Neural networks for predicting the temperature-dependent viscoelastic response of PEEK under constant stress rate loading

High-performance polymer composites are used in demanding applications in civil and aerospace engineering. Often, structures made from such composites are monitored using structural health monitoring systems. This investigation aims to use a multilayer perceptron neural network to model polymer response to a non-standard excitation under different temperature conditions. Model could be implemented into health monitoring systems. Specifically, the neural network was used to model PEEK material's creep behavior under constant shear stress rate excitation at different temperatures. Optimal neural network topology, the effect of the amount of training data and its distribution in a temperature range on prediction quality were investigated. The results showed that based on the proposed optimization criterion, a properly trained neural network can predict polymeric material behavior within the experimental error. The neural network also enabled good prediction at temperatures where stress-strain behavior was not experimentally determined.

Automated summary

In this study, a multilayer perceptron neural network was used to model the response of high-performance polymer composites to non-standard excitation at different temperatures. The research found that a properly trained neural network can predict polymeric material behavior within the experimental error.