Deep learning framework for predictive modeling and optimization of mechanical properties in multiscale hybrid polymer nanocomposites

Polymer nanocomposites exhibit improved mechanical, thermal, and barrier properties due to the integration of nanoscale fillers, making them suitable for progressive engineering applications. However, predicting their complex behavior is challenging because of multiscale interactions and heterogeneous structures. To address this, a Uniform Physics Informed Neural Network (UPINN) based predictive modeling and optimization of mechanical characteristics in multiscale hybrid polymer nano composites (MHPNC-UPINN) is proposed. Initially, the input data is collected from experimental studies on Multiscale poly (3-hydroxybutyrate) (P3HB)-driven nanocomposites. Then, the data is fed to UPINN to predict the mechanical properties of the multiscale P3HB based nanocomposite. In General, UPINN does not incorporate optimization techniques to determine optimum parameters to assure accurate estimation. Hence, Secretary Bird Optimization Algorithm (SBOA) used to optimize the weight parameters of UPINN which accurately predict the mechanical properties. The proposed method is executed in the MATLAB platform and analyzed against existing techniques, including deep neural networks (DNNs), Fully Convolutional Network (FCN) and Artificial Neural Network (ANN). The proposed MHPNC-UPINN model achieved testing R 2 values of 0.955 for Young’s Modulus, 0.942 for Impact Strength, 0.925 for Elongation at Break, and 0.845 for Tensile Strength. It obtained the lowest prediction errors, with MSE values ranging from 0.028 to 0.034 and MAE values ranging from 0.095 to 0.109, demonstrating superior prediction accuracy and strong generalization capability compared with the existing models. The proposed MHPNC-UPINN framework effectively predicts the mechanical characteristics of multiscale P3HB-driven nanocomposites with high accuracy, demonstrating its potential for efficient material design and optimization.