Creep modeling of composite materials based on improved gene expression programming

In this article, a new method for creep modeling and performance prediction of composite materials is presented. Since Findley power-law model is usually suitable for studying one-dimensional time-dependent creep of materials under low stress, an intelligent computing method is utilized to derive three temperature-related sub-functions, the creep model as a function of time and temperature is established. In order to accelerate convergence rate and improve solution accuracy, an improved gene expression programming (IGEP) algorithm is proposed by adopting the probability-based population initialization and semi-elite roulette selection strategy. Based on short-term creep data at seven temperatures, a bivariate creep model with certain physical significance is developed. At fixed temperature, the univariate creep model is acquired. R-2, RMSE, MAE, RRSE statistical metrics are used to verify the validity of the developed model by comparison with viscoelastic models. Shift factor is solved by Arrhenius equation. The creep master curve is derived from time-temperature superposition model, and evaluated by Burgers, Findley and HKK models. R-square of IGEP model is above 0.98 that is better than classical models. Moreover, the model is utilized to predict creep values at t=1000 h. Compared with experimental values, the relative errors are within 5.2%. The results show that the improved algorithm can establish effective models that accurately predict the long-term creep performance of composites.

Automated summary

This article presents a new method for creep modeling and performance prediction of composite materials. An intelligent computing method is used to derive sub-functions related to temperature, and an improved gene expression programming algorithm is proposed. The validity of the developed model is verified through experiments, and the shift factor is solved using the Arrhenius equation. The creep master curve is derived and evaluated using different models, showing the effectiveness of the improved algorithm.