| Project Title | Ballast and Soil Performance Separation by Using Instrumented Geo-grid & Machine Learning |
| University | Penn State Altoona |
| Principal Investigator(s) | PI: Hai Huang |
| Funding Source(s) and Amounts Provided (by each agency or organization) | Penn State Altoona Federal Share — $73,948 Penn State Altoona Match Share — $74,351 |
| Total Project Cost | $148,299 |
| Start and End Dates | 02/01/2022 - 05/01/2024 |
| Brief Description of Research Project | To study how different combinations of ballast and soil yield the same tie displacement under the same load, a simple FEM track model (Figure 2) was built with different combinations of ballast and soil moduli (Table 1). The results show that it is possible for tracks with different combinations of moduli to have the same overall track displacement under the same load. However, the interface between the ballast and the soil does show very different stress-strain characteristics for different scenarios although the overall vertical track displacement might be similar. There is a clear dividing line between the track with higher subgrade modulus and the one with lower subgrade modulus no matter what the ballast condition might be. In another words, the vertical stress vs. horizontal strain relationship at the interface of ballast and subgrade together with the track modulus measurement might be able to separate the ballast and soil performances. To measure and further study the vertical stress and the horizontal stain at the ballast-soil interface, this research team is proposing to install instrumentations such as stress cells and strain gauges on geogrids, which are typically installed in between the ballast and soil to improve the track bearing capacity. The final objective of this research is to develop ballast and soil performance characterization algorithms based on the instrumented geogrid data (both in the lab and the field) by using supervised machine learning techniques including the Logistic Regressions (LR) and the Supporting Vector Machine (SVM). |