Data-Based Modeling Approaches for Short-Term Prediction of Embankment Settlement Using Magnetic Extensometer Time-Series Data

Faisal Siddiqui, Paul Sargent, Gary Montague

Research output: Contribution to journalArticlepeer-review

Abstract

Developing data-driven predictive models is highly desirable for monitoring the condition of infrastructure assets but is dependent on the generation of large data sets that are regularly updated. This represents a challenge in modern geotechnical infrastructure projects such as earth embankments, where the size of settlement monitoring data sets is generally small and of low resolution. While long-term settlement predictions for embankment structures are useful for design engineers, short-term predictions are more valuable to site engineers who are required to make operational decisions regarding construction. Their challenge is greater on sites where ground conditions are complex. The purpose of this study is to explore the applicability of parametric data-driven methods (namely polynomial curve fitting and transfer function methods) to forecast the trend of soil settlement in real-time using magnetic extensometer and embankment fill-level data. An industrial data set was sourced for a highway earth embankment, which was founded on a sequence of interbedded glacial soils. Polynomials models were more effective in predicting settlement during earlier stages of embankment construction when information on the influence of loading on settlement is limited. As this information grows, transfer functions are preferable in terms of quality of prediction. The findings from this study highlight the potential for wider use of data-driven approaches to assist in earth embankment construction.
Original languageEnglish
JournalInternational Journal of Geomechanics
Volume22
Issue number2
DOIs
Publication statusPublished - 22 Nov 2021

Fingerprint

Dive into the research topics of 'Data-Based Modeling Approaches for Short-Term Prediction of Embankment Settlement Using Magnetic Extensometer Time-Series Data'. Together they form a unique fingerprint.

Cite this