Análise do comportamento plástico de solos para otimização dos ensaios de deformação permanente em pavimentos

• Ano de defesa: 2023
• Autor(a) principal: Araújo, Matheus Jesus Ribeiro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Pavimentos asfálticos; Método de dimensionamento nacional; Redes Neurais Artificias; Random forest; Deformação permanente; Materiais granulares; Uso do solo; pavements; prediction models; Artificial Neural Networks; Random Forest; Permanent Deformation; granular materials; CNPQ::ENGENHARIAS::ENGENHARIA DE TRANSPORTES
• Link de acesso: http://repositorio.ufc.br/handle/riufc/74617

The new guidelines of the National Design Method (MeDiNa) propose the use of Permanent Deformation (PD) in the analysis of the plastic behavior of subgrade soils and structural layers of flexible pavements. However, factors such as laboratory procedure time and the cost of triaxial equipment might jeopardize this method. In this sense, this research suggests the analysis of PD test parameters for the development of optimized prediction models for geotechnical materials applied to pavement. For this purpose, 7 representative soil samples from the state of Ceará were assessed. After characterizing the PD of the soils using 9 stress pairs and 150,000 load application cycles, 3 main approaches were developed. In the first approach, predictive models based on geotechnical properties and physical indices were adjusted using Random Forest algorithms and Artificial Neural Networks. Both models performed satisfactorily, with correlation coefficients (R²) greater than 0.99. In the second approach, neural networks were implemented to predict the plastic behavior of soils in each single-stage test. After defining an optimal architecture, it was possible to identify a methodology to facilitate soil characterization with a reduced number of load application cycles. Thus, the initial 1,000 cycles of the PD test were removed when training the models, verifying that the rutting obtained from the MeDiNa software by reducing to 30,000 load cycles were similar to those with 150,000 cycles recommended by the regulations. The third approach analyzed the reduction of cycles in tests that showed plastic shakedown. The recommended method did not generate significant errors in the rutting obtained after around 30,000 cycles, while the development of a decision tree enabled the classification of tests in the A range of accommodation within the first 10,000 cycles. It is therefore hoped that the results of this work will permit the preliminary analysis of materials for use in paving works in the light of pavement mechanics, as well as helping to reduce the Permanent Deformation test period, thereby saving time and resources in road construction.