Identification of expansive and collapsible soils in northeastern Brazil from Artificial Neural Networks generated in Pernambuco

• Autor(a) principal: Holanda, Maria Julia de Oliveira
• Data de Publicação: 2021
• Outros Autores: Ferreira, Silvio Romero de Melo, Amorim, Samuel Franca, Borges, Jesce John Silva, Silva, Larissa Ferreira da
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/22541
• topic: Expansive Soils; Collapsible Soils; Artificial Neural Networks; Classification equation.; Suelos Expansivos; Suelos Colapsables; Redes Neuronales Artificiales; Ecuación de clasificación.; Solos Expansivos; Solos Colapsíveis; Redes Neurais Artificiais; Equação de classificação.

Collapsible and expansive soils are problematic in Civil Engineering, causing pathologies in buildings due to the variation in volume with the change in humidity. The identification of these soils in the design phase is important. The paper aims to develop an Artificial Neural Network architecture trained with soils from Pernambuco, to identify expansive and collapsible soils, and expand its application to soils from other states in Northeastern Brazil. Developed from 87 samples, divided between training (53 samples), selection (17 samples) and test (17 samples) groups, according to 4 input variables, percentage of sand, percentage of clay, plasticity and activity indices. The best network architecture consists of 4 neurons at the input and 1 at the output. For the blind validation of the model, the network was applied to 45 samples of collapsible and expansive soils from other Northeastern states. The performance analysis of the classification accuracy of the network with data from Pernambuco showed an accuracy rate of 76.5% and in the validation in the other Northeastern states, pattern recognition was even higher, reaching an accuracy of 91.1%, demonstrating capacity capturing trends in soil surface movement and aiding in problem solving.