Banca de DEFESA: Luciana Barbosa Amancio

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : Luciana Barbosa Amancio
DATE: 27/09/2023
TIME: 09:00
LOCAL: Sala Teams (On-line)
TITLE:

MULTILAYER PERCEPTRON NEURAL NETWORKS APPLIED IN THE FORECAST OF THE LOAD CAPACITY OF ISOLATED AND GROUPED PILES


KEY WORDS:

Multilayer perceptrons. Bearing capacity. Residual Loads. Instrumented load tests. Standard Penetration Test.


PAGES: 251
BIG AREA: Engenharias
AREA: Engenharia Civil
SUBÁREA: Geotécnica
SPECIALTY: Fundações e Escavações
SUMMARY:
Several researches have reported that the semi-empirical methodologies applied in the prediction
of the load capacity of isolated piles provide very dispersed results in relation to the values
acquired from load tests and, moreover, they disregard the effects of residual loads in their lateral
and tip portions. Considering that neural modeling with multilayer perceptrons has been
highlighted as a successful tool in predicting the behavior of isolated and grouped piles, it was
decided to use it in this research to develop prediction models for the lateral and tip plots of
isolated piles and also for the load capacity of groups of piles using the error back propagation
algorithm. To this end, results from SPT tests and instrumented static load tests were collected
from 120 isolated piles and from static load tests performed in 60 groups. With the information
related to the isolated piles, the lateral and tip plots of the isolated piles were calculated and
also the correction of these plots considering the presence of residual loads. This correction
implied a reduction in the lateral parcels and an increase in the tip parcels, of 22%. From the
load-strength curves corresponding to the 60 pile groups, the load capacities were determined
through the method of Terzaghi (1943). Then, the input variables for each of the neural models
were chosen, being the diameter, the resistance to penetration - SPT, the elevation of the water
level, the type of pile, the type of soil and the speed of the applied load those used in the models
referring to the isolated piles and the geometry of the group, the spacing between the piles, the
condition of the element that joins the piles and the load capacity provided by the neural model
developed in this work, those used in the models referring to the groups. After that, different
models were trained and validated with the help of the QNET2000 program. By comparing
the results obtained, it was possible to select a neural model for each variable that was to be
estimated. The correlation coefficients obtained in the validation phase of these models are between
0.88 and 0.99, a range that can be considered satisfactory for the prediction of a complex
phenomenon. Given the data used in the modeling, the proposed models for isolated piles had a
good performance for some types of piles (continuous and metallic propeller) while for others
(excavated) this did not occur. The model proposed to estimate the load capacity of pile groups
had an average performance, since it provided values far from the desired ones, especially for
the groups composed by hollowed piles or continuous flight augers. When applying the proposed
models in different cases, it was verified that the models related to the magnitudes of
the isolated piles had a better performance than the usual semi-empirical methodologies - Aoki
& Velloso (1975) and Décourt & Quaresma (1978) - and the model of the groups had a more
realistic behavior when the connecting element is not in contact with the soil. Therefore, it can
be stated that the multilayer perceptron is a promising tool for understanding the load transfer
mechanism that occurs both in isolated piles and in groups. It is worth remembering that the
difference between the models already available and those proposed in this work lies in the simplicity
of the architectures, in the fact that the lateral portion of the load capacity is calculated
as a function of depth and not just its total value, in the possibility of evaluating up to seven
types of piles, two types of loads (SML and QML) and different geotechnical profiles.

COMMITTEE MEMBERS:
Externo à Instituição - GUILHERME DE ALENCAR BARRETO - UFC
Externo à Instituição - CRISTHIAN CAMILO MENDOZA BOLAÑOS - UNC
Interno - 2161425 - JUAN FELIX RODRIGUEZ REBOLLEDO
Interno - 1549489 - MANOEL PORFIRIO CORDAO NETO
Presidente - 1122864 - RENATO PINTO DA CUNHA
Notícia cadastrada em: 04/09/2023 11:09
SIGAA | Secretaria de Tecnologia da Informação - STI - (61) 3107-0102 | Copyright © 2006-2024 - UFRN - app20.sigaa20