Mohsen Mirkhalaf

Accurate modeling of Short Fiber Reinforced Composites (SFRCs) remains computationally expensive for full-field simulations. Data-driven surrogate models using artificial neural networks have been proposed as an efficient alternative to numerical modeling, where Recurrent Neural Networks (RNNs) are increasingly being used for path-dependent multiscale modeling. However, recently, transformer models have been developed and they offer scalability and efficient parallelization, yet have not been systematically compared with RNNs in this field. In this study, we perform a systematic comparison between RNNs and transformer models trained on sequences of homogenized response of SFRC. We study the effect on two types of hyperparameters, namely architectural hyperparameters (such as the number of GRU layers, hidden size, number of attention heads, and encoder blocks) and training hyperparameters (such as learning rate and batch size). Both sets of hyperparameters are tuned using Bayesian optimization. We then analyze scaling laws with respect to dataset size and inference accuracy in interpolation and extrapolation regimes. The results show that while transformer models remain competitive in terms of accuracy on large datasets, the RNNs demonstrate better accuracy on small datasets. Furthermore, under extrapolation, the RNN is more accurate, while the transformer model is seven times faster at inference.

Biodegradable and renewable biocomposites have gained interest as solutions to reduce the environmental impact of composites. In this work, pulp fibres/thermoplastic starch biocomposites were fabricated with a single-step water-assisted reactive extrusion and characterised by thermomechanical analysis. This specific manufacturing process led to a reinforcement mechanism that, through both the upper-bonding theory and traditional simulation methods, cannot be properly captured. We investigated the relevance of the interface in such phenomena through micromechanical simulations performed via full-field representative elementary volume finite elements. The deviation between the experimental and simulated results led to a deepening of the investigation of the reinforcement mechanism at the matrix/fibres interface, where the modelling hypotheses failed to describe the system. This work pioneers a joint effort between modelling and experimentation in the overarching need for theoretical descriptions of outstanding reinforced short fibre polymer composites when the interactions between polymer matrix and reinforcement exceed the ‘perfect’ adhesion of the classical micromechanics. Graphical Abstract

Development of lightweight materials with enhanced mechanical properties has been a long-standing challenge in science and engineering. Lightweight auxetic metastructures (AMSs) provide attractive solutions to this problem. AMSs' negative Poisson's ratio is a unique characteristic which results in very interesting practical properties such as high energy absorption and improved toughness. Different properties of metastrcutures, including anisotropy, are dependent, in addition to their original material, on the unit cell shape and geometrical features which have very high variations. Over the past few years, researchers have developed AMSs with various unit cells, either introducing new internal architecture or enhancing and optimizing the existing ones. Despite the progress made in this field, there is no comprehensive review of the AMS unit cells. This review describes the cellular AMSs associated with a classified set of metastructures, comprising more than 100 distinct auxetic repetitive unit cells. Besides, achievable range of negative Poisson's ratios are provided for different categories of AMSs. Finally, the future perspective of the research field and potential developments and applications in this field are discussed.