scholarly journals NP-ODE: Neural process aided ordinary differential equations for uncertainty quantification of finite element analysis

2021 ◽  
pp. 1-16
Author(s):  
Yinan Wang ◽  
Kaiwen Wang ◽  
Wenjun Cai ◽  
Xiaowei Yue
Author(s):  
S. N. Medyanik ◽  
N. Vlahopoulos

The Energy Finite Element Analysis (EFEA) has been developed for modeling coupled structural-acoustic systems at mid-to-high frequencies when conventional finite element methods are no longer computationally efficient because they require very fine meshes. In standard Finite Element Analysis (FEA) approach, governing differential equations are formulated in terms of displacements which vary harmonically with space. This requires larger numbers of elements at higher frequencies when wavelengths become smaller. In the EFEA, governing differential equations are formulated in terms of energy density that is spatially averaged over a wavelength and time averaged over a period. The resulting solutions vary exponentially with space which makes them smooth and allows for using much coarser meshes. However, current EFEA formulations require exact matching between the meshes at the boundaries between structural and acoustic domains. This creates practical inconveniences in applying the method as well as limits its use to only fully compatible meshes. In this paper, a new formulation is presented that allows for using incompatible meshes in EFEA modeling, when shapes and/or sizes of elements at structural-acoustic interfaces do not match. In the main EFEA procedure, joints formulations between structural and acoustic domains have been changed in order to deal with non-matching elements. In addition, the new Pre-EFEA procedure which allows for automatic searching and formation of the new types of joints is developed for models with incompatible meshes. The new method is tested using a spherical shaped structural-acoustic interface. Results for incompatible meshes are validated by comparing to solutions obtained using regular compatible meshes. The effects of mesh incompatibility on the accuracy of results are discussed.


2020 ◽  
Vol 57 (10) ◽  
pp. 1566-1580
Author(s):  
Bipin K. Gupta ◽  
Dipanjan Basu

A nonlinear analysis framework for laterally loaded piles is presented that is as accurate as equivalent three-dimensional nonlinear finite element analysis, but computationally one order of magnitude faster. The nonlinear behavior of sands and clays are account for by using hyperbolic modulus–reduction relationships. These nonlinear–elastic constitutive models are used to calculate the reduced modulus at different points in the soil based on the soil strains induced by lateral pile displacement. The reduced modulus at different points in the soil domain are spatially integrated to calculate the reduced soil resistance parameters associated with the differential equation governing the lateral pile displacement. The differential equations governing the lateral displacements of pile and soil under equilibrium are obtained by applying the principle of virtual work to a continuum-based pile–soil system. These coupled differential equations are solved using the one-dimensional finite difference method following an iterative algorithm. The accuracy of the analysis is verified against equivalent three-dimensional nonlinear finite element analysis, and the validity of the analysis in predicting the field response is checked by comparisons with multiple pile load test results. Parametric studies are performed to gain insights into the lateral pile response.


Sign in / Sign up

Export Citation Format

Share Document