Neural network-based prediction of the long-term time-dependent mechanical behavior of laminated composite plates with arbitrary hygrothermal effects

Recurrent neural network (RNN)-based accelerated prediction was achieved for the long-term time-dependent behavior of viscoelastic composite laminated Mindlin plates subjected to arbitrary mechanical and hygrothermal loading. Time-integrated constitutive stress-strain relation was simplified via Laplace transform to a linear system to reduce the computational storage. A fast converging smooth finite element method named cell-based smoothed discrete shear gap was employed to enhance the data generation procedure for straining RNNs with a sparse mesh. This technique is applicable under varying hygrothermal conditions for real engineering structure problems with fluctuating temperature and moisture. Hence, accurate RNN-based long-term deformation prediction for laminated structures was realized using the history of environmental temperature and moisture condition.

Automated summary

This study utilized RNN for accelerated prediction of the long-term deformation behavior of viscoelastic composite materials, simplifying the time-integrated constitutive relation and employing a smooth finite element method to reduce computational storage cost. The technique can be applied in real engineering structures under varying hygrothermal conditions.