Application of ANN to coupled hygro- Thermo-mechanical analysis of porous media with sorption hysteresis

Capillary condensation of water vapour plays an important role during hygrothermo- mechanical processes in meso-porous materials. An Artificial Neural Network is used to account for a set of laboratory test results concerning adsorption and desorption of water vapour. We include the relative humidity - saturation relationship by incorporating a suitably trained ANN into a FE program that models the coupled hygro-thermo-mechanical phenomena in deformable porous media. Computation of derivatives of the saturation inside the smooth segments of its graphs is also performed using the ANN, trained with the experimental results. The numerical performance of the developed hybrid ANN-FE computer program, called HMTRA-ANN, is tested on 1- and 2- dimensional examples dealing with a process of periodic drying - wetting of concrete elements. The experimental sorption data have been applied for training of ANN using the back propagation algorithm. The effect of different descriptions of the capillary condensation hysteresis on the hygro-thermal and mechanical behaviour of concrete is analysed. Essential differences of results concerning the water content distribution as well as the displacements and stresses are observed when different cases of hysteresis curves were compared to the solutions without hysteresis. We conclude that the type of the internal sorption curves has an effect on the results obtained.