Self-supervised learning and prediction of microstructure evolution with convolutional recurrent neural networks

Summary

Microstructural evolution is a key aspect of understanding and exploiting the processing-structure-property relationship of materials. Modeling microstructure evolution usually relies on coarse-grained simulations with evolution principles described by partial differential equations (PDEs). Here we demonstrate that convolutional recurrent neural networks can learn the underlying physical rules and replace PDE-based simulations in the prediction of microstructure phenomena. Neural nets are trained by self-supervised learning with image sequences from simulations of several common processes, including plane-wave propagation, grain growth, spinodal decomposition, and dendritic crystal growth. The trained networks can accurately predict both short-term local dynamics and long-term statistical properties of microstructures assessed herein and are capable of extrapolating beyond the training datasets in spatiotemporal domains and configurational and parametric spaces. Such a data-driven approach offers significant advantages over PDE-based simulations in time-stepping efficiency and offers a useful alternative, especially when the material parameters or governing PDEs are not well determined.

Highlights

• •

  A convolutional recurrent neural net is trained to predict microstructure evolution

• •

  The model accurately predicts short-term dynamics and long-term statistical properties

• •

  It is significantly more efficient than the traditional PDE-based modeling approaches

• •

  It can predict microstructure evolution based on incomplete information of systems

The bigger picture

Material microstructures are small structural features intermediate between atoms and macroscopic products with often decisive impact on the performance of engineering materials. A major goal of modern materials science is to improve material properties through the control of microstructure evolution during material processing and service. Microstructure evolution is traditionally simulated by continuum models based on partial differential equations. Here we demonstrate that convolutional recurrent neural networks, a type of machine-learning method, can be trained to predict various microstructure evolution phenomena with significantly improved efficiency. The method can learn the evolution rules from microstructure image sequences and make reliable predictions even with incomplete information about the systems or underlying mechanisms. This work illustrates the increasing power of data-driven approaches to address the computational challenges in microstructure modeling.

Abstract

Material microstructure plays a key role in the processing-structure-property relationship of engineering materials. Microstructure evolution is commonly simulated by computationally expensive continuum models. Yang et al. apply convolution recurrent neural networks to learn and predict several microstructure evolution phenomena of different complexities. The method is significantly faster than the traditional approach and capable of predicting the evolution process in systems with unknown material parameters. It provides a useful data-driven alternative to microstructure simulation.