Estimating fibres’ material parameter distributions from limited data with the help of Bayesian inference

Abstract

Numerous materials are essentially structures of discrete fibres, yarns or struts. Considering these materials at their discrete scale, one may distinguish two types of intrinsic randomness that affect the structural behaviours of these discrete structures: geometrical randomness and material randomness. Identifying the material randomness is an experimentally demanding task, because many small fibres, yarns or struts need to be tested, which are not easy to handle. To avoid the testing of hundreds of constituents, this contribution proposes an identification approach that only requires a few dozen of constituents to be tested (we use twenty to be exact). The identification approach is applied to artificially generated measurements, so that the identified values can be compared to the true values. Another question this contribution aims to answer is how precise the material randomness needs to be identified, if the geometrical randomness will also influence the macroscale behaviour of these discrete networks. We therefore also study the effect of the identified material randomness to that of the actual material randomness for three types of structures; each with an increasing level of geometrical randomness.

Infusion Mechanisms in Brain White Matter and Their Dependence on Microstructure: An Experimental Study of Hydraulic Permeability

“…Such methodologies, which have been successfully used for the determination of e.g. materials linear and non-linear characteristics, viscoelasticity, anisotropy and multi-scale parameters in fibre composites [51]- [54], can also shed light on the choice of model and model uncertainty [55] when using different methods to study fluid flow in biological tissue, for example by using fractional order derivatives [55].…”

A Review of Recent Trends and Challenges in Computational Modeling of Paper and Paperboard at Different Scales

“…In consequence, this method needs to be adopted also in the modeling of paper and paperboard with the corresponding statistical distributions and effects. One approach going into this direction has been published recently in [209], where a scheme to identify material parameter distributions using only a limited number of fibres has been developed, which is based on probability density functions (PDF) and Bayesian inference.…”

Bayesian inference of non-linear multiscale model parameters accelerated by a Deep Neural Network

“…However, in the case of polymericbased composites, such dispersion can be important: in [28], PA06 tensile modulus measured at at constant temperature and strain rate ranges from 1200 to 3400 MPa, and tensile tests conducted on PA06-GF30 lead to a Young's modulus ranging from 6200 to 9500 MPa. To capture this dispersion, properties spatial distributions can be inferred [29], or BI can be adapted in order to infer the parameters of an assumed distribution of the material properties instead of the material properties themselves [30]. This so-called distribution-based BI approach requires a double MCMC sampling process in the non-linear range, which is the reason why in [16] the authors have inferred the distribution of the linear material constants only.…”