A simple estimuon miethod of soil properties using neural network for prediction of ground vibration

In the prediction of ground vibration induced by artificial excitations, soil properties am essential data. Conventional exploration techniques and estimation methods of soil properties are not so practical in such a field as environmental vibration, because large-scale and/or costly investigations are not performed except in special cases. The authors propose a simplified method to estimate soil properties using a kind of back-analysis. In this method, ground is modeled to a visco-elastic half-space, and neural network is built to estimate soil properties from maximum displacements at three response points on the ground surface and time delays of maximum peaks in each displacement wave. Numerical simulation for layered ground indicates that the present method has applicability to practical use. In practical application of this method, however, excitation equipment plays an important role. In this paper, a falling weight impactor is tentatively made and neural network built for this impact force is used to estimate soil properties of actual ground.