Banca de DEFESA: Abenezer Tefera Tanga

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : Abenezer Tefera Tanga
DATE: 27/09/2022
TIME: 15:00
LOCAL: TEAMS
TITLE:

MACHINE LEARNING FOR GEOMEMBRANE-SAND INTERFACE ANALYSIS 

KEY WORDS:

 Random Forest, Machine Learning, Geomembrane, Interface Shear & Friction angle

PAGES: 84
BIG AREA: Engenharias
AREA: Engenharia Civil
SUMMARY:

 Evaluation of soil-geosynthetic interaction is important for analyzing the stability of the overall structure. This is because the interaction between the reinforcing geosynthetics and the reinforced soil can be a start for breakage which may cause a structural failure. Several researchers have been studying factors determining the interface shear strength in the laboratory and identified various components which affect the overall strength outcome. Machine Learning has thatgreat potential for the analysis of parameters which are influenced by many variables. This dissertation brings a discussion about the use of Random Forest regression, which is a Machine Learning algorithm, for predicting geomembrane-sand interface friction angle.  

 

 The interfaces subjected for strength parameter investigation include geomembrane and cohesionless soil, and 495 interfaces from various literature were collected. The acquired interface data is utilized for the overall statistical and Machine Learning analysis. Fourteen parameters were recorded from the referred literatures as the factors determining the interface shear strength. The fourteen parameters are from three main interface components which are; laboratory test type, geomembrane properties and soil properties.  

 The presented data has been studied by using simple linear regression before initializing the Random Forest, to evaluate the interdependence between pairs of influencing parameters and their correlation with interface friction angle. The Pearson's correlation coefficient results are indicating, the influence level between the interface components is mostly not strong. These correlation values imply the nonlinear distribution of the database and the importance of a multivariate and nonlinear algorithm for studying the referred types of interfaces.  

 After the data analysis, an optimized Random Forest has been initialized to predict interface friction angle. It is observed only for 3% of the training set and 6% of the testing set that the friction angle estimation has exceeded ±5° from the laboratory records. The coefficient of determination measures shows strong coherence between friction angles from laboratory studies and Random Forest estimations by resulting R² = 0.93 and R² = 0.92; for the training and testing sets respectively.  Thus, the Random Forest has forecasted interface friction angle adequately

BANKING MEMBERS:
Externo à Instituição - KRISHNA RAJIREDDIGARI REDDY,
Interno - 404375 - ENNIO MARQUES PALMEIRA
Presidente - 2517970 - GREGORIO LUIS SILVA ARAUJO