System and method for cost-effective classification of an object under inspection

• US 6,738,450 B1
• Filed: 12/10/2002
• Issued: 05/18/2004
• Est. Priority Date: 12/10/2002
• Status: Expired due to Fees

• Current Assignee: Agilent Technologies, Inc.
• Original Assignee: Agilent Technologies, Inc.
• Inventors: Barford, Lee A.
• Primary Examiner(s): BRUCE, DAVID VERNON
• Application Number:
  US10/316,299
• Time in Patent Office: 525 Days
• Field of Search: 378/51, 378/54, 378/58, 378/62
• US Class Current: 378/58
• CPC Class Codes: G01N 21/95684 Patterns showing highly reflecting parts, e.g. metallic elements
• CPC Class Codes: G01N 23/044 using laminography or tomosynthesis
• CPC Class Codes: G01R 31/71 Testing of solder joints
• CPC Class Codes: G05B 2219/37217 Inspect solder joint, machined part, workpiece, welding result

Abstract

The system and method enable cost-effective classification of objects, such as solder joints, that are under inspection. A classifier is operable to receive a feature vector f of an object under inspection and compute a probability that such object is properly assigned membership in a first of a plurality of different classes. For instance, a “good” class may be available for objects that meet a pre-determined criteria, and a “bad” class for objects that do not meet such criteria. The classifier analyzes f to compute the probability that an object is properly classified as “good.” Embodiments of the invention enable the classifier to be tuned as desired for a given inspection section for proper risk management. A cost variable is input to the classifier and used to determine whether, given the computed probability that an object is “good,” it is cost-effective to classify the object as “good.”

Claims

1. A method for classifying an object as being a member of one of a plurality of classes, the method comprising:

computing a probability that an object under inspection is properly assigned membership in a first of a plurality of classes; and

using a cost variable to determine whether it is cost-effective to assign the object to said first of said plurality of classes given the computed probability.

2. The method of claim 1 wherein said object under inspection comprises a solder joint.

3. The method of claim 1 wherein said plurality of classes comprises a good class for objects that meet a pre-defined criteria and a bad class for objects that do not meet said pre-defined criteria.

4. The method of claim 3 wherein said first of said plurality of classes comprises said good class.

5. The method of claim 3 wherein said cost variable comprises a variable $K=\frac{C_{\mathrm{GB}}}{C_{\mathrm{BG}}},$

wherein C_GB is a cost associated with incorrectly classifying a good object as bad and C_BG is a cost associated with incorrectly classifying a bad object as good.

6. A method for classifying an object under inspection, said method comprising:

inputting a feature vector for an object under inspection to a classifier for said object;

inputting a cost variable to said classifier that specifies a cost associated with incorrectly classifying said object under inspection in a first of a plurality of classes;

said classifier using said feature vector to compute a probability that said object under inspection is properly assigned membership in said first of a plurality of classes; and

said classifier using said cost variable to determine whether it is cost-effective to assign the object to said first of said plurality of classes given the computed probability.

7. The method of claim 6 wherein said object under inspection comprises a solder joint.

8. The method of claim 6 wherein said plurality of classes comprises a good class for objects that meet a pre-defined criteria and a bad class for objects that do not meet said pre-defined criteria.

9. The method of claim 8 wherein said cost variable comprises a variable $K=\frac{C_{\mathrm{GB}}}{C_{\mathrm{BG}}},$

wherein C_GB is a cost associated with incorrectly classifying a good object as bad and C_BG is a cost associated with incorrectly classifying a bad object as good.

10. A method for inspecting an object, said method comprising:

constructing a classifier for a first type of object that is operable to compute for an object of said first type that is under inspection a probability that such object under inspection is properly assigned membership in a first of a plurality of different classes; and

inputting a cost variable to said classifier to tune said classifier for cost-effective classification of said object of said first type that is under inspection.

11. The method of claim 10 wherein said cost variable specifies a cost associated with incorrectly classifying said object of said first type that is under inspection in said first of a plurality of classes.

12. The method of claim 10 further comprising:

using the cost variable to compute, for each of the plurality of different classes, a cost associated with assigning the object under inspection to the respective class based at least in part on the determined probability that such object is properly assigned membership in the respective class; and

determining the most cost-effective one of the plurality of different classes to which to assign the object under inspection based at least in part on the computed costs associated with each of the plurality of different classes.

13. The method of claim 10 wherein said first type of object comprises a first type of solder joint.

14. The method of claim 13 wherein said first type of solder joint comprises one selected from the group consisting of:

J-lead solder joint, gullwing solder joint, and ball grid array (BGA) solder joint.

15. The method of claim 10 wherein said first type of object comprises a first type of electrical connection.

16. The method of claim 15 wherein said first type of electrical connection comprises one selected from the group consisting of:

solder joint, conductive epoxy, mechanical bond, tungsten bond, and eutectic bond.

17. A system for classifying an object under inspection, said system comprising:

at least one processor; and

computer-executable code stored to a computer-readable medium, said computer-executable code executable by said at least one processor to

receive a feature vector for an object under inspection,

receive a cost variable that specifies a cost associated with incorrectly classifying said object under inspection,

use said feature vector to compute, for each of a plurality of classes, a probability that said object under inspection is properly assigned membership in the respective class, and

use said cost variable to determine the most cost-effective one of the plurality of classes to which to assign the object given the computed probability of each of the plurality of classes.

18. The system of claim 17 wherein said computer-executable code executable to use said cost variable to determine the most cost-effective one of the plurality of classes comprises:

computer-executable code executable to use the cost variable to compute, for each of the plurality of different classes, a cost associated with assigning the object under inspection to the respective class based at least in part on the computed probability that such object is properly assigned membership in the respective class; and

computer-executable software code executable to determine the least costly one of the plurality of different classes to which to assign the object under inspection.

19. The system of claim 17 further comprising:

input means for inputting said cost variable to said system.

20. The system of claim 17 further comprising an image processing means for generating said feature vector from at least one image of said object under inspection.

21. The system of claim 17 wherein said object under inspection comprises a solder joint.