Optimisation of Tunnel Boring Machine Performance by Machine Learning

Shan, Feng

Optimisation of Tunnel Boring Machine Performance by Machine Learning

Shan, Feng

Permalink

Publication Type:: Thesis
Issue Date:: 2024

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (6.45 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Shan, Feng
dc.date.accessioned	2024-10-04T03:48:55Z
dc.date.available	2024-10-04T03:48:55Z
dc.date.issued	2024
dc.identifier.uri	http://hdl.handle.net/10453/181183
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Tunnel boring machines (TBMs) have been widely utilised in tunnel construction due to their efficiency and reliability. However, tunnel collapse, rock bursting, water inrush, and machine jamming remain severe challenges in complex geotechnical conditions. The main objective of this thesis is to investigate if and to what extent TBM operation can be forecasted in real time by machine learning, and if such forecasts can result in TBM tunnelling optimisation, dissecting it into four aspects: (1) penetration rate regression, (2) penetration rate forecasting, (3) cutterhead torque and thrust force forecasting, and (4) rock mass classification. Penetration rate regression is to explore the relationship between penetration rate and other related parameters. Traditional theoretical and empirical methods often provide less accurate results, suited for overall project time management before the start of a project. Machine learning models, using data from the Pahang-Selangor water tunnel, demonstrate better results with support vector machine (SVM), random forest (RF), and artificial neural network (ANN). While other models inclusive of operational and geological parameters can have high accuracy, their applicability is questioned because operational parameters are not accessible during training. Real-time forecasting of penetration rates leverages historical data to predict unknown penetration rates in the future. Accurate forecasts, even for a short distance ahead, can aid in refining TBM operations, resource allocation, decision-making process, and early warning of unexpected geotechnical conditions. We build recurrent neural network (RNN) and long short-term memory (LSTM) models to predict short-term and long-term penetration rates. These models are trained on data from Changsha but tested on data from Zhengzhou. In addition, a random walk theory is used to explain time lags. It is found that data smoothing enhances accuracy at the potential cost of losing original characteristics. Centring on real-time forecasting of cutterhead torque and thrust force, the thesis underscores their significance in optimising performance and ensuring operational safety. By integrating the historical operational parameters and current setting values, the aware-context RNN model outperforms other RNN models, as reflected in the low-frequency data from the Yinsong water diversion project. Lastly, we accentuate the need for accurate rock mass classification, avoiding inappropriate operation and low safety of excavation. Utilising supervised and semi-supervised learning, our results emphasise the superiority of the RF-based self-training classifier over other RF classifiers, especially when leveraging unlabelled data. Overall, the findings serve as a comprehensive guide for TBM professionals, aiming to streamline and enhance TBM operations.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2024 Feng Shan
dc.rights	au.edu.uts.lib/cph
dc.rights	https://opus.lib.uts.edu.au/bitstream/10453/181183/1/thesis.pdf
dc.title	Optimisation of Tunnel Boring Machine Performance by Machine Learning	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Tunnel boring machines (TBMs) have been widely utilised in tunnel construction due to their efficiency and reliability. However, tunnel collapse, rock bursting, water inrush, and machine jamming remain severe challenges in complex geotechnical conditions. The main objective of this thesis is to investigate if and to what extent TBM operation can be forecasted in real time by machine learning, and if such forecasts can result in TBM tunnelling optimisation, dissecting it into four aspects: (1) penetration rate regression, (2) penetration rate forecasting, (3) cutterhead torque and thrust force forecasting, and (4) rock mass classification. Penetration rate regression is to explore the relationship between penetration rate and other related parameters. Traditional theoretical and empirical methods often provide less accurate results, suited for overall project time management before the start of a project. Machine learning models, using data from the Pahang-Selangor water tunnel, demonstrate better results with support vector machine (SVM), random forest (RF), and artificial neural network (ANN). While other models inclusive of operational and geological parameters can have high accuracy, their applicability is questioned because operational parameters are not accessible during training. Real-time forecasting of penetration rates leverages historical data to predict unknown penetration rates in the future. Accurate forecasts, even for a short distance ahead, can aid in refining TBM operations, resource allocation, decision-making process, and early warning of unexpected geotechnical conditions. We build recurrent neural network (RNN) and long short-term memory (LSTM) models to predict short-term and long-term penetration rates. These models are trained on data from Changsha but tested on data from Zhengzhou. In addition, a random walk theory is used to explain time lags. It is found that data smoothing enhances accuracy at the potential cost of losing original characteristics. Centring on real-time forecasting of cutterhead torque and thrust force, the thesis underscores their significance in optimising performance and ensuring operational safety. By integrating the historical operational parameters and current setting values, the aware-context RNN model outperforms other RNN models, as reflected in the low-frequency data from the Yinsong water diversion project. Lastly, we accentuate the need for accurate rock mass classification, avoiding inappropriate operation and low safety of excavation. Utilising supervised and semi-supervised learning, our results emphasise the superiority of the RF-based self-training classifier over other RF classifiers, especially when leveraging unlabelled data. Overall, the findings serve as a comprehensive guide for TBM professionals, aiming to streamline and enhance TBM operations.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/181183