Choose timezone
Your profile timezone:
CD Technical Meeting (ML4): 3D variational autoencoder for fingerprinting microstructure volume elements
→
Europe/London
Description
Machine Learning, Uncertainty Quantification and Data Science
Abstract
Microstructure quantification is an important step towards establishing structure-property relationships in materials. Machine learning-based image processing methods have been shown to outperform conventional image processing techniques and are increasingly applied to microstructure quantification tasks. In this work, we present a 3D variational autoencoder (VAE) for encoding microstructure volume elements (VEs) comprising voxelated crystallographic orientation data. Crystal symmetries in the orientation space are accounted for by mapping to the crystallographic fundamental zone as a preprocessing step, which allows for a continuous loss function to be used and improves the training convergence rate. The VAE is then used to encode a training set of VEs with an equiaxed polycrystalline microstructure with random texture. Accurate reconstructions are achieved with a relative average misorientation error of 9x10-3 on the test dataset, for a continuous latent space with dimension 256. We show that the model generalises well to microstructures with textures, grain sizes and aspect ratios outside the training distribution.
The main aim of this project is to present the 3D VAE as an application-agnostic method for parameterising microstructure for input into downstream tasks where microstructure dependence is required. As a proof of concept, a simple surrogate model for uniaxial crystal plasticity (CP) simulations, with a fixed load path and microstructural dependence is presented. Microstructural fingerprints, obtained by encoding VEs with the trained VAE encoder, parameterise the VEs in a low-dimensional latent space and are stored alongside the volume-averaged stress response, at each strain increment. This is then used to train a fully connected neural network mapping the input fingerprint to the resulting stress response, which acts as a surrogate model for the CP simulation. The fingerprint-based surrogate model is shown to accurately predict the microstructural dependence in the CP stress response, with a mean relative error of 2.75 MPa on unseen test data. This approach offers a significant speed-up on the order of 108 for a stress-strain curve prediction, compared to running a CP simulation.
