Geotechnically Informed Monitoring-Data Fusion and Gradient-Boosted Regression for Instrumented Driven Piles
- 1Ministry of Transportation of Ontario, Toronto, ON M9M 2L4, Canada
- 2Department of Civil and Environmental Engineering, University of Waterloo, Waterloo, ON N2L 3G1, Canada
- 3Geomechanics and Geotechnics, Institute of Geosciences, Kiel University, 24118 Kiel, Germany
Abstract
Long-term monitoring can test whether pile-design assumptions remain appropriate as consolidation and soil-pile load transfer develop after construction. This study combines records from three Ontario transportation projects: Ramsayville, Locha Creek, and Nash Road. The database contains 53,218 weekly movement records, 152 sensor channels, 1714 axial-load profiles, and ten instrumented piles. Baseline-referenced axial-force changes were interpreted together with pore pressure, ground and pile movement, soil-test results, pile geometry, load tests, and construction history. Conventional total-stress and effective-stress calculations were retained as engineering baselines, while gradient-boosted models were used to examine nonlinear residuals and the relative influence of measured inputs. A bounded cross-site mobilization relation achieved an in-sample RMSE of 100.0 kN (R2=0.71) and a leave-one-site-out RMSE of 110.2 kN. In a separate Ramsayville daily forecasting analysis, a seven-day moving average outperformed LightGBM and CatBoost, showing that greater model complexity did not improve short-term prediction. The confirmed cost of the full Ramsayville research monitoring program is CAD 678,700: CAD 267,200 for the test-pile program and CAD 411,500 for the production-pile program. A reduced regional seed-site program is estimated at CAD 225,700. At Locha Creek, the project reference exceeded the conditional 95% screening value by 7.9%, equivalent in cost to 2.30 m of installed pile length, or about CAD 919 per representative 29 m pile at CAD 400/m. The corrected break-even scale is approximately 246 similar piles. No screened reduction was supported at Ramsayville or Nash Road. The results define a monitoring-to-design framework that can identify both conservative and unconservative assumptions. Any design modification must still satisfy the governing geotechnical, structural, serviceability, drivability, group, founding, durability, and reliability requirements.
Related
- On The Linear Convergence of Bregman Proximal Gradient Methods with Applications to Kullback--Leibler regression
- On the Gradient Heterogeneity Dynamics of Adversarially Robust Federated Regression
- Gradient descent reaches arbitrary accuracy in a fixed number of large steps on multiclass logistic regression
- Statistical and Privacy Guarantees for Deep M-Regression via Stochastic Gradient Langevin Dynamics
- Minimax Rates for imperfect Physics-Informed Regression
BibTeX
@article{sangiuliano2026geotechnically,
title = {{Geotechnically Informed Monitoring-Data Fusion and Gradient-Boosted Regression for Instrumented Driven Piles}},
author = {Tony Sangiuliano and Mohammad Zaid and Zarghaam Rizvi and Dipanjan Basu},
journal = {Mathematics},
year = {2026},
month = oct,
doi = {10.3390/math14193632},
url = {https://www.mdpi.com/2227-7390/14/19/3632},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Geotechnically Informed Monitoring-Data Fusion and Gradient-Boosted Regression for Instrumented Driven Piles
AU - Tony Sangiuliano
AU - Mohammad Zaid
AU - Zarghaam Rizvi
AU - Dipanjan Basu
PY - 2026
DA - 2026/10/08
JO - Mathematics
DO - 10.3390/math14193632
UR - https://www.mdpi.com/2227-7390/14/19/3632
AB - Long-term monitoring can test whether pile-design assumptions remain appropriate as consolidation and soil-pile load transfer develop after construction. This study combines records from three Ontario transportation projects: Ramsayville, Locha Creek, and Nash Road. The database contains 53,218 weekly movement records, 152 sensor channels, 1714 axial-load profiles, and ten instrumented piles. Baseline-referenced axial-force changes were interpreted together with pore pressure, ground and pile movement, soil-test results, pile geometry, load tests, and construction history. Conventional total-stress and effective-stress calculations were retained as engineering baselines, while gradient-boosted models were used to examine nonlinear residuals and the relative influence of measured inputs. A bounded cross-site mobilization relation achieved an in-sample RMSE of 100.0 kN (R2=0.71) and a leave-one-site-out RMSE of 110.2 kN. In a separate Ramsayville daily forecasting analysis, a seven-day moving average outperformed LightGBM and CatBoost, showing that greater model complexity did not improve short-term prediction. The confirmed cost of the full Ramsayville research monitoring program is CAD 678,700: CAD 267,200 for the test-pile program and CAD 411,500 for the production-pile program. A reduced regional seed-site program is estimated at CAD 225,700. At Locha Creek, the project reference exceeded the conditional 95% screening value by 7.9%, equivalent in cost to 2.30 m of installed pile length, or about CAD 919 per representative 29 m pile at CAD 400/m. The corrected break-even scale is approximately 246 similar piles. No screened reduction was supported at Ramsayville or Nash Road. The results define a monitoring-to-design framework that can identify both conservative and unconservative assumptions. Any design modification must still satisfy the governing geotechnical, structural, serviceability, drivability, group, founding, durability, and reliability requirements.
SN - 2227-7390
ER -