Simple heuristic for data-driven computational elasticity with material data involving noise and outliers: a local robust regression approach

Abstract

Data-driven computing in applied mechanics utilizes the material data set directly, and hence is free from errors and uncertainties stemming from the conventional material modeling. This paper presents a simple heuristic for data-driven computing, that is robust against noise and outliers in a data set. For each structural element, we extract the material property from some nearest data points. Using the nearest neighbors reduces the influence of noise, compared with the existing method that uses a single data point. Also, the robust regression is adopted to reduce the influence of outliers. Numerical experiments on the static equilibrium analysis of trusses are performed to illustrate that the proposed method is robust against the presence of noise and outliers and, hence, is effective for dealing with real-world data.

Data-driven reduced homogenization for transient diffusion problems with emergent history effects

“…To reduce the effect of noise in a data-set and obtain smoother fields, a regression can be performed on neighboring data-points [32]. Noisy data-sets with significant outliers may create a larger problem.…”

A neural network constitutive model for hyperelasticity based on molecular dynamics simulations

“…Extensive related studies were conducted. 24,25 Since the introduction of artificial neural networks (NNs) for constitutive modeling of concrete by Ghaboussi et al, [26][27][28] a number of studies on the use of NNs to model materials have also been conducted. [29][30][31][32][33] These models can be used to acquire information about the behavior of a material directly from experimental datasets, and no significant idealizations are used in defining the model.…”

“…This approach is usually known as “data‐driven” computing, and is free from the uncertainties and errors encountered when a material is modeled. Extensive related studies were conducted 24,25 . Since the introduction of artificial neural networks (NNs) for constitutive modeling of concrete by Ghaboussi et al, 26‐28 a number of studies on the use of NNs to model materials have also been conducted 29‐33 .…”

Relaxation analysis in a data driven problem with a single outlier

“…For recent developments of the data driven approach we refer to [10] and [17], which are both concerned with numerical aspects. Finite plasticity in the context of data driven analysis is treated in [7].…”