geopsy.org

Hi,

I have undertaken MASW testing and ultimately want to replicate the soil conditions with high accuracy for an explicit finite element analysis. Despite this, I find I get quite different Dinver ground profiles based on my specified input parameters.

I have read on the forum that people recommend keeping density and poisson’s ratio constant for all layers, but ideally I would like something more realistic/accurate than this (FEM requires density, Poisson and Young’s modulus for each layer).

1. Are there any recommendations for maximising the accuracy of density/poisons ratio for each layer? I have deliberately been creating more layers than I believe to exist and allowing all 4 variables to vary at each layer (with only Vp linked to Vs)…is this likely to deliver more accurate soil properties (if I run enough model scenarios)? Also, should density be linked to Vs?

2. Is it possible to get a ground profile plot for poisons ratio (like that available for Vs, Vp and density)?

3. After obtaining material properties using Dinver, would forward modelling using FEM be the best approach to verify/tweak the layer material properties?

Many thanks,

David

I'm not sure I completely understood: you record real MASW signals, you extracted the dispersion with Geopsy or another software and you inverted them in dinver. In a second step, you would like to reproduce the signal time series with FEM. Is that correct?

1. If you do not have separate information about density, MASW inversion will not tell you much about the real values. Poisson's ratio in Dinver is not a parameter, it is a condition. It helps defining two profiles: a minimum and a maximum profiles between which Poisson's ratio must stand. If you have the same limits for all Poisson's ratio layers, this is strictly equivalent to a single layer. Having more than one layer for Poisson's ratio is reserved for very special cases or for research/development use. I should add a warning about that. Linking layer means only linking the bottom depths of linked layers (no constrain on parameter values). Be careful about the number of layers. I does not offer a more precise solution but explore more degrees of freedom. Usually we increase the number of layers while the minimum achievable misfit decreases significantly from one run to the next. It defines the minimum number of layers required to adjust the curve in an acceptable way. In a second step, we can add other layers (usually with constrains in depth) to explore the non-uniqueness,... but not to get more precise results.

2. You have to use gpdcreport: 3. this is a kind of full waveform inversion if I correctly understood. This is to be experimented. I've never explored such options.