Data-driven prediction and inverse design of fabric tensile properties via machine learning

Accurate prediction of fabric mechanical behavior constitutes a fundamental challenge in textile material design and optimization. As the primary mechanical property, tensile failure determines fabric durability and fastness, making its reliable prediction essential for quality assessment. To address the high cost and low efficiency issues arose from empirical fabric design, this study aims to establish a hybrid finite modeling - machine learning (FEM-ML) framework for predicting and inversely designing woven fabric tensile properties based on structural parameters. The FEM was firstly conducted to generate simulation datasets for ML training; then DNN ML models, were adopted to predict the tensile properties; An inverse design model was further established to directly determine structural parameters from the targeted tensile properties. Experimental results demonstrate that the optimal model achieved a coefficient of determination ( R ²) of 0.913 and a root mean square error (RMSE) of 0.032, with mean prediction errors 8.3 %. In addition, the inverse design model designed six sets of structural parameters, which were practically fabricated and tested. The average deviation between the measured and target tensile values remained 7.8 %, confirming the model’s predictive accuracy and practical applicability. Overall, the proposed FEM–ML hybrid approach enables both forward prediction and inverse optimization of fabric structures, offering a prospective data-driven strategy for intelligent textile design.