Simulated powdered model generation for neural networks

1 . A method comprising:

generating, by a processor, simulated powdered models that simulate application of powder to a part during additive manufacture of the part in accordance a computer aided design (CAD) model of the part;

generating, by the processor, synthetic images of the part using the simulated powdered models, where in the synthetic images the part is at least partially encapsulated by powder residue; and

training, by the processor, a neural network to recognize objects within images of the objects captured after post-additive manufacture transfer of the objects from a three-dimensional (3D) printing device,

wherein the objects in the images are at least partially encapsulated by the powder residue adhering to and/or accumulating on the objects during additive manufacture,

wherein the neural network is trained based on the synthetic images, and

wherein the trained neural network is subsequently used to recognize an actual object after additive manufacture that is at least partially encapsulated by the powder residue, within a captured image of the actual object.

2 . The method of claim 1 , wherein generating the simulated powdered models comprises extending vertices of the CAD model to simulate the powder adhering to the part during additive manufacture.

3 . The method of claim 1 , wherein generating the set of simulated powdered models simulate different amounts of the powder applied to outer surfaces of the part.

4 . The method of claim 1 , wherein generating the simulated powdered models comprises simulating the powder that accumulates more on concave surfaces than on convex surfaces of the part during additive manufacture.

5 . The method of claim 1 , further comprising:

additively manufacturing the actual object, using the 3D printing device; and

upon transfer of the actual object after additive manufacture, using the trained neural network to recognize the actual object within the captured image.

6 . The method of claim 5 , further comprising:

upon recognition of the actual object after additive manufacture, physically interacting with the actual object, by an actuator device.

7 . The method of claim 6 , wherein the actual object is transferred to a container after additive manufacture,

and wherein the actuator device is to physically interact with the actual object to remove the actual object from the container.

8 . A system comprising:

a processor; and

a memory storing instructions executable by the processor to:

generate a simulated powdered models that simulate application of powder to a part during additive manufacture of the part in accordance with a computer aided design (CAD) model of the part;

generate synthetic images of the part using the simulated powdered models, where in the synthetic images the part is at least partially encapsulated by powder residue;

generate annotations identifying the part in the synthetic images; and

train a neural network to recognize real objects within images of the objects captured after post-additive manufacture transfer of the objects from a three-dimensional (3D) printing device,

wherein the objects in the images are at least partially encapsulated by the powder residue adhering to and/or accumulating on the objects during additive manufacture,

wherein the neural network is trained based on the synthetic images and the annotations, and

wherein the trained neural network is subsequently used to recognize an actual object after additive manufacture that is at least partially encapsulated by the powder residue, within a captured image of the actual object.

9 . The system of claim 8 , wherein the instructions are executable by the processor to generate the synthetic images by:

simulating transfer of the part after post-additive manufacture transfer of the part from 3D printing device, using the simulated powdered models;

rendering images of the part after post-additive manufacture transfer, in accordance with simulation of the transfer; and

stitching backgrounds to the rendered images to generate the synthetic images.

10 . The system of claim 8 , wherein the instructions are executable by the processor to generate the annotations by:

changing the part in a given synthetic image to a first color;

changing a remainder of the given synthetic image to a second color;

determining a contour of the part; and

associating the contour and a label for the part with the given synthetic image.

11 . The system of claim 8 , wherein the instructions are executable by the processor to further determine locations of the part in the synthetic images,

and wherein the instructions are executable by the processor to generate the annotations based on the locations.

12 . The system of claim 8 , further comprising:

a vision device to capture the images of the real objects after post-additive manufacture transfer of the objects from the 3D printing device and to use the neural network to recognize the objects in the images.

13 . The system of claim 12 , further comprising:

an actuator device to physically interact with the objects after post-additive manufacture transfer of the objects from the 3D printing device, based on recognition of the objects within the images using the trained neural network.

14 . The system of claim 13 , wherein the objects are transferred to a container after additive manufacture,

and wherein the actuator device is to physically interact with the objects to remove the objects from the container.

15 . A non-transitory computer-readable medium storing instructions that when executed by a processor of a computing device, cause the processor to:

generate simulated powdered models that simulate application of powder to parts during additive manufacture of the parts in accordance with computer aided design (CAD) models of the parts;

generate synthetic images of the parts with and without the powder applied thereto, using the simulated powdered models and the CAD models, where in the synthetic images of the parts with the powder, the parts are at least partially encapsulated by powder residue;

generate annotations identifying the parts in the synthetic images; and

train a neural network to recognize objects within images of the objects captured after post-additive manufacture transfer of the objects from a three-dimensional (3D) printing device,

wherein the objects in the images are at least partially encapsulated by powder residue adhering to and/or accumulating on the objects during additive manufacture,

wherein the neural network is trained based on the synthetic images and the annotations, and

wherein the trained neural network is subsequently used to recognize an actual object after additive manufacture that is at least partially encapsulated by the powder residue, within a captured image of the actual object.

16 . The non-transitory computer-readable medium of claim 12 , wherein the annotations for the synthetic images are automatically generated based on locations of the parts in the synthetic images.

17 . The non-transitory computer-readable medium of claim 15 , wherein the instructions, when executed by the processor, further cause the processor to:

cause the 3D printing device to additively manufacture the objects.

18 . The non-transitory computer-readable medium of claim 17 , wherein the instructions, when executed by the processor, further cause the processor to:

upon transfer of the actual object after additive manufacture, recognize the actual object within the captured image, using the trained neural network.

19 . The non-transitory computer-readable medium of claim 18 , wherein the instructions, when executed by the processor, further cause the processor to:

upon recognition of the actual object after additive manufacture, cause an actuator device to physically interact with the actual object.