The purpose of this study is to investigate the role that a deep learning approach could play in computational mechanics. In this paper, a convolutional neural network technique based on modified loss function is proposed as a surrogate of the finite element method (FEM). Several surrogate-based physics-informed neural networks (PINNs) are developed to solve representative boundary value problems based on elliptic partial differential equations (PDEs). According to the authors’ knowledge, the proposed method has been applied for the first time to solve boundary value problems with elliptic partial differential equations as the governing equations. The results of the proposed surrogate-based approach are in good agreement with those of the conventional FEM. It is found that modification of the loss function could improve the prediction accuracy of the neural network. It is demonstrated that to some extent, the deep learning approach could replace the conventional numerical method as a significant surrogate model.
This study aims at generation of a novel artificial bee colony algorithm using surrogate finite element method with neural network technique. In this paper, theory of surrogate finite element method with physics-informed neural networks (PINNs) are generated and applied to deal with the geometrically nonlinear optimization problem of size, shape and topology for single-layer domes. In the artificial bee colony algorithm, the feedforward neural network is used to surrogate finite element analyses. Three numerical examples of 10-bar truss, Lamella dome, and Kiewit dome are carried out to verify feasibility and accuracy of the proposed method. Results of the present study are in good agreement with ones from literature. It is indicated that optimization processes can be considerably accelerated using the modified algorithm. That is, using the neural network surrogate-based models could significantly increase computational efficiency of structural optimum design for single-layer domes.
With the rapid development of machine learning and data science, computer performance continues to improve. It has become possible to integrate finite element analyses and machine learning technology. In this study, a surrogate-based finite element method enhanced by a deep learning technique is proposed to predict the displacement and stress fields of prestressed concrete beams with openings. Physics-informed neural networks (PINNs) were used to conduct a finite element analysis for the prestressed concrete structures. The displacement and stress of all nodal points were extracted to train the surrogate-based model. Then, the surrogate-based model was used to replace the original finite element model to estimate the displacement and stress fields. The results from the trained neural networks are in good agreement with experimental data obtained in a laboratory. It is demonstrated that the accuracy and efficiency of the proposed PINNs are superior to conventional approaches.
23 March 2009 in Bioscience, Biotechnology, and Biochemistry
Cell death can proceed via at least two distinct pathways, apoptosis and oncosis. Apoptosis is an energy-dependent process characterized morphologically by cell shrinkage, whereas oncosis is defined as a prelethal pathway leading to cell death associated with cellular swelling, organelle swelling, and increased membrane permeability. In this study, we found that overexpression of chromatin modifying protein 6 (CHMP6) induced cell death by a series of experiments, including morphological observation, intracellular ATP determination, caspase-3 activity, and flow cytometry. Typical morphological characteristics consistent with oncosis were observed by transmission electron microscopy. Simultaneously, we obtained some results that indicated apoptosis, but the anti-apoptotic gene Bcl-xL and caspase family inhibitor Z-VAD-FMK had little effect on CHMP6-induced cell death. These results suggest that CHMP6 overexpression can cause cell death, predominantly via oncosis and to a certain extent via apoptosis, and th...
For concrete structures in offshore areas, chloride ion erosion is one of the main factors affecting durability. It is crucial to evaluate the chloride ion permeability resistance of concrete structures. In this paper, a finite element simulation of the chloride ion diffusion process in concrete is conducted. A mass diffusion finite element model based on a random aggregate approach is established to investigate the influences of an aggregate, the interface transition zone, and micro-cracks on the chloride ion diffusion coefficients in concrete. The results show that the mass diffusion finite element analysis based on the exponential function model and the power function model can effectively simulate the chloride ion diffusion process in concrete. In addition, the data reveal that volume fraction and distribution aggregates considerably affect chloride ion diffusivity in concrete. Also, the interface transition zone significantly accelerates chloride ion diffusion in concrete. Moreover, this acceleration effect exceeds the barrier effect of an aggregate.
This study aimed at exploring what role artificial intelligence techniques could play in the futural numerical analysis. In this paper, a convolutional neural network techniques based on modified loss function is proposed as a surrogate of finite element method(FEM). Several surrogate-based physics-informed neural networks(PINNs) are developed to solve representative boundary value problems based on elliptic partial differential equations (PDEs). Results from the proposed surrogate-based approach are in good agreement with ones from conventional FEM. It is found that modification of the loss function could improve the prediction accuracy of the neural network. It is indicated that to some extent the artificial intelligence technique could replace conventional numerical analysis as a great surrogate model.