Abstract:

The conventional numerical simulation applied to the excavation of foundation pit may result in inaccurate excavation deformation because it does not fully integrate the case history with the observed performance data. The self-learning method for engineering simulation is a neural network-based inverse analysis technique that combines the finite element analysis and the artificial intelligence. The excavation simulation process includes three steps ： （1） obtaining the case observation data, （2） calculating the soil stress and measuring the soil deformation, and （3） performing the finite element analysis using the obtained data and predicting the deformation in the following excavation stage. This method possesses the self-adaptability and the self-organization, learning, association, fault tolerance and anti-interference abilities of neural network and can reveal the nonlinear correlation between the historical information and the field observed data. The application of the self-learning method to a simulated case and three historical cases indicates that the method is feasible in the excavation simulation of foundation pit.

Key words: foundation pit, excavation, self-learning method, engineering simulation, constitutive model, numeri-cal simulation