As a result of our synthetic data experiments, we make the following generalizations:
Reasonable single layer estimates can be obtained using dispersion and attenuation from only the fundamental mode.
Varying half space by±50 % had no discernable effect on other parameter estimates.
For a single-layer model, assuming an incorrectfixed value of layer thick- ness has a modest effect on estimates of shear modulus and viscosity;
however, assuming an incorrect fixed value of shear modulus can effect a very large error in viscosity estimates.
Labile changes from primary to companion modes (and vice versa) could cause the inversion process to converge to the wrong mode, entirely. Avoiding this problem requires that the model be initialized using vis- cosities near the true values, and that a sequential inversion process be employed.
Attempts to estimate viscosity using dispersion data, alone, were only successful when inclusion of viscosity created an obvious shelf that was not present in the purely elastic model.
Attempts to estimate layer thickness and shear modulus using only atten- uation information were unsuccessful.
It is better to assume a model with too many layers than it is to assume a model with too few layers.
For two and three-layer models, the inversion procedure is sensitive to initial assumptions about layer thickness. For the two-layer model, con- straining total layer thickness or effective layer thickness improved esti- mates of shear modulus and viscosity. For the three-layer model, reason- able (non-negative) estimates of shear modulus and layer thickness could not be obtained without constraining either total layer thickness or effec- tive layer thickness.
CHAPTER 7:
CONCLUSIONS AND RECOMMENDATIONS
FOR FURTHER RESEARCH
We set out to determine whether or not Love wave inversion could be a viable means for determining viscoelastic soil properties. The answer is a qualifiedyes. We demon- strated a method for winnowing Love wave modes from the types of data typically collected in the field, and we developed a straightforward inversion process that can be used to identify and characterize soil layering. As we have seen, the inversion process is relatively stable, and almost always yields a solution. For single-layer models, the method gives reasonable estimates of layer thickness, shear modulus, and viscosity, even when dispersion and attenuation information is only available for a single mode. As the number of layers increases, accurate estimates of shear modulus and viscos- ity require that constraints on layer thickness be incorporated into the model. This suggests that surveys using Love waves should be used in conjunction with downhole or crosshole methods: With layer thicknesses from these latter methods as a control, Love wave seismic methods can be used to extend the area of investigation.
We saw that the process for determining dispersion information was relatively straightforward; however, obtaining amplitude information of sufficient quality to estimate attenuation coefficients requires that extra care be used in the design and conduct of field experiments. Given a sufficient level of care, we have shown how data from a source walk-away experiment can be used to extract attenuation information of sufficient quality to estimate viscous soil parameters.
Depending on local conditions and equipment availability, a receiver walk-away experiment may afford better reproducibility, and we have suggested a method for estimating dispersion curves and attenuation coefficient for receiver walk-away exper- iments.
Our analyses in Chapters 4 through 6 were conducted using layer thicknesses between 5 and 10 m, and useful data was obtained from modes between about 10 and 50Hz. As discussed in Chapter 1,Vs30 calculations specified by theInternational
Building Code require that viscoelastic properties be measured to a depth of at least 30m. Using the scaling rules derived in Chapter 4, the first four modes of a 30meter layer structure would make their appearances at frequencies less than about 10 Hz. Higher modes might be detectable above this frequency range, but as we saw at the end of Chapter 4, the behavior of higher modes can very difficult to predict. Receivers with a two to three Hertz lower limit should be considered for Vs30 surveys.
Mode ambiguity was an impediment to inverting the data. The potential for a mode to transition from a primary to a companion mode, or vice versa, was the usual culprit when the inversion process became bistable. Nevertheless, choosing initial values of layer thickness, shear modulus, and viscosity that ensured the proper mode type (primary vs. companion) prevented this issue.
The current inversion process is very labor intensive, and requires between sixteen and twenty hours per inversion. In particular, the process of identifying and tracking roots in order to create predictive dispersion curves is very time consuming, and it is only automated at frequencies far from each mode’s cut-off frequency. Attempts to fully automate the procedure in the cut-off frequency region were unsuccessful.
task of separating and measuring their amplitudes is both labor intensive and prone to error. Many attempts were made to filter the data; however, most of these schemes required a-priori knowledge of the system’s dispersive behavior.