Evaluation of delayed compression of gypsies soils with emphasis on neural network approach

Dissertant

al-Nuaymi, Muhammad Abd al-Latif Mahmud

University

University of Technology

Faculty

-

Department

Department of Building and Construction Engineering

University Country

Iraq

Degree

Ph.D.

Degree Date

2006

English Abstract

In this research, long - term behavior of gypsums soils is studied.

Four sites were selected.

Gypsum content in these soils ranges in between (37.5 -g4 5%).

Classification and chemical tests are performed on the selected soils to ensure the suitability of such tests on the gypsums soils.

Moreover, standard tests (one dimensional compression, single collapse test and double odometer tests) are carried out to define the mechanical characteristics of tested soils.

Odometer and permeability-leaching cells are used to study the behavior 0f gypseous soil under long term (soaking and washing) at (200 kPa) applied stress.

Furthermore, for this purpose, physical model is built to simulate the field conditions.

The soils were examined during (30 days) in the cells while in the laboratory model, the period was (180 days) for Al-Qarma and Al-Jaghaify soils and (30 days) for Al-Anbar University and Ain Al-Timor Soils.

Also, gypsums soils are examined under repeated loading with various intervals using laboratory model.

It was found from results of repeated loading that most compression occurs in the early cycles of loading.

Then deformation continues at slow rate until it reaches constant rate in final cycles.

It was found that one day of soaking is not enough to define the problem of severity of gypsums soils since the settlement continues along soaking period which refers to the delayed compression or creep occurrence in the soils.

Therefore, the collapsibility of gypsums soils may be considered as time dependent.

In this work, the results of testing are analyzed in statistical models using STATISTICA Computer Package.

Artificial Neural Networks (ANNs) are used to relate the properties of gypsums soils and evaluate the values of delayed compression for such types of soils under different conditions.

Therefore, multi-layer perception trainings using back propagation algorithm are used to assess the validity of application of ANNs for modeling the delayed compression (creep) of gypsums soils.

In this work, the geometry and construction of ANNs are investigated to obtain the optimum performance of Ann's model.

It was found that ANNs have ability to predict the secondary compression of gypsums soil due to soaking, leaching process and repeated loading with high degree of accuracy.

Also, performance of ANNs showed that one hidden layer with three hidden nodes is practically enough for the neural network analysis to concern the creep parameters and no clear relationship between the proportion of the data is used in each subsets and model performance.

The sensitivity analysis indicated that the type of soil and time have the most significant effect on the predicated delayed compression of gypsums soils followed by the gypsum content.

The i exults also, 'how that the type of test has the smallest impact on delayed compression of gypsums soils, while the initial void ratio, plasticity index and specific gravity have a moderate impact on the secondary compression of gypsums soils.

Finally, it was found from results of artificial neural network that the regression values are more than those obtained by using STATISTICA computer package.

Therefore, ANNs can be used instead of statistical software to show the nonlinear function of any problems.

Main Subjects

Civil Engineering

Topics

• Mechanics
• Soil physics
• Soil mechanics
• Neural networks(Computer science)

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-306160