Método de previsão de defeitos em estradas vicinais de terra com base no uso das redes neurais artificiais: trecho de Aquiraz - CE

Abstract

The representative extension of the unpaved roads in Ceará and in Brazil and their relevance for the socioeconomic development, they justify the need of programs of regular and preventive maintenance in those roads, whose badly conservation state systematically delays and it burdens the drainage of the production, the circulation of asset, services and of people between the rural areas and the urban centers. This dissertation presents a method of forecast of problems in unpaved roads, elaborated with base in the Nets Artificial Neural (ANN), technique that have been applied in the recognition of patterns. The area of the case study is the municipal district of Aquiraz/CE that presents representative characteristics. The developed method looked for to make possible the forecast of the problems and of the respective severities, starting from selected local attributes, to subsidize the activities of a roads management system, especially the process of decision with regard to the allocation of the available financial resources for the interventions to objectify the use of these roads during the whole year. The method was served as experimental procedures as: the selection and identification of the attributes; the identification and evaluation of the species of defects with relationship to the severity degrees with base in the suitable methods and adaptations offered by the author; the ordination of the inputs and exit for the application of RNA; the formation of the partitions of data; the choice by Multilayer Percerptron (MLP) and of the parameters of training of ANN, culminating with the training processes, validation and test. A ANN was structured with use of the software Easy NN Plus. After this, there was a development of an interface that simulates the operation of the ANN model and it favors the use of the method for no experts users, which suit with the reality of the municipal districts. This application of ANN was shown viable to the view of the acting of 91% obtained, what took to the practical consideration of the identity relationship between the found results and the problems and severities observed in reality. The method is applicable to other areas, being enough the procedures to be adopted in agreement with the particularities and local restrictions.