Thermal integrity profiling of concrete piles incorporating three-dimensional defects using deep convolutional neural networks
File(s)
Author(s)
Sanchez Fernandez, Javier
Ruiz Lopez, Agustin
Taborda, David
Type
Journal Article
Abstract
Thermal Integrity Profiling (TIP) has emerged as a powerful non-destructive method for evaluating cast-in-place concrete piles by monitoring heat of hydration (HoH) temperature profiles using temperature probes embedded in the piles. However, conventional TIP
interpretation relies either on manual inspection of a series of temperature-depth profiles over time, which is inherently prone to subjective bias, or on back-analysis using simplified modelling approaches. This study presents, for the first time, a fully three-dimensional TIP
framework combining Finite Element (FE) simulations with two deep neural networks (NN). A multi-label classifier detects up to three defects per pile, while a regressor quantifies defect volume, location, reinforcement-cage displacement, and hydration parameters. A database with hydration temperature profiles from 19,445 simulations encompassing randomised pile lengths, soil stratigraphies, cage misalignments and multi-defect configurations, is used to train the two NNs. The classifier achieves a per-defect accuracy of 93.6% (96.8% per sample). The
regression model achieves near-perfect accuracy for cage displacement (R² = 0.992) and defect volume (mean absolute error of 0.03 m³), and an adjusted R² of 0.92 for defect location. This novel automated approach significantly enhances the objectivity, test speed, and reliability of TIP-based defect detection across a wide range of deep pile foundations.
interpretation relies either on manual inspection of a series of temperature-depth profiles over time, which is inherently prone to subjective bias, or on back-analysis using simplified modelling approaches. This study presents, for the first time, a fully three-dimensional TIP
framework combining Finite Element (FE) simulations with two deep neural networks (NN). A multi-label classifier detects up to three defects per pile, while a regressor quantifies defect volume, location, reinforcement-cage displacement, and hydration parameters. A database with hydration temperature profiles from 19,445 simulations encompassing randomised pile lengths, soil stratigraphies, cage misalignments and multi-defect configurations, is used to train the two NNs. The classifier achieves a per-defect accuracy of 93.6% (96.8% per sample). The
regression model achieves near-perfect accuracy for cage displacement (R² = 0.992) and defect volume (mean absolute error of 0.03 m³), and an adjusted R² of 0.92 for defect location. This novel automated approach significantly enhances the objectivity, test speed, and reliability of TIP-based defect detection across a wide range of deep pile foundations.
Date Acceptance
2026-06-15
Citation
Geomechanics and Geoengineering, pp.1-16
ISSN
1748-6025
Publisher
Taylor and Francis Group
Start Page
1
End Page
16
Journal / Book Title
Geomechanics and Geoengineering
Copyright Statement
Copyright This paper is embargoed until publication. Once published the author’s accepted manuscript will be made available under a CC-BY License in accordance with Imperial’s Research Publications Open Access policy (www.imperial.ac.uk/oa-policy).
License URL
Publication Status
Published online
Date Publish Online
2026-06-22