The objective of this research is is to develop a neural network trained from numerical simulations to model the stress diffusion in a granular soil based on the loading, soil fabric, stress field and stress history. Loadings to be studied include footings and laterally loaded piles. A photoelastic experimental study will be used to to verify results. Such a trained network would not only be able to reproduce the examples from which it was taught, but through its ability to generalize it, would also be able to find solutions to an entire set of problems that it has not previously encountered.
