1. Draft Additive Manufacturing File Format Hod.lipson@cornell.edu Chair, ASTM F42/Design Task Group on File Formats July 8, 2010 With Jonathan Hiller Disclaimer: Information in this presentation does not constitute the final standard. Actual specifications is subject to change until finalized by ASTM

2. Goals Identify needs of new file format Propose new standard  Reach consensus Catalyze adoption

3. Outline of presentation Summarize survey Outline proposed standard Process and timetable for moving forward

4. Prioritized features from survey n=162. Error bars = Standard Error

5. Proposed Format

6. Key considerations Technology independence – Describes target object, not how to make it – Every machine can make it to the best of its ability Simplicity – Easy to understand and implement Scalability – Can handle complex objects, microstructures, repetitions Performance – File size, read/write time, processing, accuracy Backwards compatible – Can covert to/from STL without additional info Forward compatible – Easy to extend new features in the future

7. Name AMF – Additive Manufacturing Format – Additive Manufacturing File

8. XML Meta-format: Format of formats – Text based – Easy to read/write/parse – Existing editing tools – Extensible – Highly compressible Mentioned by a number of constituents – E.g. Materialise – Based on work by J. Hiller (Cornell)

9. General Concept Parts (objects) defined by volumes and materials – Volumes defined by triangular mesh – Materials defined by properties/names Color properties can be specified – Color – Texture mapping Materials can be combined – Graded materials – Lattice/Mesostructure Objects can be combined into constellations – Repeated instances, packing, orientation

10. Basic AMF Structure 0 1.32 3.715 0 1.269 2.45354... 0 1 3 1 0 4... Addresses vertex duplication and leaks of STL

11. Compressibility Comparison for 32-bit Floats; need to look at double precision

12. File Size Stored either as text or compressed (zip) Both versions have AMF extension Reader can determine which and decompress during read

13. Read/Write/Parse time Write (seconds) Read + parse + construct data structure (seconds) Still negligible compared to slicing/processing time

14. Increasing Geometric Accuracy Flat triangles do not scale well for complex geometry, esepcially: – Curved surfaces – Microstructures Typical objects require millions of triangles – 10M triangles not uncommon Likely to get worse with increasing printer resolution – 10cm sphere at 10  m requires 20,000 triangles

15. Geometric fidelity is a high priority Must be addressed

16. CURVED PATCH (Curved using vertex normals) PLANNAR PATCH Optionally add normal/tangent vectors to some triangle mesh edges to allow for more accurate geometry. CURVED PATCH (or curved using edge tangents) Curved patches

17. ... 0 0.707... 0 0.577 1 0.707 0 0.707......... Only needed for curved surfaces Only needed for curved edges (rare)

18. Recursive Triangle Subdivision Importer temporarily subdivides each curved triangle into a set of 4 n planar triangles then uses those to calculate slice

21. Icosahedrons (20 Triangles)

22. Flat triangles, error = 10.26% of diameter Error

23. Icosahedrons (still 20 triangles) One subdivision, error = 3.81%

24. Icosahedrons (still 20 triangles) Two-fold subdivisions, error = 1.49%

25. Icosahedrons (20 Triangles) Three-fold subdivisions, error = 0.84%

26. Icosahedrons (still 20 triangles) Four-fold subdivisions, error = 0.67%

27. Icosahedrons (still 20 triangles) Five-fold subdivisions, error = 0.635%

28. Icosahedrons (still 20 triangles) Six-fold subdivisions, error = 0.625%

29. Curving the triangle patches using surface normal reduces error #-fold subdivisions Error on units sphere

30. Double Icosahedrons (n=80)

31. Flat triangles, error = 3.29%

32. Double Icosahedrons (n=80) One subdivision, error = 0.946%

33. Double Icosahedrons (n=80) Two-fold subdivision, error = 0.290%

34. Double Icosahedrons (n=80) Three-fold subdivision, error = 0.121%

35. Double Icosahedrons (n=80) Four-fold subdivision, error = 0.079%

36. Double Icosahedrons (n=80) Five-fold subdivision, error = 0.068%

37. Double Icosahedrons (n=80) Six-fold subdivision, error = 0.065%

38. Curving the triangle patches using surface normal reduces error #-fold subdivisions Error on units sphere

39. Curved Triangles 1001,00010,000100,0001,000,00010 Number of Triangles Error on unit sphere STL AMF curved AMF curved 2-fold AMF curved 3-fold AMF curved 4-fold AMF curved 5-fold AMF curved 6-fold x4 error reduction with every subdivision

