12. Application of Smartphones in Pavement Deterioration Identification Using Artificial Neural Network

The new generation of smartphones, equipped with various sensors such as a three-axis accelerometer, have shown potential as an intelligent, low-cost, crowd-based infrastructure monitoring platform over the past few years. This paper reports the results of an experimental study on using smartphones to identify different types of pavement deteriorations using artificial neural network (ANN). In an experimental study conducted in Blacksburg, VA, 92 responses, i.e., acceleration versus time responses in z direction, were recorded using smartphone accelerometers located in a moving vehicle. These responses were collected from different types of pavement deteriorations including speedbump, pothole, alligator cracking, and intact pavement. Then, ten different features were selected using signal-processing-based statistical techniques in both time-domain and frequency-domain to distinguish between different pavement deterioration types. ANN was then used for classification. The training techniques were Patternnet, Learning Vector Quantization 1 (LVQ1), and LVQ2 algorithms and their combination, i.e., being first trained using one of these techniques and being again trained using another technique. For model evaluation, repeated hold-out, leave-one-out cross-validation, and accuracy were used, and the average errors were reported for model comparison. According to the results, Patternnet and Patternnet+ LVQ2 provided the most accurate results with 93.48% and about 90% accuracies, respectively, while LVQ1 and LVQ1+LVQ2 did not reveal acceptable results.