Artificial neural network approach to tailor composite materials with nonlinear viscoelasticity

Material characterization is a major challenge for viscoelastic composite materials. Due to nonlinearity of materials properties, it is overwhelming to test each composite material for every application case under the combined effects of temperature and strain rate. Machine learning methods can help by using the existing datasets to predict properties over a different combination of parameters. This work focuses on building an artificial neural network (ANN) architecture to help in predicting properties and compositions of viscoelastic materials. The high density polyethylene (HDPE) syntactic foam is used as a case study material. Four types of HDPE syntactic foams were tested using dynamic mechanical analysis (DMA). Then, ANN was used to build the master relation of viscoelastic properties with respect to frequency, temperature, particle volume percentage and strain. The master curve for storage modulus was then transformed to time domain relaxation function and used to predict the stress-strain relations. The elastic modulus was extracted and compared to experimental results from tensile tests. The results show good agreements in properties of syntactic foams with both tested and extrapolated compositions. These results show that ANN can help in designing composite materials using machine learning methods on a limited dataset.