Modelling Fiber Network Materials:Micromechanics, Constitutive Behaviour and AI

Abstract: This thesis focuses on understanding the mechanical behavior of fiber-based materials by utilizing various modeling approaches. Particular emphasis is placed on their structural variability, anisotropic properties, and damage behavior. Furthermore, the study explores moisture diffusion phenomena within these materials, leveraging machine learning techniques. The research employs a blend of multiscale modeling, experimental investigation, machine learning, and continuum modeling to enhance the predictive capabilities for modelling fiber-based materials.In Paper I, the work investigates the impact of stochastic variations in the structural properties of thin fiber networks on their mechanical performance. A multiscale approach that includes modeling, numerical simulation, and experimental measurements is proposed to assess this relationship. The research also considers the influence of drying conditions during production on fiber properties. The study finds that spatial variability in density has a significant impact on local strain fields, while fiber orientation angle with respect to drying restraints is a key influencer of the mechanical response. In Paper II, the research delves into the investigation of anisotropic properties and pressure sensitivity of fiber network materials. It draws a comparison between the Hoffman yield criterion and the Xia model, which are widely utilized for simulating the mechanical response in fiber-based materials. The study performs a detailed analysis of these models under bi-axial loading conditions, assessing their numerical stability and calibration flexibility. Further supporting the research community, the paper provides open-source access to the user material implementations of both models and introduces a calibration tool specifically for the Xia model, thereby promoting ease of usage and facilitating further research in this domain. In Paper III a novel thermodynamically consistent continuum damage model for fiber-based materials is introduced. Through the integration of elastoplasticity and damage mechanisms, the model employs non-quadratic surfaces comprised of multi sub-surfaces, augmented with an enhanced gradient damage approach. The model’s capability is demonstrated by predicting the nonlinear mechanical behavior under in-plane loading. This study provides valuable insights into the damage behavior of fiber-based materials, showcasing a range of failure modes from brittle-like to ductile. In Paper IV, the study examines moisture penetration in fiber-based materials and the resultant out-of-plane deformation, known as curl deformation, using a combination of traditional experiments, machine learning techniques, and continuum modeling. The paper compares the effectiveness of two machine learning models, a Feedforward Neural Network (FNN) and a Recurrent Neural Network (RNN), in predicting the gradient of the moisture profile history. The study finds that the RNN model, which accounts for temporal dependencies, provides superior accuracy. The predicted gradient moisture profile enables simulating the curl response, offering a deeper understanding of the relationship between moisture penetration and paper curling.