Modelling Present Serviceability Rating of Highway Using Artificial Neural Network

8 Pages Posted: 7 Apr 2014

See all articles by Oladapo Abiola

Oladapo Abiola

Tshwane University of Technology - Department of Civil Engineering

W. K. Kupolati

Tshwane University of Technology - Department of Civil Engineering

Date Written: April 6, 2014

Abstract

Reliable pavement performance prediction models are essential for pavement design and preservation effort. Pavement performance is defined as the serviceability trend of the pavement over a design period of time. Serviceability indicates the ability of the pavement to serve and sustain the demand of the traffic in the existing condition. Pavement condition can be evaluated in four aspects: roughness, surface distress, structural capacity and skid resistance. In the analysis of the results of the road test conducted by American Association of State Highway Officials (AASHO), the subjective evaluation of serviceability by users was called the Present Serviceability Rating (PSR). The data used in modelling Pavement Serviceability Index (PSI), as reported by some authors, violate the basic assumptions of linear regression modelling in that it does not follow normal distribution. The objective of this study is to explore the relationship between the subjective Pavement Serviceability Rating (PSR) and objective index called Present Serviceability Index for highway sections in South-East, Nigeria. Artificial Neural Network (ANN) model was used to explore the relationship. The method of rating PSR is based on a five point scale: 0-1 (very good); 1-2 (good); 2-3 (fair); 3-4 (critical) and 4-5 (poor). International roughness Index (IRI) was converted to Slope Variance (SV). The input variables are rut depth, cracking, patching and SV. Back-propagation of ANN models with different activation function and number of hidden layers were trained and tested. The dataset was randomly split into three subsets, namely training (60%), testing (20%) and validation (20%) for the ANN model. The optimal models were evaluated with respect to forecasting error and coefficient of determination. Both Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for all predictions are plotted. Considering the architecture (4-18-1) with minimum MAE, RMSE and coefficient of determination, the table and figures show that the topology with one hidden layer with hyperbolic transfer function and hyperbolic transfer function for the output layer is the best. Comparison was made with multiple linear regression model which attempts to obtain a relationship between two or more explanatory variables and a response variable by fitting a linear equation to observed data. The results showed that the coefficient of determination for ANN model is 0.90 compared to 0.34 for regression model; ANN has demonstrated its ability to model non-linear data. This result confirms that the input variables are non-linear, and the ANN has shown to forecast with high degree of accuracy over regression analysis.

Keywords: Artificial neural network; pavement condition; present serviceability index; pavement serviceability rating; roughness.

Suggested Citation

Abiola, Oladapo and Kupolati, W. K., Modelling Present Serviceability Rating of Highway Using Artificial Neural Network (April 6, 2014). OIDA International Journal of Sustainable Development, Vol. 07, No. 01, pp. 91-98, 2014. Available at SSRN: https://ssrn.com/abstract=2420968

Oladapo Abiola (Contact Author)

Tshwane University of Technology - Department of Civil Engineering ( email )

Staatsartillerie Rd
Philip Nel Park
Pretoria, 0183
South Africa

W. K. Kupolati

Tshwane University of Technology - Department of Civil Engineering ( email )

Staatsartillerie Rd
Philip Nel Park
Pretoria, 0183
South Africa

Register to save articles to
your library

Register

Paper statistics

Downloads
191
Abstract Views
878
rank
157,331
PlumX Metrics