Coupling numerical modeling and machine-learning for back analysis of cantilever retaining wall failure

Amichai Mitelman, Gili Lifshitz Sherzer

Research output: Contribution to journalArticlepeer-review

1 Scopus citations

Abstract

In this paper we back-analyze a failure event of a 9 m high concrete cantilever wall subjected to earth loading. Granular soil was deposited into the space between the wall and a nearby rock slope. The wall segments were not designed to carry lateral earth loading and collapsed due to excessive bending. As many geotechnical programs rely on the Mohr-Coulomb (MC) criterion for elastoplastic analysis, it is useful to apply this failure criterion to the concrete material. Accordingly, the back-analysis is aimed to search for the suitable MC parameters of the concrete. For this study, we propose a methodology for accelerating the back-analysis task by automating the numerical modeling procedure and applying a machine-learning (ML) analysis on FE model results. Through this analysis it is found that the residual cohesion and friction angle have a highly significant impact on model results. Compared to traditional back-analysis studies where good agreement between model and reality are deemed successful based on a limited number of models, the current ML analysis demonstrate that a range of possible combinations of parameters can yield similar results. The proposed methodology can be modified for similar calibration and back-analysis tasks.

Original languageEnglish
Pages (from-to)307-314
Number of pages8
JournalComputers and Concrete
Volume31
Issue number4
DOIs
StatePublished - 1 Apr 2023
Externally publishedYes

Keywords

  • back-analysis
  • cantilever wall
  • concrete
  • failure
  • machine-learning
  • Mohr-Coulomb

ASJC Scopus subject areas

  • Computational Mechanics

Fingerprint

Dive into the research topics of 'Coupling numerical modeling and machine-learning for back analysis of cantilever retaining wall failure'. Together they form a unique fingerprint.

Cite this