Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data

Research output: Chapter in Book/Report/Conference proceedingConference contributionResearchpeer-review

Abstract

The tunnel boring machine (TBM) which is currently excavating the exploratory tunnel Ahrental-Pfons of the Brenner Base Tunnel records parameters like cutter head torque or advance pressure on a ten second interval. TBM data like this and derived indicators (e.g.: specific penetration, torque ratio…) are often used as additional help for assessing the response of the rockmass towards the excavation. The goal of this paper is to explore the applicability of a special type of artificial neural network (ANN) for an automatic online classification of the rockmass behavior solely based on TBM data. An ensemble of Long Short Term Memory (LSTM) networks with additional one-dimensional convolutional layers on top, is used to classify individual features of TBM data in mini-batches. The 1D convolutional input layers enhance the ANN’s ability to extract significant features of the data. After an experimental phase, the best performance was achieved with an ensemble of eight convolutional LSTM – networks, where four networks each were deployed on the features torque - ratio and torque. Although the final categorical classification of the ensemble only achieved an overall accuracy of 74.4%, the probabilistic, relative output still yields valuable information about the rockmass behavior and could be used to aid geotechnicians in a real-world scenario.
Original languageEnglish
Title of host publicationInformation Technology in Geo-Engineering
Subtitle of host publicationProceedings of the 3rd International Conference (ICITG), Guimarães, Portugal
PublisherSpringer
DOIs
Publication statusPublished - 2019

Publication series

NameGeomechanics and Geoengineering
PublisherSpringer

Fingerprint

Tunnels
Neural networks
Torque
Excavation
Long short-term memory

Keywords

  • Artificial intelligence
  • Artificial neural networks
  • TBM-Data
  • Online classification

Cite this

Erharter, G. H., Marcher, T., & Reinhold, C. (2019). Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data. In Information Technology in Geo-Engineering: Proceedings of the 3rd International Conference (ICITG), Guimarães, Portugal (Geomechanics and Geoengineering). Springer. https://doi.org/10.1007/978-3-030-32029-4_16

Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data. / Erharter, Georg Hermann; Marcher, Thomas; Reinhold, Chris.

Information Technology in Geo-Engineering: Proceedings of the 3rd International Conference (ICITG), Guimarães, Portugal. Springer, 2019. (Geomechanics and Geoengineering).

Research output: Chapter in Book/Report/Conference proceedingConference contributionResearchpeer-review

Erharter, GH, Marcher, T & Reinhold, C 2019, Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data. in Information Technology in Geo-Engineering: Proceedings of the 3rd International Conference (ICITG), Guimarães, Portugal. Geomechanics and Geoengineering, Springer. https://doi.org/10.1007/978-3-030-32029-4_16
Erharter GH, Marcher T, Reinhold C. Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data. In Information Technology in Geo-Engineering: Proceedings of the 3rd International Conference (ICITG), Guimarães, Portugal. Springer. 2019. (Geomechanics and Geoengineering). https://doi.org/10.1007/978-3-030-32029-4_16
Erharter, Georg Hermann ; Marcher, Thomas ; Reinhold, Chris. / Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data. Information Technology in Geo-Engineering: Proceedings of the 3rd International Conference (ICITG), Guimarães, Portugal. Springer, 2019. (Geomechanics and Geoengineering).
@inproceedings{9a1db8449fa945ecb002f82dcd20820b,
title = "Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data",
abstract = "The tunnel boring machine (TBM) which is currently excavating the exploratory tunnel Ahrental-Pfons of the Brenner Base Tunnel records parameters like cutter head torque or advance pressure on a ten second interval. TBM data like this and derived indicators (e.g.: specific penetration, torque ratio…) are often used as additional help for assessing the response of the rockmass towards the excavation. The goal of this paper is to explore the applicability of a special type of artificial neural network (ANN) for an automatic online classification of the rockmass behavior solely based on TBM data. An ensemble of Long Short Term Memory (LSTM) networks with additional one-dimensional convolutional layers on top, is used to classify individual features of TBM data in mini-batches. The 1D convolutional input layers enhance the ANN’s ability to extract significant features of the data. After an experimental phase, the best performance was achieved with an ensemble of eight convolutional LSTM – networks, where four networks each were deployed on the features torque - ratio and torque. Although the final categorical classification of the ensemble only achieved an overall accuracy of 74.4{\%}, the probabilistic, relative output still yields valuable information about the rockmass behavior and could be used to aid geotechnicians in a real-world scenario.",
keywords = "Artificial intelligence, Artificial neural networks, TBM-Data, Online classification",
author = "Erharter, {Georg Hermann} and Thomas Marcher and Chris Reinhold",
year = "2019",
doi = "10.1007/978-3-030-32029-4_16",
language = "English",
series = "Geomechanics and Geoengineering",
publisher = "Springer",
booktitle = "Information Technology in Geo-Engineering",

}

TY - GEN

T1 - Artificial Neural Network Based Online Rockmass Behavior Classification of TBM Data

AU - Erharter, Georg Hermann

AU - Marcher, Thomas

AU - Reinhold, Chris

PY - 2019

Y1 - 2019

N2 - The tunnel boring machine (TBM) which is currently excavating the exploratory tunnel Ahrental-Pfons of the Brenner Base Tunnel records parameters like cutter head torque or advance pressure on a ten second interval. TBM data like this and derived indicators (e.g.: specific penetration, torque ratio…) are often used as additional help for assessing the response of the rockmass towards the excavation. The goal of this paper is to explore the applicability of a special type of artificial neural network (ANN) for an automatic online classification of the rockmass behavior solely based on TBM data. An ensemble of Long Short Term Memory (LSTM) networks with additional one-dimensional convolutional layers on top, is used to classify individual features of TBM data in mini-batches. The 1D convolutional input layers enhance the ANN’s ability to extract significant features of the data. After an experimental phase, the best performance was achieved with an ensemble of eight convolutional LSTM – networks, where four networks each were deployed on the features torque - ratio and torque. Although the final categorical classification of the ensemble only achieved an overall accuracy of 74.4%, the probabilistic, relative output still yields valuable information about the rockmass behavior and could be used to aid geotechnicians in a real-world scenario.

AB - The tunnel boring machine (TBM) which is currently excavating the exploratory tunnel Ahrental-Pfons of the Brenner Base Tunnel records parameters like cutter head torque or advance pressure on a ten second interval. TBM data like this and derived indicators (e.g.: specific penetration, torque ratio…) are often used as additional help for assessing the response of the rockmass towards the excavation. The goal of this paper is to explore the applicability of a special type of artificial neural network (ANN) for an automatic online classification of the rockmass behavior solely based on TBM data. An ensemble of Long Short Term Memory (LSTM) networks with additional one-dimensional convolutional layers on top, is used to classify individual features of TBM data in mini-batches. The 1D convolutional input layers enhance the ANN’s ability to extract significant features of the data. After an experimental phase, the best performance was achieved with an ensemble of eight convolutional LSTM – networks, where four networks each were deployed on the features torque - ratio and torque. Although the final categorical classification of the ensemble only achieved an overall accuracy of 74.4%, the probabilistic, relative output still yields valuable information about the rockmass behavior and could be used to aid geotechnicians in a real-world scenario.

KW - Artificial intelligence

KW - Artificial neural networks

KW - TBM-Data

KW - Online classification

U2 - 10.1007/978-3-030-32029-4_16

DO - 10.1007/978-3-030-32029-4_16

M3 - Conference contribution

T3 - Geomechanics and Geoengineering

BT - Information Technology in Geo-Engineering

PB - Springer

ER -