Automated learning of model classifications

• US 7,639,868 B1
• Filed: 06/16/2004
• Issued: 12/29/2009
• Est. Priority Date: 06/16/2003
• Status: Active Grant

• Current Assignee: Drexel University
• Original Assignee: Drexel University
• Inventors: Ip, Cheuk Yiu, Sieger, Leonard, Regli, William C.
• Primary Examiner(s): Chawan; Sheela C
• Application Number:
  US10/869,061
• Time in Patent Office: 2,022 Days
• Field of Search: 382/159, 382/228, 382/224, 382/225, 382/118, 382/190, 706/25, 706/45, 706/15, 706/20, 706/934, 707/102, 707/3, 707/E17.003, 707/100, 707/101, 707/6, 704/245, 704/9, 704/239, 704/246, 704/E15.014, 704/E17.003, 704/230, 704/240, 704/256, 704/232, 704/244, 704/243, 725/46, 725/9
• US Class Current: 382/159
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching

Abstract

A method of providing an automated classifier for 3D CAD models wherein the method provides an algorithm for learning new classifications. The method enables existing model comparison algorithms to adapt to different classifications that are relevant in many engineering applications. This ability to adapt to different classifications allows greater flexibility in data searching and data mining of engineering data.

Claims

1. A method of classifying solid models comprising the steps of:

using a processor, wherein the processor performs the steps of;

providing a plurality of training models;

determining a first set of values based on predetermined properties of said training models;

defining classifications based upon said first set of values;

providing a query model;

determining a second set of values based on predetermined properties of said query model;

comparing said second set of values to said classifications; and

determining at least one of said classifications and said query model based on said comparing step, wherein said query model and said training models are models of 3D objects or CAD models.

2. The method of claim 1, wherein said predetermined properties comprise a first set of distances between points selected on said training models or said query model.

3. The method of claim 2, wherein said first set of distances comprises a set if predetermined shape interrogation measures.

4. The method of claim 3, wherein said predetermined shape interrogation measures include IN distances, OUT distances, and MIXED distances.

5. The method of claim 3, wherein said IN distances, OUT distances, and MIXED distances are measured at point pairs placed on a surface of said training models or said query model to maximize discrimination.

6. The method of claim 5, wherein said point pairs are randomly selected.

7. The method of claim 5, wherein said point pairs are selected according to a predetermined grid function.

8. The method of claim 1, wherein said step of defining classifications includes a step of weighting said first set of values.

9. The method of claim 8, wherein said weighting step further comprises:

assigning one of said classifications for each of said training models; and

normalizing said first set of values for each of said assigned classifications.

10. The method of claim 9, wherein said assigned classification for each of said training models comes from a group consisting of shapes and functionality of said training models.

11. The method of claim 9, wherein said assigned classification for each of said training models comes from a group consisting of manufacturing processes to build said training models.

12. The method of claim 9, further comprising a step of adding a new one to said classifications for another one of said training models.

13. The method of claim 1, wherein said defining step comprises a step of forming shape distribution histograms.

14. A method of searching for a solid model comprising the steps of:

using a processor, wherein the processor performs the steps of;

providing a query model;

determining a first set of values based on predetermined properties of said query model;

comparing said first set of values to definitions for classification, wherein said definitions for said classification are based on a second set of values that is determined by predetermined properties for a group of training models; and

determining a classification of said query model based on said comparing step, wherein said query model and said training models are models of 3D objects or CAD models.

15. The method of claim 14, wherein said predetermined properties comprise a first set of distances between points selected on said training models and said query model.

16. The method of claim 15, wherein said first set of distances comprises a set of predetermined shape interrogation measures.

17. The method of claim 16, wherein said predetermined shape interrogation measures included IN distances, OUT distances, and MIXED distances.

18. The method of claim 16, wherein said IN distances, OUT distances, and MIXED distances are measured at point pairs placed on a surface of a training model or said query model to maximize discrimination.

19. The method of claim 18, wherein said point pairs are randomly selected.

20. The method of claim 18, wherein said point pairs are selected according to a predetermined grid function.

21. The method of claim 14, wherein said determining step includes a step of weighting said second set of values.

22. The method of claim 21, wherein said weighting step further comprises:

assigning one of said classifications for each of said training models; and

normalizing said second set of values for each of said assigned classifications.

23. The method of claim 22, wherein said assigned classification for each of said training models comes from a group consisting of shapes and functionality of said training models.

24. The method of claim 22, wherein said assigned classification for each of said training models comes from a group consisting of manufacturing processes to build said training models.

25. The method of claim 22, further comprising a step of adding a new one to said classifications for a training model.

26. The method of claim 14, wherein said determining step includes a step of forming shape distribution histograms.

27. A system for classifying solid models comprising:

a database for storing a plurality of training models and classification definitions, wherein said classification definitions are based upon a first set of values determined by predetermined properties of said training models; and

a host unit connected to said database for receiving a query model, and determining a second set of values based on said predetermined properties of said query model, said host determining a classification of said query model by comparing said second set of values to said classification definitions.

28. The system of claim 27, wherein said predetermined properties comprise a first set of distances between points selected on said training models or said query models.

29. The system of claim 28, wherein said first set of distances comprises a set of predetermined shape interrogation measures.

30. The system of claim 29 wherein the predetermined shape interrogation measures included IN distances, OUT distances, and MIXED distances.

31. The system of claim 29, wherein said IN distances, OUT distances, and MIXED distances are measured at point pairs placed on a surface of said training models or said query model to maximize discrimination.

32. The system of claim 31, wherein the point pairs are randomly selected.

33. The system of claim 31 wherein the point pairs are selected according to a grid function.

34. The system of claim 27, wherein said host adds new classification definitions to said database.

35. The system of claim 27, wherein said query model and said training models, are CAD models.

36. A method for training a system for classifying solid models, comprising the steps of:

using a processor, wherein the processor performs the steps of;

providing a plurality of training models;

selecting points on said training models, wherein said selected points comprise a first set of point pairs selected from points located on an interior of said training model, a second set of point pairs located on an exterior of said training model, and a third set of point pairs located on both said interior of said training model and said exterior of said training model;

determining a first set of values based upon distances between said first set of point pairs;

determining a second set of values based upon distances between said second set of point pairs;

determining a third set of values based upon distances between said third set of point pairs; and

defining classifications based upon said first set, said second set, and said third set of the values, wherein said training models are models of 3D objects or CAD models.

37. The method of claim 36, wherein said defining step includes a step of weighting said first, second, and third sets of values.

38. The method of claim 36, wherein said weighting step further comprises:

assigning one of said classifications for each of said training models; and

normalizing said first, second, and third sets of values for each of said assigned classification assigned.

39. The method of claim 38, wherein said assigned classification for each of said training models comes from a group consisting of shapes and functionality of said training models.

40. The method of claim 38, wherein said assigned classification for each of said training models comes from a group consisting of manufacturing processes to build said training models.

41. The method of claim 38, further comprising a step of adding a new one to said classifications for a training model.