Identificação da condição funcional de pavimentos urbanos auxiliado por machine learning

Abstract

The predominance of the highway mode in Brazil and its intensive use by the population make the responsible managers responsible for the maintenance and rehabilitation of the built roads. However, many municipalities do not have sufficient resources to perform data surveys to assess the current road situation under their jurisdiction. Due to the recurring financial difficulties in several parts of the country to maintain its roads, this work proposes a model for identifying the functional conditions of urban pavements, to help managers make decisions. The proposed model involves the use of an unsupervised Machine Learning algorithm called k-means, which, together with the accelerometer, gyroscope and receiver of the Global Positioning System (GPS), present in smartphones, generate georeferenced images. which indicate, every second, the condition of the pavement interpreted by the model. Data from 135 km of stretches in the city of Fortaleza studied about linear acceleration from the accelerometer and angular acceleration from the gyroscope were processed in conjunction with the vehicle location at the time of survey and statistical parameters obtained as a function of numerical values of accelerations, thus forming four labeled groupings of different types of pavement conditions. One of its possible uses of this model is the generation of situation maps to help managers make decisions regarding road maintenance and rehabilitation. It was realized that the model produced has the potential to become a useful tool for municipal managers as a viable alternative to replace more costly and time consuming methods, especially in places with no or limited financial, material or human resources for this purpose.