Use of ANN and visual-manual classification for prediction of soil properties for paving purposes

ABSTRACT

The prospecting and preliminary analysis of soils for the understanding of their potential performance as subgrade, based on the AASHTO classification and CBR, generally require the investment of considerable financial resources, which often has a major impact on the final costs of the road project. An alternative approach that could facilitate the preliminary identification of the properties of materials would thus be an extremely positive advance for road construction. This study aims to developed models for the prediction of the AASHTO classification and the CBR in the normal (CBR-N) and intermediate energies (CBR-I) of the soils of the Brazilian state of Ceará. These models were based on the Multilayer Perceptron (MLP) type of Artificial Neural Network (ANN) modelling. The input data of the ANN consisted of the visual-manual classification of the soils according to their colour and particle size. The geotechnical database was derived from 1790 samples obtained from highway projects in Ceará, Brazil. The proposed models obtained an accuracy rate of 94.35% for the AASHTO classification, 90.14% for the CBR-N, and 95.11% for the CBR-I, with mean squared error (MSE) of 0.041, 0.082, and 0.045, respectively.

KEYWORDS:

• Low volume roads
• artificial neural network
• subgrade
• pavement modelling

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Conselho Nacional de Desenvolvimento Científico e Tecnológico.