40. Curved Triangles 1001,00010,000100,0001,000,00010 Number of Triangles Error on unit sphere STL AMF curved AMF curved 2-fold AMF curved 3-fold AMF curved 4-fold AMF curved 5-fold AMF curved 6-fold Three orders of magnitude improvement in accuracy for same number of triangles

41. Curved Triangles 1001,00010,000100,0001,000,00010 Number of Triangles Error on unit sphere STL AMF curved AMF curved 2-fold AMF curved 3-fold AMF curved 4-fold AMF curved 5-fold AMF curved 6-fold Three orders of magnitude reduction in number of triangles for same accuracy

42. Accuracy

43. Fabricate 10cm diameter sphere with 10μm Precision – STL: 20,480 Flat Triangles 500K Compressed Binary STL – AMF: 320 Curved Triangles 10K Compressed AMF Fabricate 1m Sphere with 1nm precision – AMF: 1M Triangles – STL: !? Examples

44. Simple to implement 1.If tangents t 0 or t 1 not specified, compute tangents from normals 2.Compute center point v 01 = h( 0.5 ) and center tangent t 01 using Hermite curve h(s)=(2s 3 -3s 2 +1)v 0 +(s 3 -2s 2 +s)t 0 +(-2s 3 +3s 2 )v 1 +(s 3 -s 2 )t 1 3.Repeat for three triangle edges, then split triangle into four 4.Recurse as much as possible (diminishing returns after ~4 levels) 5. No ambiguities. Detailed procedure in specification. v0v0 v1v1 d n0n0 t0t0 t1t1 n1n1 n 01 v 01 t 01

45. Multiple Materials StiffMaterial FlexibleMaterial MediumMaterial 0.4 0.6 VerticallyGraded z 10-z Checkerboard floor(x+y+z%1)+0.5) 1-floor(x+y+z%1)+0.5).........

46. Graded Materials StiffMaterial FlexibleMaterial MediumMaterial 0.4 0.6 VerticallyGraded z 10-z Checkerboard floor(x+y+z%1)+0.5) 1-floor(x+y+z%1)+0.5).........

47. Microstructure StiffMaterial FlexibleMaterial MediumMaterial 0.4 0.6 VerticallyGraded z 10-z Checkerboard floor(x+y+z%1)+0.5) 1-floor(x+y+z%1)+0.5)......... Can also reference a texture map

48. Periodic functions can be used to describe linear and nonlinear lattice materials

49. Material properties By name – ABS – Nylon 1234 By physical property – 2GPa

50. Color and Graphics Can be assigned to – A material – A region – A vertex Specified – Fixed RGBA values – By formula – By reference to an image

51. Print Constellation Print orientation Duplicated objects Sets of different objects Efficient packing Hierarchical

52. Metadata John Doe”> SolidX 2.3”> Product 1> 12A”> Part A >

53. Not addressed (for now) Tolerances Surface/depth textures Data encryption, copyright External references and subassemblies Process control Non-volumetric support structures Non mesh geometry specification methods – Voxel, FRep

54. Next steps

56. Join The Discussion http://groups.google.com/group/stl2

57. http://ccsl.mae.cornell.edu/amf

58. Building consensus Contact hod.lipson@cornell.edu to be added or removedhod.lipson@cornell.edu

59. Required Tools Viewer – Geometry (with curved triangles) – Metadata – Constellations – Color – Materials STL   AMF Converter (available) Verifier

60. Certification? Canonical files – Computational verification – Physical verification XML Schema

61. Anticipated adoption stages CAD companies initially only implement AMF geometry export Full-fledged viewers are developed Equipment manufacturers have incentive to implement imports that support their unique capabilities – Companies supporting color implement color – Companies supporting multimaterial implement material parsing (volume, graded) – Companies interested in lattice structures implement lattice materials Small/new CAD programs allow designers to create parts with microstructure, color, graded materials etc. Big CAD follows Other applications emerge: 3D scanning, CNC

62. Conclusion AMF Standard converging – Voice your support or suggest alternatives – Ballot Sep 1 2010 Essential to moving the AM field forward – STL is holding back our field – Machines can do much more than current CAD/STL can support