Modeling of heterogeneous materials at high strain rates with machine learning algorithms trained by finite element simulations

Abstract

Great progress has been made in the dynamic mechanical properties of concrete which is usually assumed to be homogenous. In fact, concrete is a typical heterogeneous material, and the meso-scale structure with aggregates has a significant effect on its macroscopic mechanical properties of concrete. In this paper, concrete is regarded as a two-phase composite material, that is, a combination of aggregate inclusion and mortar matrix. To create the finite element (FE) models, the Monte Carlo method is used to place the aggregates as random inclusions into the mortar matrix of the cylindrical specimens. To validate the numerical simulations of such an inclusion-matrix model at high strain rates, the comparisons with experimental results using the split Hopkinson pressure bar are made and good agreement is achieved in terms of dynamic increasing factor. By performing more extensive FE predictions, the influences of aggregate size and content on the macroscopic dynamic properties (i.e., peak dynamic strength) of concrete materials subjected to high strain rates are further investigated based on the back-propagation (BP) artificial neural network method. It is found that the particle size of aggregate has little effect on the dynamic mechanical properties of concrete but the peak dynamic strength of concrete increases obviously with the content increase of aggregate. After detailed comparisons with FE simulations, machine learning predictions based on the BP algorithm show good applicability for predicting dynamic mechanical strength of concrete with different aggregate sizes and contents. Instead of FE analysis with complicated meso-scale aggregate pre-processing, time-consuming simulation and laborious post-processing, machine learning predictions reproduce the stress–strain curves of concrete materials under high strain rates and thus the constitutive behavior can be efficiently predicted.

High‐Throughput Method–Accelerated Design of Ni‐Based Superalloys

“…So far, many algorithms have been employed to analyze plenty of data in materials science. [67][68][69][70] Among these algorithms, the frequency distribution statistics and unsupervised machine learning have been proved to exhibit outstanding accuracy and efficiency in narrowing the design ranges of alloys. In this work, the performances of these two algorithms were compared by selecting the composition ranges with superior microstructural stability and creep resistance, respectively.…”

Image Reconstruction for ECT under Compressed Sensing Framework Based on an Overcomplete Dictionary

“…According to the above analysis, three variations of the Landweber algorithm, the Tikhonov regularization algorithm, and the improved Newton-Raphson algorithm are used to achieve image reconstruction, which are called Non-Sparse-Landweber, Non-Sparse-Tikhonov, and Non-Sparse Newton-Raphson, and their derivation results are given below. NS-Landweber: (12) NS-Tikhonov:…”

“…where ( 12) and ( 14) are iterative algorithms and ( 12) is a direct algorithm. θ in (12), μ in ( 13), and γ in ( 14) are tunable parameters. e in ( 12) and ( 14) is the error between adjacent iterations.…”

“…There are two stages used by the K-SVD algorithm in training the overcomplete dictionary, which are sparse coding and dictionary updating. In the sparse coding stage, in order to obtain the sparse matrix, the sparse coefficient vector corresponding to each sample is calculated [38][39][40]. In the dictionary updating stage, atoms in the overcomplete dictionary are updated according to the non-zero elements of the sparse matrix.…”

“…Deep learning has shown great advantages in many fields, especially in image processing [9,10]. Deep learning has many advantages such as strong learning ability, wide coverage, and good portability [11][12][13]. However, deep learning models are more complex and time-consuming.…”