What does this research mean for the field?

An interpretable deep learning framework integrating a GRU model with spatial and temporal attention can accurately predict jacking force in weathered rock and identify drive-specific operational parameter influences. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To establish an interpretable spatial-temporal deep learning framework using recurrent neural networks and attention mechanisms for predicting pipe jacking forces in highly weathered rock.

August 29, 2026Tunnelling and Underground Space TechnologyOpen Access

A two-dimensional spatial-temporal decision-support framework for pipe jacking in weathered rock

View Full Paper

Ask AI

Bookmark

Authors

LYLit Yen Yeo CCChung Siung Choo SLSue Han Lee

Discussion

Member takes

Overview

Deep learning study demonstrates accurate jacking force prediction in weathered rock microtunnelling drives, indicating stage-dependent operational influence across varying ground conditions.

Key Points

To establish an interpretable spatial-temporal deep learning framework using recurrent neural networks and attention mechanisms for predicting pipe jacking forces in highly weathered rock.
Benchmarked GRU, LSTM, and Conv1D architectures via random grid search optimization on field data from two full-scale microtunnelling drives in weathered phyllite (120 m) and sandstone (140 m).
Integrated the selected GRU network with Bahdanau spatial and temporal attention using a leak-free section-wise train-test split, mapping relative feature importance via post-hoc dot-product scoring.
Evaluated model generalizability across differing geological and mechanical conditions using direct cross-drive transfer and targeted fine-tuning.
The GRU architecture achieved the highest predictive performance, yielding an R² of 0.82 and RMSE of 7.93 ton in phyllite, and an R² of 0.97 and RMSE of 12.22 ton in sandstone.
Direct cross-drive transfer learning exhibited poor direct accuracy across distinct drives, but predicted-drive fine-tuning substantially improved model adaptability.
Spatial-temporal attention mapping demonstrated drive-specific dependencies, showing primary reliance on face support and thrust response in phyllite, compared to drive length, elapsed time, and jacking speed in sandstone.

Cite This Study

Yeo et al. (2026) studied this question.

synapsesocial.com/papers/6a92995f8e5d7d1fc0c113fc https://doi.org/10.1016/j.tust.2026.108069

View Full Paper

Ask AI

Bookmark

August 29, 2026Tunnelling and Underground Space TechnologyOpen Access

A two-dimensional spatial-temporal decision-support framework for pipe jacking in weathered rock

View Full Paper

Ask AI

Bookmark

Authors

LYLit Yen Yeo CCChung Siung Choo SLSue Han Lee

Discussion

Member takes

Overview

Deep learning study demonstrates accurate jacking force prediction in weathered rock microtunnelling drives, indicating stage-dependent operational influence across varying ground conditions.

Key Points

To establish an interpretable spatial-temporal deep learning framework using recurrent neural networks and attention mechanisms for predicting pipe jacking forces in highly weathered rock.
Benchmarked GRU, LSTM, and Conv1D architectures via random grid search optimization on field data from two full-scale microtunnelling drives in weathered phyllite (120 m) and sandstone (140 m).
Integrated the selected GRU network with Bahdanau spatial and temporal attention using a leak-free section-wise train-test split, mapping relative feature importance via post-hoc dot-product scoring.
Evaluated model generalizability across differing geological and mechanical conditions using direct cross-drive transfer and targeted fine-tuning.
The GRU architecture achieved the highest predictive performance, yielding an R² of 0.82 and RMSE of 7.93 ton in phyllite, and an R² of 0.97 and RMSE of 12.22 ton in sandstone.
Direct cross-drive transfer learning exhibited poor direct accuracy across distinct drives, but predicted-drive fine-tuning substantially improved model adaptability.
Spatial-temporal attention mapping demonstrated drive-specific dependencies, showing primary reliance on face support and thrust response in phyllite, compared to drive length, elapsed time, and jacking speed in sandstone.

Cite This Study

Yeo et al. (2026) studied this question.

synapsesocial.com/papers/6a92995f8e5d7d1fc0c113fc https://doi.org/10.1016/j.tust.2026.108069

View Full Paper

Ask AI

Bookmark