Effect of tensile testing parameters on the mechanical performance of Poulsenia Armata fiber cloth: Experimental analysis and machine learning modeling

This study investigates the influence of tensile testing parameters on the mechanical response of Poulsenia Armata fiber cloth (PAFC), a novel lignocellulosic material sourced from the Ecuadorian Amazon. A Taguchi L9 design of experiments was employed to systematically evaluate the effects of gauge length (25–75 mm) and crosshead speed (2.5–10 mm/min) on ultimate tensile strength (UTS), tensile modulus, and elongation at fracture. Microstructural analysis demonstrates a hierarchical fibrillar architecture with inherent defects, which govern damage initiation and promote progressive failure mechanisms such as microfibril debonding, pull-out, and fibrillation. The tensile behavior of PAFC is governed by a coupled interaction between defect-controlled size effects and strain-rate-dependent viscoelasticity. Increasing gauge length reduced UTS by up to 28% and elongation at fracture by up to 55%, while increasing crosshead speed enhanced tensile strength and modulus by up to 43% and 23%, respectively, due to the suppression of microfibril relaxation and improved load transfer efficiency. Statistical analysis confirmed that both parameters were significant (p < 0.05), with gauge length exerting the dominant influence. A suite of machine learning regressors, including Random Forest, Gaussian Process, K-Nearest Neighbors, and Multi-Layer Perceptron, was developed to predict tensile strength. After hyperparameter optimization, Random Forest and Gaussian Process Regression achieved the best predictive performance, reaching test-set R² values of 0.863 and maintaining robust behavior during cross-validation. The integration of experimental design, microstructural characterization, and machine learning establishes a robust framework for understanding, predicting, and optimizing the mechanical performance of natural fiber systems, highlighting the critical importance of testing parameter standardization for reliable material characterization.