Deep learning approach for chemistry and processing history prediction from materials microstructure

Abstract

Finding the chemical composition and processing history from a microstructure morphology for heterogeneous materials is desired in many applications. While the simulation methods based on physical concepts such as the phase-field method can predict the spatio-temporal evolution of the materials’ microstructure, they are not efficient techniques for predicting processing and chemistry if a specific morphology is desired. In this study, we propose a framework based on a deep learning approach that enables us to predict the chemistry and processing history just by reading the morphological distribution of one element. As a case study, we used a dataset from spinodal decomposition simulation of Fe–Cr–Co alloy created by the phase-field method. The mixed dataset, which includes both images, i.e., the morphology of Fe distribution, and continuous data, i.e., the Fe minimum and maximum concentration in the microstructures, are used as input data, and the spinodal temperature and initial chemical composition are utilized as the output data to train the proposed deep neural network. The proposed convolutional layers were compared with pretrained EfficientNet convolutional layers as transfer learning in microstructure feature extraction. The results show that the trained shallow network is effective for chemistry prediction. However, accurate prediction of processing temperature requires more complex feature extraction from the morphology of the microstructure. We benchmarked the model predictive accuracy for real alloy systems with a Fe–Cr–Co transmission electron microscopy micrograph. The predicted chemistry and heat treatment temperature were in good agreement with the ground truth.

Combining simulation and experimental data via surrogate modelling of continuum dislocation dynamics simulations

“…However, it has to be remarked that the GBT does not include the material history in its prediction, which is contradicting plasticity theory on a first glance [55]. In literature, LSTMs are frequently employed for data series predictions since it has been found that they are able to take into account the influence of preceding microstructure evolution [27,28,56]. This is not explicitly found in our study since the GBT shows better results compared to the LSTM.…”

Designed assembly and disassembly of DNA in supramolecular structure: From ion regulated nuclear formation and machine learning recognition to running DNA cascade

“…[35] To get high prediction accuracy we also use transfer learning method. [36] For our dataset, we obtained 7000 fluorescent microscopy images (Figure S14) to confirm whether dependences between Mg 2+ concentration and the structures of assembled capsules and their fluorescence intensities existed. We formed the reaction-diffusion systems containing 0 mM, 10 mM, 50 mM, 200 mM, or 500 mM Mg 2+ .…”

Machine Learning in Soft Matter: From Simulations to Experiments

“…Meanwhile, in order to overcome the limitation of PCA as a 1D dimensionality reduction method, some studies have utilized tensor decomposition [135,136] instead of PCA or employed autoencoders as a nonlinear feature extraction method. [137][138][139][140] Moreover, more complex and effective models for the prediction of spatiotemporal sequences, such as ConvLSTM, [137][138][139] E3D-LSTM, [141] and PredRNN, [142] have also been applied to predict long-term microstructures with heightened efficiency and precision. In addition, for systems with overdamped kinetics, such as defect kinetics in lamellar morphology, some studies have successfully employed acceleration schemes without time dependence, like FFT+CNN, [143] which predict the microstructure at time t + Δt based on that at time t. Since this task is similar to traditional deep learning tasks, such as video prediction, [144] future research might develop more advanced prediction algorithms for enhancing the speed and accuracy of predictions.…”