Deep learning for predicting stress and instantaneous modulus in closed-cell PVC foams

Abstract

Closed-cell polyvinyl chloride (PVC) foam offers several benefits, including its lightweight nature, high stiffness and specific strength, moisture resistance, and low thermal conductivity. These features make it an excellent core material for composite sandwich structures. This study aims to develop a deep neural network (DNN) model to accurately predict both stress and instantaneous modulus up to the densification phase based on density, strain, strain rate, and anisotropy ratio for all commercially available closed-cell H series PVC foams. To achieve this, the DNN model has been trained on stress-strain data from various studies published in the existing literature, encompassing different strain rates, densities, and anisotropy ratios of commercially available closed-cell H series PVC foams. The developed DNN model demonstrates exceptional accuracy in predicting stress and effectively captures the general trend of instantaneous modulus despite inherent noise in the actual data. This is evidenced by $R^2$ scores of 0.99 for stress and 0.77 for instantaneous modulus. These results underscore the model’s robustness, reliability, and potential as a powerful tool for accurately modeling the mechanical properties of closed-cell H series PVC foams.