Abstract

Nowadays, composite materials are often used in structural design for various engineering applications thanks to their good mechanical properties coupled to their lightness. Those materials are presently more and more used in renewable marine energy structures for the manufacturing of structural parts such as offshore windmill blades. During their life-service, these components are submitted to harsh environments which, combined to classic mechanical loadings, may lead to a premature aging of the structure. Among these various aggressive phenomena, water absorption is of first importance since the involving hygroscopic swelling may activate or worsen a damage mechanism such as crack initiation or propagation. In this work, we thus focus on the impact of water absorption on structural parts made of composites the polymer matrix of which can be hydrophilic. The diffusion model is the classical Fick’s law and the mechanical problem is solved under a linear elasticity assumption.

Experiments conducted in order to quantify the diffusion parameters (diffusion tensor and maximum moisture absorption capacity) showed a quite large dispersion was observed on the identified diffusion parameters. A stochastic study seemed necessary in order to well apprehend the uncertainties on the various output fields such as local water content or stress fields induced by the so-called hygroscopic swelling. Here is the purpose of the present work.

We adopt a parametric vision of the uncertainties which leads to a probabilistic model based on independent random variables. These random variables help in the modeling of parameters such as the water diffusion coefficient or the maximum moisture absorption capacity. We focus on the propagation of uncertainties through the proposed physical model governed by stochastic partial differential equations. Several methods are available to achieve this task depending on the probabilistic quantities one seeks to obtain (Monte-Carlo methods, reliability analysis etc.). Among them, spectral stochastic methods are good candidates in order to get an explicit solution with respect to the basic random variables modeling the diffusion coefficients, for instance. They consist in representing the random solution on a suitable approximation basis.

The efficiency of the proposed method is shown through numerical studies. We conduct different analyses at both microscale and macroscale for which we focus on various local and global stochastic quantities of interest. The impact of uncertainties associated with the input parameters on the outputs response is finally discussed.