process step and result in detecting, classifying, and reporting a pre-specified defect type. The defect pre-specification is determined based upon a training method related to inputting into the defect classification system a limited number of defect types associated with the process step under inspection. The defect classification system subsequently can attempt to detect and classify similar defects when placed in operation.
Notably, the placement of defect classification systems in multiple positions throughout an assembly line in manufacturing processes such as textile formation, textile printing, technical ceramic component manufacturing, printed circuit board manufacturing, and integrated circuit manufacturing generally results in the collection of an enormous volume of defect-related data. In particular, the defect-related data includes product images, defect classifications, tool associations, lot numbers and product specifications. In consequence, manufacturers store the collected data in large databases in an attempt to maintain a historical record of product and process quality issues. More particularly, manufacturers store the collected data with the intention of subsequently querying the database with present defect data in order to localize, isolate, and correct an errant manufacturing process step.
Notwithstanding, present methods cannot efficiently retrieve historical defect data from a database as a result of the enormous quantity of data stored therein. Specifically, present methods lack the ability to efficiently classify and retrieve defect imagery from a large defect database. Rather, present methods include either inaccurate computer retrieval of inconsistently characterized defect imagery or slow manual retrieval of defect imagery. Thus, an unsatisfied need exists for an efficient automated method for data mining defect imagery that can be used to statistically localize and isolate an errant process step.
SUMMARY OF THE INVENTION
A method for localizing and isolating an errant process step in accordance with the inventive arrangement satisfies the long-felt need of the prior art by integrating content-based image retrieval [CBIR] with a large, sorted database of defect imagery and corresponding defect characterization data to efficiently diagnose a defective product and identify an errant process apparatus causing the defective product. Using a CBIR engine, the present invention can compare image content extracted from a query image to image content extracted from a group of images. Generally, imagery sharing similar image content represent similar manufacturing conditions. Thus, the present invention extends manufacturing-specific CBIR by determining an estimate of a conditional probability that associates the query image with an errant tool and defect type. The inventive arrangements have advantages over all defective process step identification techniques, and provides a system and apparatus for errant process step identification based on content-based image retrieval.
A method for localizing and isolating an errant process step comprises the steps of: retrieving from a defect image database a selection of images, each image having image content similar to image content extracted from a query image depicting a defect, each image in the selection having corresponding defect characterization data; deriving from the defect characterization data a conditional probability distribution of the defect having occurred in a particular process step; and, identifying a process step as a highest probable source of the defect according to the derived conditional probability distribution.
In particular, the retrieving step comprises the steps of: providing to a content-based image retrieval engine, a query image depicting a defect; retrieving from the content-based image retrieval engine, a selection of images, each image having image content similar to image content extracted from the query image; and, ranking the selection of images according to a similarity metric.
The deriving step comprises the steps of: calculating a process step conditional probability distribution from the defect characterization data; calculating a defect class conditional probability distribution from the defect characterization data; selecting a process step included in the process step conditional probability distribution having a highest probability; selecting a defect class included in the defect class conditional probability distribution having a highest probability; and, merging the selected process step with the selected defect class to produce a probable source process step of the defect. Subsequently, the identifying step can comprise the steps of: ordering a ranked list of probable source process steps of the defect; and, reporting to a user the ranked list, whereby the ranked list localizes and isolates a probable source process step of the defect.
A method for process step defect identification comprises four steps. First, product anomalies can be characterized. Specifically, the product anomalies can be detected by an imaging system. Moreover, the characterizing step can comprise the steps of: detecting a product anomaly passing within imaging range of a computer vision system; forming an image of the product anomaly; assigning defect characterization data to the formed image; and, storing the formed image and the assigned image defect characterization in a defect characterization database, whereby each formed image and assigned image defect characterization in the defect characterization database can be retrieved for subsequent comparison to a query image depicting a product anomaly. Significantly, the assigning step comprises the step of assigning to the acquired image defect a defect class label and an associated process step. Furthermore, the storing step comprises the step of sorting each acquired image and assigned defect characterization data in a hierarchical search tree structure using an unsupervised clustering algorithm.
Second, a query image of a product defect can be acquired. The acquiring step can comprise the step of generating an image of a type selected from the group consisting of optical imagery, laser scattering imagery, interferogram imagery, scanning electron microscopy imagery, and atomic force microscopy imagery.
Third, a particular characterized anomaly can be correlated with the query image. The correlating step can comprise the steps of: retrieving from the defect characterization database, a selection of images, each image having image content similar to image content extracted from the query image; and, ranking the selection of images according to a similarity metric. In particular, the ranking step can com
Ferrell Regina K.
Karnowski Thomas P.
Tobin, Jr. Kenneth W.
Jarrett Ryan
Picard Leo
Senterfitt Akerman
Ut-Battelle LLC
LandOfFree
Method for localizing and isolating an errant process step does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for localizing and isolating an errant process step, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for localizing and isolating an errant process step will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3078399