Artificial neural network modeling for predicting mechanical properties of Mediterranean green composites
Abstract
The growing demand for sustainable materials has positioned lignocellulosic fiber-reinforced polypropylene composites as promising alternatives to conventional composites. However, accurately predicting key mechanical properties, including Young’s modulus (YM), ultimate tensile strength (UTS), and Elongation at Break (ELO), remains difficult because of the inherent variability in natural fiber composition. This study addresses this challenge by developing artificial neural network (ANN) models specifically tailored to Mediterranean lignocellulosic fibers derived from lemon and fig leaves. A computational dataset was generated from a limited set of mechanical, physical, and chemical-property observations reported in published studies on lemon- and fig-leaf-reinforced polypropylene composites. The literature-based property ranges were discretized into representative values and systematically combined to create ANN training observations. Controlled noise, outlier screening, and normalization were then applied to enhance data variability and suitability for model development. Several ANN architectures were evaluated, including narrow, medium, wide, bilayered, and trilayered networks. The bilayered ANN with a 10 × 10 architecture achieved the best performance for YM, with an R2 of 0.99 and RMSE values of 0.0378 for training and 0.0411 for testing. For UTS, the trilayered 10 × 10 × 10 ANN achieved an R2 of 0.99, with RMSE values of 0.0409 for training and 0.0414 for testing. The results demonstrate a clear trade-off between network complexity and predictive performance, while simpler architectures remained competitive for ELO prediction. Overall, the study highlights the potential of ANN models to optimize green composites, support industrial adoption, and advance the development of high-performance, eco-friendly materials. Future work will extend the models to additional natural fibers and composite formulations.