作者：Minhan Song,Wei Wang,Chong Sun
關鍵字：Machine learning,BP neural network,Equibiaxial tensile stress
論文來源：期刊
具體來源：https://doi.org/10.1016/j.polymer.2025.129030
發表時間：2025年
Equibiaxial tensile test data are essential for determining the parameters of rubber constitutive models. However, the equibiaxial tensile testing process is both expensive and time-consuming due to its complexity. To overcome these challenges, we propose a novel approach to predict the equibiaxial tensile stress. In this investigation, based on machine learning, we developed the BP-ANNs (Backpropagation artificial neural networks) model
trained with two algorithms: Levenberg-Marquardt algorithm and Bayesian Regularization algorithm. The developed models demonstrate a strong capability to map uniaxial tension and pure shear deformation data to equibiaxial stress fields. The results indicate that the BP neural network trained with the Bayesian Regularization algorithm can effectively predict the equibiaxial tensile stress of natural rubber-based compounds over a wide strain range. This method can predict the equibiaxial tensile stress of natural rubber with a prediction error of
less than 5 %. In contrast, although the BP neural network trained using the Levenberg-Marquardt algorithm demonstrates good performance during training, it yields relatively large prediction errors for equibiaxial tensile stress. Furthermore, we also considered scenarios where pure shear test data are unavailable. The Yeoh constitutive model, fitted using only uniaxial tensile data, was employed to calculate the pure shear stress, which was then used as an input set for the BP neural network to predict the equibiaxial tensile stress. The prediction error of this method is less than 10 %. In conclusion, these two methods provide a fast and reliable alternative for obtaining the equibiaxial tensile stress of rubber materials. This approach can significantly reduce testing costs, shorten testing cycles, and effectively improve the accuracy of fitting rubber constitutive models.