The use of data analytics for monitoring and predicting the short-term future performance of geotechnical embankments

  • Faisal Siddiqui

Student thesis: Doctoral Thesis


For earth embankment construction projects, magnetic extensometers are installed within the foundation soils to verify the ultimate settlement post-construction predicted at the design stage. The uses of settlement datasets collected from these instruments have been under-utilised before embankment completion due to their limited frequency and low resolution. Therefore, short-term predictions have not yet been thoroughly investigated for datasets collected from full-scale embankments. Such predictions would be valuable to site engineers who make operational decisions regarding embankment construction. This research considers these opportunities and proposes a comprehensive data-driven framework comprising data preprocessing and short-term predictive modelling.
At the data preprocessing stage, principal component analysis was used to explore raw soil settlement data and identify outliers. Furthermore, filtering methods (including moving average, Gaussian-weighted moving average, Savitzky-Golay and zero phase) were used to remove noise from the raw data. Statistically, these methods showed similar performances in removing noise. These preprocessing steps are essential before using data to train predictive models.
For short-term predictions, polynomial curve fitting and transfer function were examined, as parametric data-driven methods, for forecasting soil settlement trends in real-time. When validated through the moving time window approach to compare consecutive 7th day predictions with field measurements, the transfer function model performed better (R-squared value for validation = 0.98) than the polynomial curve fitting model (R-squared value for validation = 0.89). Nevertheless, both models captured on-site behaviour to a standard that meets ground engineering requirements.
The findings from this research highlight the potential of small datasets in extracting rich information suitable for data-based decision-making during earth embankment construction. This study offers a complementary data-driven approach intended to reduce disruption to construction schedules for geotechnical infrastructure.
Date of Award1 Oct 2022
Original languageEnglish
Awarding Institution
  • Teesside University
SupervisorGary Montague (Supervisor), Paul Sargent (Supervisor) & Nashwan Dawood (Supervisor)

Cite this