Maintenance and Rehabilitation Project Selection Using Artificial Neural Networks

When determining a preservation plan for a pavement management system, it is important to select projects of highest priority, as limited funding cannot meet all preservation needs. Currently, in the state of Texas, each district uses locally developed methods to prioritize projects based on the subjective judgment of engineers rather than on formal quantitative assessment. Artificial neural networks (ANN) can be trained with a series of statewide pavement preservation plans and corresponding PMIS data. This study applied the concept of artificial neural networks (ANN’s) to select projects for the 4 year plan of the Texas Department of Transportation (TxDOT). The data obtained for this study included the maintenance and rehabilitation schedules for selected road sections from 2009 to 2013 collected from the 4 year plans of all 25 TxDOT districts. TxDOT’s corresponding Pavement Management Information System (PMIS) data for 2009 was used as input data in the neural network. Based on attributes from the PMIS data, such as traffic and pavement distresses, each section in the PMIS was related to the 4 year plan data and identified as a project (=1) or not a project (=0). The dataset was then randomly divided into a train set (80% of data) and a test set (20% of data) for the ANN training and validation, respectively. After applying the data to an ANN using the data mining program, WEKA, an optimal network structure was developed that resulted in 99.85% of the training data and 98.90% of the test-data being correctly classified.