Predicting Highway Friction on an Annual Basis on Texas Network

Details

• Creators:
  Sabillon-Orellana, Christian ; Inoue, Danilo ; Li, Ruohan ; Hooton, Anna ; Huang, Robing ; Xu, Hongbin ; Prozzi, Jorge A.
• Corporate Creators:
  University of Texas at Austin. Center for Transportation Research
• Corporate Contributors:
  Texas Department of Transportation. Research and Technology Implementation Office ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  • Pavement surface course
  • Friction
  • Skid resistance tests
  • Machine learning
  • Artificial intelligence
  • Data analysis
  • Regression analysis
  • Mathematical prediction
  • skid number
  • grip number
• Publication/ Report Number:
  FHWA/TX-27/0-7031-01-1
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States ; Texas
• Corporate Publisher:
  University of Texas at Austin. Center for Transportation Research
• Abstract:
  The research project developed a network-level model to predict highway friction at the network level on an annual basis. By analyzing pavement texture data and employing advanced artificial intelligence and machine learning techniques, the study established a quantifiable relationship between surface texture characteristics and friction. Extensive field data was collected using a specialized data collection prototype across various pavement types and climate conditions, leading to a robust predictive model for friction. Traditional methods, such as regression, showed limitations, including underfitting and the need for frequent recalibration due to surface deterioration. To address these issues, the researchers used a representative database of Texas pavement surfaces, ensuring the models' statewide applicability and employed machine learning combining unsupervised and supervised learning techniques. Cluster analysis grouped pavement sections with similar characteristics then, regression models were used to predict pavement friction. These clusters served as labels to train a classification algorithm, achieving an F1 score of 0.903 with the random forest model. An advanced ensemble linear regression model captured the complex relationships between texture indices, surface information, and pavement friction, achieving an R2adj of 0.767, a mean absolute error (MAE) of 0.050, and a root mean square error (RMSE) of 0.071. This model outperformed a deep neural network regression model, which had an R2adj of 0.669, an MAE of 0.066, and an RMSE of 0.085. The algorithms were integrated into the Texan Texture Friction Forecaster application, capable of processing raw laser sensor data, computing indices, and predicting friction using the developed models.
• Format:
  PDF
• Funding:
  0-7031-01
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/93505/dot_93505_DS1.pdf
• File Type:
  [PDF - 4.55 MB]
• Collection(s):
  US Transportation Collection
• Main Document Checksum:
  urn:sha-512:895fec5a7200f699237119b9320da75ac6e6ffde94f7357e498baa4eb127f3d9959f3f79ee8f39126d78153c2bfed8ad0a74935688e9c45d4a3e1e994f9d33a1