US 12,481,880 B2
Defect removal from manufactured objects having morphed surfaces
David Patrick Lovell, Moseley (GB); Daniela Sofia Seixas Sousa, Stafford (GB); and Chin-Yi Cheng, San Francisco, CA (US)
Assigned to Autodesk, Inc., San Francisco, CA (US)
Filed by Autodesk, Inc., San Francisco, CA (US)
Filed on Apr. 28, 2023, as Appl. No. 18/141,262.
Application 18/141,262 is a continuation of application No. 16/524,912, filed on Jul. 29, 2019, granted, now 11,676,007.
Prior Publication US 2023/0289596 A1, Sep. 14, 2023
Int. Cl. G06N 3/08 (2023.01); G05B 19/4097 (2006.01); G06F 30/20 (2020.01)
CPC G06N 3/08 (2013.01) [G05B 19/4097 (2013.01); G06F 30/20 (2020.01); G05B 2219/35134 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
obtaining, in a computer aided manufacturing program,
a first three dimensional model of at least one actual deformed three dimensional surface of a manufactured object, and
a second three dimensional model of at least one non-deformed source three dimensional surface used as input to a manufacturing process that generated the manufactured object, wherein the first three dimensional model is different from the second three dimensional model because of: (1) at least one surface defect, and (2) deformation of the manufactured object;
producing, by the computer aided manufacturing program, a two dimensional image by projection of points between the first three dimensional model and the second three dimensional model, wherein each of multiple pixel values in the two dimensional image is a difference along a projection from a point in the second three dimensional model to a corresponding point in the first three dimensional model;
providing, by the computer aided manufacturing program, the two dimensional image to an image-to-image translation based machine learning algorithm, wherein the image-to-image translation based machine learning algorithm was previously trained using pairs of input images, each of the pairs of input images being
a first image representing differences between a nominal three dimensional surface and a corresponding deformed version of the nominal three dimensional surface, and
a second image representing differences between the nominal three dimensional surface and a corresponding deformed version of the nominal three dimensional surface with one or more included surface defects
wherein the image-to-image translation based machine learning algorithm was previously trained to translate the second image into the first image, resulting in removal of the one or more included surface defects;
receiving, by the computer aided manufacturing program, a translated version of the two dimensional image from the image-to-image translation based machine learning algorithm; and
producing, by the computer aided manufacturing program, a third three dimensional model of at least one deformed three dimensional surface corresponding to the at least one non-deformed source three dimensional surface by projection of points between the first three dimensional model and the second three dimensional model using differences represented by pixel values in the translated version of the two dimensional image, wherein the third three dimensional model includes the deformation of the manufactured object and removes the at least one surface defect.