Uma abordagem alternava para obtenção da classificação de solos da AASHTO usando redes neurais artificiais

Abstract

The soil survey and laboratory tests to analyze the soil general rang as subgrade of roads using the AASHTO classificaon, usually have a high financial cost for roads pro-jects, in Ceará state the cost of geotechnics services for pavement design is esmated at 30%. An alternave way to idenfy preliminarily a soil’s qualies rapidly just with field soil analysis would be posive to paving. The aims of this paper are an arficial neural network framework that processes qualitave field test data to predicon AASHTO soil classificaon. The data of the Visual-Manual classificaon of soils, which makes it possi-ble to verify preliminary the parcle size and color of the material, were used as explan-atory variables. Thus, was created a database with 1790 soil samples, which were ex-tracted from pre-exisng projects, provided by Naonal Department of Transportaon Infrastructure and Department of Transportaon of Ceará state. The proposed model presented an accuracy rate of 94.5%, in the average of the esmates for the AASHTO classificaon, and an error of the order of 0.04, considered the mean square of errors (MSE).