In this study, the diffusive-discrete crack transition scheme, originally developed for quasi-static brittle fracture, is enhanced to represent dynamic fracture within the finite strain framework. The developed approach simultaneously realizes the prediction of the diffusive crack propagation problem in the context of non-local damage theory and the diffusive-discrete crack transition utilizing the advantages of the finite cover method. Accordingly, a series of dynamic fracture events involving the crack initiation, propagation, bifurcation, divisions of an original object into multiple portions, and independent motions of divided portions can be continuously simulated. After presenting the formulation of the employed non-local damage model, as well as its spatial and temporal discretizations using the finite cover method and the Newmark method are described, several representative numerical examples are presented to demonstrate the performance and capabilities of the developed approach.

This study discusses the numerical properties of finite element procedures for eigenvalue analysis of free vibration in elastic solid structures. Exact solutions for a three-dimensional rectangular domain with the slip boundary condition, in which stationary P and S wave modes can be obtained separately, were adopted for comparison with numerical solutions. The distributions and numerical errors of the eigenvalues obtained by different finite element schemes were characterized in detail. Numerical performance was evaluated by utilizing particular stationary P and S wave modes in the cylindrical domain.

This research focuses on Physics-Informed Neural Network (PINN), which is a neural network that approximates the solution function of the initial-boundary value problem of partial differential equations. In particular, numerical experiments are conducted to evaluate the effect of point set features on the computational accuracy of predictive models by PINN. As point sets, in addition to general pseudo-random number sequences, low-discrepancy sequences such as Halton sequences and Sobol sequences, which have been shown to be useful in quasi-Monte Carlo methods, are also used. Furthermore, this research proposes the use of a finite element mesh smoothed by centroidal Voronoi tessellation as a technique to make it easier to apply PINN to regions with arbitrary boundary shapes.

In direct tensile test of concrete, bending moment and strain gradient occurs due to heterogeneity, which degrades the estimation accuracy of tensile stress-strain curve. This paper creates new estimation method of tensile stress-strain curve under the condition of bending moment and strain gradient. This new method uses data assimilation method and can individually estimate stress-strain curves, which are different by place. In the new method, block universe model is assumed in data assimilation. The block universe model is characterized by four-dimensional space, which consists x-y-z space and time flow. The space and time direction in block universe counterparts to the update of displacement and elastoplastic matrix respectively in FEM analysis. In block universe model, simulation parameters are assimilated in space and time directional alternately like EM algorithm and ADDM. In the space directional assimilation, the sensitivity of elastoplastic matrix’s constant factor is used in sequential data assimilation method. Then, node displacement in specimen is assimilated by using gradient decent method. In the time directional assimilation, simulation parameters of stress-strain-curve are perturbated and the influence of the perturbation to inner variables is predicted. Then, elastoplastic matrix is assimilated. By using this new method, the simulation parameters assimilated to observation data and the stress-strain curves were individually estimated.

The extended subloading surface model is endowed with the capability for predicting the elastoplastic deformation behavior in engineering problems involving cyclic loadings. In the conventional plasticity theory, the interior of the yield surface is always considered as a pure elastic domain. In contrast to this, the extended subloading surface model is capable of predicting the evolution of plastic strains even under a stress state below the yield stress. In this model, the elastic domain bounded by the subloading surface shrinks during elastic unloading, and then expands towards the normal-yield surface during plastic loading. The standard loading/unloading criterion for conventional plasticity models is not applicable to the extended subloading surface model, since it is not able to capture the transition from elastic unloading to plastic reverse loading during the above-mentioned process. Hence, a suitable scheme for loading/unloading judgment is required for the extended subloading surface model. We thus developed a novel stress-update algorithm for the subloading surface model equipped with an improved loading/unloading criterion. Furthermore, a series of numerical examples with regard to cyclic loadings along proportional and non-proportional strain paths were performed to demonstrate that the proposed algorithm enables highly accurate stress calculation even in arbitrarily large strain increments.

