Banca de DEFESA: Lucas de Paula Vasques

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : Lucas de Paula Vasques
DATE: 25/07/2023
TIME: 09:00
LOCAL: PECC
TITLE:

PREDICTON OF THE BEHAVIOR OF PORTLAND CPIV CEMENT THROUGH THE USE OF MACHINE LEARNING TECHNIQUES WITH DIFFERENT DATABASES

KEY WORDS:

Machine Learning; Pozzolanic cement; Calcined clay; Compressive strength; Artificial Neural Networks

PAGES: 118
BIG AREA: Engenharias
AREA: Engenharia Civil
SUBÁREA: Construção Civil
SPECIALTY: Materiais e Componentes de Construção
SUMMARY:

The use of pozzolanic cements has grown in the field of civil construction, so more studies are needed to understand their properties. Machine Learning (ML) techniques have been used in recent research to analyze databases of properties of materials such as concrete and Portland cement. It was noted that there is an absence in the literature of studies analyzing the prediction of compressive strength of pozzolanic cements produced by a single factory, even more so with the use of inputs that are easier to obtain. Therefore, in this study, Artificial Neural Networks techniques were used in order to predict the compressive strength of cements containing calcined clay in their composition. Physical-chemical characterization data (such as oxide and main compound contents) of cement from a single industry were used to assemble 5 databases, varying in the types of inputs adopted. Each database was used to train a different algorithm that later had its performance analyzed using quality indicators such as R², RMSE and MAPE. Two of the algorithms stood out among the others, one using oxide contents (A1) and the other with principal compound contents (B2), the first being more efficient in forecasting than the second. For A1 and B2, respectively, values of MAPE (2.36% and 3.12%), RMSE (1.204 and 1.513 MPa) and mean absolute error (0.999 and 1.300 MPa) were considered efficient according to the existing literature, despite lower-than-expected R² values (0.50 and 0.51). Through the results obtained, it was possible to make considerably reliable predictions (with low relative and absolute errors) of the compressive strength at 28 days of the pozzolanic cement. This fact makes the models produced promising alternatives for the evaluation of this property in future research.

COMMITTEE MEMBERS:
Presidente - 1734221 - JOAO HENRIQUE DA SILVA REGO
Externo ao Programa - 2517970 - GREGORIO LUIS SILVA ARAUJO - UnBExterna à Instituição - ANDRIELLI MORAIS DE OLIVEIRA - UFG