| dc.description.abstract | Soil liquefaction is a phenomenon caused by seismic activity in the
ground which may result in surface settlement, the formation of sand
boils, lateral spreading that ultimately damages the super-structure and
loss of lives. This kind of natural disasters has been reported vastly from
last few decades in different regions of the world. Soil Liquefaction
triggering occurs in silty and sandy soils. The huge damage due to
liquefaction at Niigata, Japan, and Alaska due to the earthquake that
occurred in 1964, extensively grabbed the attention of many geotechnical
researchers. SPT based empirical relationship is usually used to evaluate
soil liquefaction. However, a few parameters involved in the analysis are
associated with a great extent of uncertainties. A reliability-based
analysis provides an approach to consider various uncertainties and
provides the probability for the failure of the structure. Due to site
conditions and other reasons, it is difficult to obtain complete information
about a random variable. Therefore, very often we come across censored
samples. It is important that we design an reliable engineering structure
based on censored samples.
The primary objective of the research is to perform a reliability
analysis based on censored samples. The research focuses on developing
the deterministic, probabilistic and reliability-based models to calculate
soil liquefaction resistance using historical liquefaction database based
on the SPT. The principle of maximum entropy is incorporated to develop
probability density function that includes various uncertainties associated
with soil and site parameters.
With the development of computing techniques like artificial
intelligence, it is possible to frame the empirical relationship between the
seismic load and resistance offered by the soil. Standard penetration test
based database of soil liquefaction is used in the artificial neural network
to predict the liquefaction index. Further, the developed liquefaction
index model is utilized for modeling the empirical relationship between
clean sand equivalence corrected standard penetration test-N count and
cyclic resistance ratio. The deterministic model is developed, and the
relationship for estimating the resistance offered by soil to liquefaction is
established by identifying the best fit curve. Bayesian mapping theory is
used for determining the function for liquefaction probability. With the
knowledge about the expected values from the database, maximum
entropy distributions are plotted for seismic, site and soil parameters.
The developed probability density function of the random variables are
utilized for performing the first order reliability analysis. Using
sensitivity analysis, the degree of conservatism is identified and
eliminated from performance function. Finally, the calibrated
performance function is framed which can be used for performing
reliability analysis on truncated samples.
The truncated normal and log-normal probability density
function are developed using the information available on censored
samples. The parameters of the truncated normal distribution are
estimated using maximum likelihood method and Newton-Raphson’s
iterative procedure. When dealing with censored samples, the flow of
iteration points has a limitation in reliability analysis. For this reason, a
new algorithm is proposed to identify the reliability index for liquefaction
potential based on global search. | en_US |
