Uso de machine learning para a classificação e predição da capacidade de autocura de pasta cimentícea por meio do uso de bactérias

• Ano de defesa: 2023
• Autor(a) principal: Oliveira, Franciana Sokoloski de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Mato Grosso; Brasil; Instituto de Ciências Exatas e da Terra (ICET) – Araguaia; UFMT CUA - Araguaia; Programa de Pós-Graduação em Ciência de Materiais
• Palavras-chave em Português: Bactéria self-healing; Linguagem R; Materiais cimentíceos; CNPQ::CIENCIAS EXATAS E DA TERRA; Self-healing bacteria; R language; Cementitious materials
• Link de acesso: http://ri.ufmt.br/handle/1/5608

The research is designated as the study of self-healing in cement paste using self-healing bacteria using the Support Vector Machine-Machine Learning algorithm. The data used to implement the forecast were collected from studies carried out previously. There were 38 (thirty-eight) samples, using six bacteria: Bacillus cohnii, Bacillus subtilis, Bacillus megaterium, Bacillus sphaericus, Bacillus mucilaginous and Bacillus pseudofirmus. Input data referring to the amount of cement, water, sand, aggregate, calcium lactate and plasticizer, water/cement factor, the size of the cured crack, the curing time and the percentage of selfhealing were used. In this study, the R programming language was used, where the parameters gamma, cost, type and coef0 were kept for all data sets, the kernel and cross validation values were changed. The kernel varied between sigmoid, radial and linear and the cross value was 10, 5 or 2. The study aims to evaluate the accuracy of the prediction of self-curing of concrete with statistical modeling in R language based on the evaluation criteria of performance: explained variance (R²), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). For each data set, an R², MAE and RSME and their respective percentages were obtained. The R² of the samples that had the radial kernel parameter obtained better results with values of 0,897 and 0,976. The RSME was reached for all samples, which had values close to zero, with the highest value being close to 0,14. Thus, there are few significant errors in the model. The MAE values ranged from 0,0625 to 0,0264, equivalent to 6,52% to 2,64%, with the lowest value being that of the radial kernel. Water, sand, lactate, plasticizer, water/cement factor and curing time are variables that receive good interaction with bacteria, however these, water, lactate and water/cement factor, are the ones that protect the most in the cement paste. Considering the three evaluative metrics R², RSME and MAE, and analyzing the graphs generated by R, it can be concluded that the predictive model has good potential for predicting the self-healing of cement-based materials.