Neural network thermal behavior predictions

1. A method for thermal behavior prediction by an electronic device, comprising:

computing, using at least one convolutional neural network, a predicted heat map of a layer corresponding to a three-dimensional (3D) model, wherein said computing is based on a contone map corresponding to the 3D model, wherein the contone map comprises a set of values indicative of agent placement;

partitioning a build area into a set of tiles;

determining one or more thermal stripes overlapping between at least two tiles in the set of tiles, wherein computing the predicted heat map is based on the one or more thermal stripes; and

printing the layer based on the predicted heat map.

2. The method of claim 1 , further comprising obtaining a thermal image of a build area, and wherein said computing the predicted heat map is further based on the thermal image.

3. The method of claim 2 , wherein said computing the predicted heat map comprises:

computing, using a first neural network, a first predicted heat map based on the contone map;

computing, using a second neural network, a second predicted heat map based on the thermal image; and

computing, using a third neural network, the predicted heat map based on the first predicted heat map and the second predicted heat map.

4. The method of claim 3 , wherein the first neural network is a convolutional neural network that accounts for thermal sourcing and sinking, the second neural network is a recurrent neural network that accounts for 3D thermal diffusion, and the third neural network is a convolutional neural network that accounts for contributions from the first neural network and the second neural network.

5. The method of claim 4 , wherein the second neural network is a long short term memory (LSTM) recurrent neural network having a convolutional neural network coupled to each network layer of the second neural network.

6. The method of claim 2 , wherein the thermal image is a thermal image of the layer at a first time, and wherein computing the predicted heat map comprises computing the predicted heat map of the layer corresponding to a second time after the first time.

7. The method of claim 1 , wherein said computing the predicted heat map is independent of thermal sensing after the at least one convolutional neural network is trained.

8. The method of claim 1 , further comprising:

obtaining a thermal image of a build area;

computing a loss function based on the predicted heat map and the thermal image; and

training the at least one convolutional neural network based on the loss function.

9. The method of claim 1 , wherein the contone map and a thermal image are inputs to the at least one convolutional neural network to compute the predicted heat map.

10. A three-dimensional (3D) printing system, comprising:

a build area;

a print head configured to print a fusing agent based on a fusing contone map and to print a detailing agent based on a detailing contone map, wherein the fusing contone map comprises a first set of values indicative of fusing agent placement and the detailing contone map comprises a second set of values indicative of detailing agent placement;

a thermal projector; and

a controller, wherein the controller is configured to:

use at least one convolutional neural network to predict thermal behavior of a layer based on the fusing contone map and the detailing contone map.

11. The 3D printing system of claim 10 , wherein the controller is further configured to control the print head based on the predicted thermal behavior.

12. A non-transitory tangible computer-readable medium storing executable code, comprising:

code for causing a processor to predict, using at least one convolutional neural network, thermal behavior of a layer corresponding to a 3D model based on voxel-level machine instructions, wherein the voxel-level machine instructions and a thermal image are inputs to the at least one convolutional neural network to predict the thermal behavior; and

code for causing the processor to print the layer or perform offline print tuning based on the predicted heat map.

13. The computer-readable medium of claim 12 , further comprising code for causing the processor to train the at least one convolutional neural network based on the predicted thermal behavior and a thermal image of the layer.