Skip to main content
ICTD 2023 Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Poster Presentation at the Conference, with a Paper in Conference Proceedings

Machine Learning & AI Applications in Pavements

Session: Meet the Author - Poster Session PS-3

1370970 - Application of Machine Learning for Joint Detection of Concrete Pavements

Friday, June 16, 2023
8:30 AM – 10:00 AM CST
Room: Governor's Grand Ballroom
Salon ABC
TRACK: Poster Presentation with a Paper in the Proceedings

    Primary Author(s)

  • Christian A. Sabillon

    Graduate Research Assistant
    University of Texas at Austin
    Austin, TX, United States

    • Presenter(s)

  • Christian A. Sabillon

    Graduate Research Assistant
    University of Texas at Austin
    Austin, TX, United States

    • Co-Author(s)

  • Joaquin B. Hernandez, n/a

    Research Engineering/ Scientist Associate II
    University of Texas at Austin, United States

  • Carlo Garcia, n/a

    Undergraduate Research Assistant
    University of Texas at Austin, United States

  • Ruohan Li, n/a

    Graduate Research Assistant
    University of Texas at Austin, United States

  • Jorge A. Prozzi

    Professor
    University of Texas at Austin, Texas, United States

To date, there is no agreed-upon accurate standard method for collecting automated pavement faulting, even though state departments of transportation are required by Federal Highway Administration (FHWA) to report average faulting measurement every 0.1 miles. This issue arises because, for network-level, faulting measurements require a joint to be identified at highway speed and automating a joint detection algorithm is a complex task.
There is a high demand for an automated joint detection algorithm that is accurate, quick, simple to understand, easy to implement, and consistent across different rigid pavement surface finishings. This study collected a sample of 573 transverse joints from jointed concrete pavements with two different surface finishings in the state of Texas using a custom-built high-speed laser profiler. These data were later used to train and validate a robust joint detection algorithm using machine learning models.
The results indicate that joint detection rates vary drastically depending on the type of surface finishing. Furthermore, selection of which machine learning model to use in joint identification depends on the type of input data available. A deep neural network architecture is preferable when only the profile heights of the 2D data are available. The accuracy of these models is in the order of 70% to 80% in terms of the F1-score. Whereas a random forest classifier is ideal when a set of fourteen texture statistics is available. This model has the shortest computation time, but could yield a lower accuracy, ranging from 50% to 80% in F1-score.

Learning Objectives:

Attendees can expect to learn the following from this session:
  • Understand why joint detection is such a critical component for faulting calculations and why it is so challenging
  • Comprehend the framework of neural networks and random forest models and the reasons for which these models where chosen as the classifiers of preference
  • Visualize an innovative data collection prototype being developed at UT Austin that is used for detecting joints and, soon, for computing pavement faulting dynamically