Automated decision making in highway pavement preventive maintenance based on deep learning

Abstract

Decision-making in highway preventive maintenance (PM) is generally costly and complicated. An inappropriate maintenance strategy could yield a low efficiency of budget usage and untreated road distress. This study describes an innovative predictive maintenance strategy that provides direct maintenance guidance to specific highway mileposts. This was achieved with the application of the artificial neural network (ANN) algorithm to mine a maintenance database. Ten-year distress measurement data at 100-m intervals, traffic load data, climatic history, and maintenance records of a chosen highway were regarded as the input data of the ANN model. A data quality control method was proposed to ensure asphalt pavement performance improvement continuity over time based on the idea of the maintenance year as the starting point for prediction. The backpropagation neural network (BPNN) model and a hybrid neural network (HNN) were applied to predict five indexes of the highway asphalt pavement performance, and the genetic algorithm (GA) was employed to optimize the hyperparameters of these models. The results indicate that the GA enhanced HNN model could increase the accuracy by 35% on average compared with traditional ANN in predicting the highway asphalt distress performance. Furthermore, a notable agreement is attained when comparing the predicted indexes to the whole-year measurement data invalidation with average coefficient of determination (R2) reaches 0.74. This study demonstrates the potential of an innovative ANN method in highway distress prediction to provide direct guidance for long-term highway asphalt pavement optimal rehabilitation and maintenance (R&M) decisions.