Deep Soil Mixing and Predictive Neural Network Models

Add to your list(s) Download to your calendar using vCal

• Ms Rakshya Shrestha, CUED
• Friday 16 March 2012, 16:00-17:00
• Engineering Department - Lecture Room 3B.

If you have a question about this talk, please contact Anama Lowday.

The work is basically a fusion of two existing research areas: Deep Soil Mixing (DSM) and Artificial Neural Networks (ANNs). The strength of the deep soil mixed walls and columns depends on a large number of factors which vary in a wide range and the dependence is complex. Soil type, grain size distribution, water content, plasticity index, liquidity index, clay content, organic matter content, binder type, water to binder ratio, mixing and curing conditions are few of the many variables that affect the strength development. The variability and uncertainty associated with these variables affect the magnitude of strength making the estimation of strength not a very straight forward task. To investigate this variability, data on a large number of these variables has been collated into a database from a large number of DSM projects worldwide. Uniformity and variability in the strength data have been analysed based on the data from these sources. The effect of parameters such as water content, liquidity index, cement content, total water to cement ratio and curing times on the UCS of the binder mixed soils has been studied by plotting these data together. The study has highlighted the importance of the variables such as liquidity index and total water to binder ratio which have not been extensively explored in previous DSM studies. Predictive Neural Network models which predict the strength gain of cement-stabilized clays as a function of clay water content, liquidity index, plasticity index, organic matter content, grain size distribution, cement content, total water to cement ratio and curing time have been developed. Results from the neural network models were found to emulate the known trends and reasonable estimates of strength as a function of the selected variables were obtained. The effectiveness of these data-driven non-linear predictive models is discussed.

This talk is part of the Engineering Department Geotechnical Research Seminars series.

This talk is included in these lists:

• Alex
• All Talks (aka the CURE list)
• Cambridge University Engineering Department Talks
• Cambridge talks
• Centre for Smart Infrastructure & Construction
• Civil Engineering Talks
• Computational Continuum Mechanics Group Seminars
• Engineering Department - Lecture Room 3B
• Engineering Department Geotechnical Research Seminars
• Featured lists
• Interested Talks
• School of Technology
• Trust & Technology Initiative - interesting events
• bld31

Note that ex-directory lists are not shown.