Analysis and eva‎luation of Pavement Condition Using Smartphone Data

2/17/2026 12:00:00 AM

Road pavement, as a fundamental transportation infrastructure, plays a crucial role in economic and social development. Timely assessment of pavement distress not only enhances user safety but also reduces maintenance and repair costs. Traditional pavement assessment methods are time-consuming, costly, and reliant on field inspections. Therefore, the use of smart tools and advanced data processing techniques is essential to improve the speed and accuracy of pavement eva‎luation. This study aims to develop the most effective machine learning model for analyzing and assessing pavement conditions using data collected from smartphone sensors. In this research, data from accelerometers, gyroscopes, and GPS sensors were collected through the Sensor Record application. These data provided precise information about the vehicle’s dynamic conditions and geographic location. After preprocessing, including noise removal, normalization, and extraction of key features from the sensor data, the processed data were used to identify and analyze pavement distress. The data analysis revealed that 39.02% of the studied route exhibited various types of distress, including patching, alligator cracking, corrugation, and transverse cracking. In eva‎luating the performance of different machine learning models, the Multilayer Perceptron (MLP) artificial neural network achieved the highest accuracy of 93.98% in detecting pavement distress along the entire route. However, some algorithms performed better in identifying specific types of distress. For instance, the Naive Bayes algorithm demonstrated superior accuracy in detecting patching and alligator cracking, while the k-Nearest Neighbors (k-NN) algorithm performed better in identifying corrugation. In contrast, detecting transverse cracking posed a challenge due to its limited contact with vehicle tires and the low-intensity vibrations it generates. These differences highlight the impact of distress characteristics on model sensitivity and provide insights for selecting appropriate algorithms in future studies. Overall, the overlap of distress types posed challenges for machine learning models in accurate classification, emphasizing the necessity of collecting higher-quality data for more comprehensive analyses. To validate the accuracy of the developed model, the collected data were compared with an external dataset, including results from field reports. This validation confirmed the model’s effectiveness in identifying and predicting pavement distress. Future research is recommended to enhance accuracy and efficiency by integrating larger datasets and employing image processing techniques.

تلفن همراه هوشمند , شتاب‌سنج , ژيروسكوپ , GPS , خرابي روسازي , يادگيري ماشين

Smartphone , Accelerometer , Gyroscope , GPS , Pavement Damage , Machine Learning

Milad Shekarbeigi

Dr. Mahmoud Ameri _ Dr. Barat Mojaradi