Generation of microstructural images of titanium alloys as a function of heat treatment conditions using conditional generative adversarial networks
The present disclosure provides for the generation of microstructural images of components (e.g., titanium alloys) using machine learning frameworks. More particularly, the present disclosure provides for the generation of microstructural images of components (e.g., titanium alloys) as a function of heat treatment conditions using conditional generative adversarial networks. The present disclosure advantageously provides ways to accelerate component designs (e.g., titanium alloy designs) by developing generative models which can produce synthetic yet realistic microstructures conditioned on heat treatment conditions.
1 . A method for the generation of microstructural images of a component comprising:
providing a machine learning framework with user-specified heat treatment conditions of a component; and
utilizing the machine learning framework to generate microstructural images of the component as a function of the user-specified heat treatment conditions of the component,
wherein the user-specified heat treatment conditions of the component are represented in an embedded space, and are used as conditional inputs to a mapping network, along with random noise to generate a latent space distribution for a class of images in a training set, and latent space vectors are then transformed as inputs to a synthesis network, which generates images at different resolutions for different random noise inputs.
2 . The method of claim 1 , wherein the component comprises a titanium alloy.
3 . The method of claim 1 , wherein the machine learning framework comprises utilizing a conditional generative adversarial network.
4 . The method of claim 1 , wherein the machine learning framework generates the microstructural images of the component utilizing a range of user-specified heat treatment conditions of the component.
5 . The method of claim 1 , wherein the generated microstructural images of the component comprise synthetic yet realistic microstructure images conditioned on the heat treatment conditions of the component.
6 . The method of claim 1 , wherein the component is a test specimen or a full-scale component.
7 . The method of claim 1 , wherein the machine learning framework allows high fidelity knowledge capture and virtual prediction of microstructure in substantially any location of the component.
8 . The method of claim 1 further comprising utilizing the generated microstructural images to predict life or risk assessments of the component.
9 . The method of claim 1 further comprising utilizing the machine learning framework to generate variations of microstructural images of the component to facilitate a user to understand uncertainty and frequency of rare events of the component.
10 . The method of claim 1 , wherein the generation of each microstructural image takes less than one second.
11 . The method of claim 1 , wherein during a training phase, the machine learning framework learns to distinguish fake images from real ones, along with classifying the generated images into class labels seen in the training set.
12 . The method of claim 1 further comprising utilizing the generated microstructural images to infer statistical representations of microstructural properties of the component.
13 . The method of claim 1 , wherein the machine learning framework comprises utilizing a conditional generative modeling approach.
14 . The method of claim 1 , wherein the machine learning framework comprises utilizing a trained conditional generative model.
15 . The method of claim 1 , wherein the machine learning framework generates microstructural images of the component for new conditions outside a training set.