Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Shield Moving Trajectory Prediction and Anomaly Detection during Tunnelling: A Deep Learning Algorithm Framework
Given that misalignment, while tunnelling in urban areas, can cause ground heaving or unfavourable soil deformations, it starts drawing great attention from scientists and engineers. In this study, the potential of applying deep learning (DL) and machine learning (ML) algorithms to earth pressure balance-tunnel boring machine (EPB-TBM) posture prediction and anomaly detection was explored. A predictive framework for the posture to a tunnelling project in loess region was proposed, including three phases: the exploratory data analysis to reduce dimensionality and highlight characteristics, the evaluation of DL-based models in posture prediction, and anomaly detection in posture deviations. The principal component analysis was used to decompose a tunnelling data set derived from Xi’an Metro Line 4, eliminating outliers and isolating its salient data features. The Pearson correlation coefficient was used to reduce irrelevant data features, while the partial dependence analysis aimed to marginalise other features and established the relationship between an individual character and the predicted target. After these EDA approaches, the feature-based subseries was normalised for training and examining the DL-based long short-term memory (LSTM) EPB-TBM posture predictors. Furthermore, several unsupervised anomaly detection methods were used to detect the posture prediction outliers. The validity of the proposed posture predictors was verified by comparing to the field measurements, and the results showed that the LSTM predictor performs well due to its complex nonlinear unit that constitutes larger deep neural networks to handle nonlinear problems. In addition, the deviation mechanism was revealed correspond to the relationship established by the exploratory data analysis and the geological conditions. These results highlight an exciting potential for the use of DL-based models to secure the sustainable development of surrounding environments from disturbing by tunnelling-induced misalignment.
Shield Moving Trajectory Prediction and Anomaly Detection during Tunnelling: A Deep Learning Algorithm Framework
Given that misalignment, while tunnelling in urban areas, can cause ground heaving or unfavourable soil deformations, it starts drawing great attention from scientists and engineers. In this study, the potential of applying deep learning (DL) and machine learning (ML) algorithms to earth pressure balance-tunnel boring machine (EPB-TBM) posture prediction and anomaly detection was explored. A predictive framework for the posture to a tunnelling project in loess region was proposed, including three phases: the exploratory data analysis to reduce dimensionality and highlight characteristics, the evaluation of DL-based models in posture prediction, and anomaly detection in posture deviations. The principal component analysis was used to decompose a tunnelling data set derived from Xi’an Metro Line 4, eliminating outliers and isolating its salient data features. The Pearson correlation coefficient was used to reduce irrelevant data features, while the partial dependence analysis aimed to marginalise other features and established the relationship between an individual character and the predicted target. After these EDA approaches, the feature-based subseries was normalised for training and examining the DL-based long short-term memory (LSTM) EPB-TBM posture predictors. Furthermore, several unsupervised anomaly detection methods were used to detect the posture prediction outliers. The validity of the proposed posture predictors was verified by comparing to the field measurements, and the results showed that the LSTM predictor performs well due to its complex nonlinear unit that constitutes larger deep neural networks to handle nonlinear problems. In addition, the deviation mechanism was revealed correspond to the relationship established by the exploratory data analysis and the geological conditions. These results highlight an exciting potential for the use of DL-based models to secure the sustainable development of surrounding environments from disturbing by tunnelling-induced misalignment.
Shield Moving Trajectory Prediction and Anomaly Detection during Tunnelling: A Deep Learning Algorithm Framework
Bai, Xue-Dong (Autor:in) / Cheng, Wen-Chieh (Autor:in)
Geo-Congress 2023 ; 2023 ; Los Angeles, California
Geo-Congress 2023 ; 287-297
23.03.2023
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
| British Library Conference Proceedings | 2023
Real-time prediction of shield moving trajectory during tunnelling
| Springer Verlag | 2022
Real-time prediction of shield moving trajectory during tunnelling using GRU deep neural network
| Springer Verlag | 2022