PSI - Issue 71

Akash S.S. et al. / Procedia Structural Integrity 71 (2025) 180–187

181

1. Introduction Unidirectional (UD) composites represent a class of advanced materials engineered with fibers predominantly aligned in one direction, bound together by a matrix material. They exhibit exceptional strength and sti ff ness along the fiber direction, making them ideal for aerospace, automotive, marine, and sporting goods industries. These composites offer efficient load-bearing capabilities, lightweight nature, and customizable mechanical properties, resultingin optimized designs with superior strength-to-weight ratios. In the domain of manufacturing unidirectional (UD) composites, ensuring precise material properties stands as a critical imperative for optimizing performance and ensuring reliability in various applications. However, the inherent variability in fiber diameters presents challenges in ensuring precise material properties, traditionally addressed through labour-intensive quality testing and finite elementmodelling. Machine learning o ff ers a promising solution by streamlining quality testing, reducing time and resource expenditure, minimizing human error, and enhancing predictive capabilities through continuous learning from new data,thusimproving the reliability and performance of UD composites in various applications. Karnik et al. (2008) explored the analysis of delamination behaviour as a function of drilling process parameters at the entrance of the CFRP plates using artificial neural network (ANN) models. Fernandez et al. (2022) had proposed a new gradient-free training algorithm based on Approximate Bayesian Computation by Subset Simulation, where the likelihood function and the weights are defined by non-parametric formulations, resulting in a flexible and fairer representation of the uncertainty and had done a case study on composite materials subject to fatigue damage, showing the ability of the proposed algorithm to consistently reach accurate predictions while avoiding gradient related instabilities. Dey et al. (2016) had employed ANN as a surrogate for quantifying the uncertainty in the stochastic first two natural frequencies of laminated composite plates by training the model using Latin Hypercube sampling, proving to be more effective than the traditional Monte Carlo simulations. Dinesh Kumar et al. (2020) had provided an efficient method for uncertainty quantification and sensitivity analysis using adaptive Sparse Polynomial Chaos expansion (SPC) approach. Dzenan Hozic et al. (2023) had proposed a method to analyze effects of material uncertainty in composite laminate structures optimized using a simultaneous topology and material optimization approach by maximization of stiffness and minimization of a failure criteria index. Chahar and Mukhopadhyay (2023) had developed multi-fidelity ML based surrogates of progressive damage in composite laminates which can use a training dataset consisting of optimally distributed high- and low-fidelity simulations.SepahvandK.K. (2021) haddeveloped an artificial neural network (ANN) model to estimate the e ff ective elastic material properties of unidirectional composites using representative volume elements (RVE) considering uncertainty in the fiber diameter. In this study, Statistically representative Volume Elements (SVEs) are used for estimating the e ff ective material properties for di ff erent fiber diameter distributions. This data is used to train the ANN model for predicting the properties whenthe mean, standard deviation and number of fibers of the fiber diameter distribution are given as input parameters. Image processing techniques are used to extract the fiber diameter distributions from microstructural images. 2. Modelling of SVEs for different fiber diameter distributions Statistically representative volume element (SVE) [YinX. et al. (2008)] is a sub-volume of a material that is representative of the overall microstructure and is large enough to contain a statistically significant numberof features such as grains, inclusions, pores, or defects that are present in the material. SVE is used in micromechanics to study the properties of a material by analyzing a smaller, representative sample rather than the entire material. Itis an extension of RVE, or representative volume element, which is a sub-volume of a material that is representativeof the overall microstructure. The main advantage of SVE is that it accounts for the statistical variations in microstructure, which arepresent in many real-world materials. This allows for a more accurate representation of the material properties, as the SVE captures the range of variations in microstructure that are present in the material. In this study, it is assumed that the UD composite material consists of cylindrical infinite fibers embedded in an elastic matrix. Regular packing of fibers is assumed. The material properties of the fiber are assumed to be