National Technical University of Athens (NTUA)
Multiscale analysis of nano-composites using surrogate models
Abstract
dc:descriptionThe thesis at hand presents a neural network (NN)-based surrogate modeling approach suitable for the geometrically nonlinear analysis of carbon nanotubes (CNTs). In this work an NN-based Equivalent Beam Element (NN-EBE) is proposed, which is capable of accurately predicting the high-order phenomena caused by size-effects that characterize the behavior of CNTs at the nano-scale and can only be predicted by micro-mechanical models. The basic idea is to approximate the residual forces of the Newton–Raphson incremental-iterative formulation of the classical Euler or Timoshenko beams of the EBE model by an NN prediction, which is based on the response of the detailed MSM model of a CNT portion. Several numerical examples are presented for straight and wavy CNTs under bending and compression, which demonstrate that the proposed methodology is capable of efficiently predicting the nonlinear response of large-scale CNT structures in a fraction computing time compared to the full-scale problem.Moreover, an efficient multiscale analysis method for extracting the nonlinear constitutive law of carbon nanotube-reinforced composites (CNT-RCs) is proposed. This is accomplished via nonlinear computational homogenization on detailed FEM models of statistical volume elements (SVEs) of the composite material that can accurately describe its microstructure in a stochastic context. To this purpose, previous work of the authors on surrogate modeling of carbon nanotubes (CNTs) is extended to include CNT/matrix interface slippage phenomena and generalized to construct realistic RVEs of the composite that contain a large representative number of randomly dispersed,oriented and shaped CNTs. Wavy CNTs are modeled as a series of nonlinear EBEs with random initial imperfections. A statistical volume element (SVE) is developed, featuring the aforementioned modeling properties. The interface between the CNTs and the matrix is simulated using a cohesive zone formulation for the modeling of slippage through a nonlinear constitutive law. Computational homogenization is finally performed via stochastic averaging using a parallel implementation strategy in order to handle the associated extreme computational cost. The proposed analysis setup enables for realistic estimation of constitutive properties of advanced composites in acomputationally efficient manner, allowing for sensitivity analysis with respect to various modeling assumptions.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Soimoiris, Georgios
- Σοϊμοίρης, Γεώργιος
Subjects
dc:subject × 19- Ανάλυση πολλαπλών κλιμάκων
- Νανοσωλήνες άνθρακα
- Μη-γραμμική ομογενοποίηση
- Υποκατάστατη μοντελοποίηση
- Νευρωνικά δίκτυα
- Στοχαστική ανάλυση
- Multiscale analysis
- Carbon nanotubes
- Nonlinear computational homogenization
- Surrogate modeling
- Neural networks
- Stochastic analysis
- Επιστήμη Πολιτικού Μηχανικού
- Επιστήμες Μηχανικού και Τεχνολογία
- Δομοστατική
- Πολιτική μηχανική
- Civil Engineering
- Engineering and Technology
- Structural Engineering
Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/49034