This study proposes a preprocessing method to address poor data integrity and excessive pipe parts information in industrial-scale CAD used for automatic model generation. Pipe connection relationships are first classified to strategically implement reconstruction steps based on a geometric approach. Critical pipe parts for simulation are then identified to reduce the final model size. Validation using actual plant CAD data revealed that the proposed method configured pipe connections although 25.5% of the data had integrity issues. The method also generated a model specification with 66% node reduction compared to unprocessed CAD. Furthermore, the thermohydraulic model built upon the simplified model specification achieved ninefold faster computational speed without compromising simulation accuracy. Additional tests showed that the model simulated actual plant operation with a maximum error of only 1.2%, demonstrating the effectiveness of the proposed method in building accurate models that enable plant digital twins.

This paper presents the improvement of the accuracy of plural crack propagation prediction by machine learning considering physical quantities limited small dataset. A dataset is obtained from the results of crack propagation analyses using s-version FEM combined with an automatic mesh generation technique. The input parameters are coordinates of the four crack tips. The output values to be predicted are crack propagation vectors and a number of crack propagation cycles of 0.25mm. Crack propagation paths and rates were predicted within 0.07 percent accuracy. This is a reason why independent multilayer perceptrons are separately configured in order to avoid an influence between different physical phenomena. Moreover, we tried to reduce the error by applying the data augmentation technique as a regularization. We observed the distribution of activations in the hidden layer to validate the generalization performance. We show that it is possible to predict with high accuracy even on small dataset with appropriate input and output parameters and appropriate configuration of training dataset.

Parameter Identification of seismic performance is important for accurate estimation of seismic performance and health monitoring of structures. In this study, we proposed a parameter identification method using quality engineering and interpretable machine learning "SHAP". Interpretable AI provided insight into the impact of the parameters on the analysis results. Understanding the importance of parameters leads to narrowing the range of parameter and efficient parameter identification. In this study, the method was validated on displacement response data obtained from a full-scale shake table experiment on 3-story wooden house. After data assimilation, the analysis results were closer to the experimental results, and the good results after data assimilation indicated their effectiveness. This method is also expected to be useful to support trial and error for the review process of analysis model.

PINN is a method for training neural networks by incorporating the error for an initial boundary value problem of partial differential equations into the loss function, and many studies have been reported. In order to improve the accuracy of PINN, it is desirable to increase the size of the training data set and to use a distribution with low discrepancy sequences. However, it is difficult to divide the point set while maintaining the characteristics of the distribution in the case of parallel processing of large training data. Therefore, this paper focuses on non-overlapping domain decomposition methods, which are known as a parallel numerical method for the finite element method. Especially, in addition to the classical Dirichlet-Neumann, Neumann-Neumann, and Dirichlet-Dirichlet algorithms, an iterative DDM algorithm based on the conjugate gradient method is developed for PINN. In addition, this paper applies the proposed method to a two-dimensional magnetostatic field problem and demonstrates numerical examples.

This paper presents an example of the verification and validation (V&V) for non-linear finite element analysis of reinforced concrete beams with shear reinforcements subjected to four-point bending. The code verification, together with the calculation verification, is performed by comparing the numerical results obtained by changing the mesh size with a reference solution based on Euler-Bernoulli beam theory. The sensitivity analysis based on the analysis of variance with an orthogonal array is then performed to quantify the degree of influence of material parameters on the numerical results. The computational model is finally validated for its intended use by comparing the results of a Monte Carlo simulation that reflects the variation in material properties with the results obtained from a validation experiment.

The virtual element method (VEM) is an approximation method for partial differential equations that does not require explicit definition of the interpolation functions for the test and trial functions within the elements. This feature allows VEM to handle elements of arbitrary polygonal or polyhedral shapes, including non-convex ones, without special treatment. Consequently, VEM is attracting attention as a generalization of the finite element method. However, research on VEM has mainly focused on the theoretical aspects, and its application to practical structural analysis has not been sufficiently explored. In this study, we propose to exploit the key advantage of VEM ―its ability to handle arbitrary polygons― in models with hanging nodes, i.e., nodes located on the edges of elements. The analysis of models comprising elements of different sizes connected via hanging nodes indicated that a size ratio of approximately 5 can provide sufficiently accurate results for practical purposes.