Microstructural inelastic fingerprints and data-rich predictions of plasticity and damage in solids

Abstract

Inelastic mechanical responses in solids, such as plasticity, damage and crack initiation, are typically modeled in constitutive ways that display microstructural and loading dependence. Nevertheless, linear elasticity at infinitesimal deformations is used for microstructural properties. We demonstrate a framework that builds on sequences of microstructural images to develop fingerprints of inelastic tendencies, and then use them for data-rich predictions of mechanical responses up to failure. In analogy to common fingerprints, we show that these two-dimensional instability-precursor signatures may be used to reconstruct the full mechanical response of unknown sample microstructures; this feat is achieved by reconstructing appropriate average behaviors with the assistance of a deep convolutional neural network that is fine-tuned for image recognition. We demonstrate basic aspects of microstructural fingerprinting in a toy model of dislocation plasticity and then, we illustrate the method's scalability and robustness in phase field simulations of model binary alloys under mode-I fracture loading.

Prediction of creep failure time using machine learning

“…In recent times, machine learning found widespread applications in predicting deformation, failure, and flow processes in disordered systems based on complex data. Predictions of irreversible deformation and failure processes were based on data describing local atomic structure in amorphous solids 12,13 , mesoscale microstructures 14 (for an overview see e.g. 15 ), as well as monitoring data obtained in macrosopic tests [16][17][18] .…”

Materials Informatics for Mechanical Deformation: A Review of Applications and Challenges

“…Through the use of unsupervised ML approaches [ 58 , 60 ] it is possible to gain insights into the history of deformation of the crystalline sample, as well as consistently predict mechanical properties such as the yield stress, without the use of stress information. Furthermore, ML methods have been applied to determine the fracture toughness of composite materials based on DIC results, typically using artificial neural networks (ANNs) [ 45 , 61 , 62 ]. Crack detection, measurement, and characterization based on DIC can be performed using image processing methods [ 63 ] and fatigue crack detection in DIC images may be automatically performed using CNNs [ 64 ].…”

“…[ 11 , 12 ], CNNs were applied to classify microconstituents of ultrahigh carbon steels (UHCS). This is a characteristic application that can be extended in a large collection of material classes [ 45 ]. Figure 3 a presents the t-distributed stochastic neighborhood embedding (t-SNE) map for the UHCS data set and the constituents were classified based on SEM images.…”

“…It is also possible to consider, as in [ 81 ], the problem of defect detection as an anomaly detection one: in such a case, self-supervised ML could be only trained on defect-free experimental images. When there is no imaging noise, applying principal component analysis (PCA) can be shown to be enough to locate defects, but CNNs may improve the model performance in the presence of imaging noise that may also originate in quenched disorder [ 45 , 52 , 58 , 59 , 82 ]. Another way to circumvent the noise issues in TEM is by using scanning transmission electron microscopy (STEM) [ 80 ] where the size and quality of STEM datasets have been increasing exponentially.…”

“…Finally, it is worth noticing that MI methods may be used towards a complete classification of dislocation ensembles. While physical tools such as local misorientations, or topological features [ 45 ] may allow for ensemble classification, Ref. [ 172 ] investigated the possibility of defining ML descriptors for the classification of dislocation microstructures.…”