{"id":{"repo_id":"colo-mines","oai_identifier":"oai:repository.mines.edu:11124/178456"},"canonical_url":"https://search.dev.ndltd.org/etd/colo-mines/oai:repository.mines.edu:11124/178456","repository":{"repo_id":"colo-mines","name":"Colorado School of Mines","base_url":"https://repository.mines.edu/server/oai/request"},"display":{"title":"Predicting adverse geology and tunnel responses utilizing tunnel-ground interaction data with machine learning techniques","abstract":"Uncertainties associated with difficult ground conditions pose major threats to tunnel engineers to ensure safe tunnel construction and maintenance. The existing tunneling practices mostly depend on limited subsoil investigation and monitoring data to deal with such uncertainties. In recent years, highly equipped tunnel boring machines (TBMs) have frequently collected machine-ground interaction data during excavation. The collected data can improve ground characterization from the limited sampling and testing of the site’s soils and rocks. However, these tunneling data need to be optimally utilized to identify their interactions with the ground, which could help deal with various difficult ground conditions for tunnels. Data-driven modeling approach, particularly machine learning (ML), has the potential to identify such interactions hidden inside the complex non-linear tunneling data. To this end, this research explores the development, use, and validation of machine learning for three distinct tunneling-related problems with difficult ground conditions utilizing field data and proposed predictive models developed by machine learning techniques. The first study used TBM data from a water conveyance tunneling project in China to predict upcoming tunnel ground collapses. The ground under TBM excavation was prone to collapse due to the natural occurrences of fault zones and karstic caves, unlocated by prior geological surveys. Three machine learning classifiers, 1) multilayer perceptron (MLP), 2) support vector machine (SVM), and 3) random forest (RF), were trained on TBM data to predict these collapse incidents. The prediction accuracy reached 98% for training data and 96% for validation data. Furthermore, the developed models can identify an “Influence Zone” associated with each collapse incident. The research demonstrated that proposed ML models could predict impending tunnel collapse and the extent of that collapsing ground ahead of the tunnel excavation face. The second study utilized EPBM (earth pressure balance machine) operational data to identify three as-encountered ground properties; 1) clay-sand mixed face conditions, 2) grain size distributions, and 3) TBM clogging potential. The EPBM operational data and geotechnical data collected from Northgate Link extension tunneling project in Seattle were used for this study. Machine learning classifiers and regressors were trained to develop prediction models using ten EPBM operational parameters as input features. The developed models can identify clay-sand mixed face conditions with a balanced accuracy of 87.42%, clogging potentials with a balanced accuracy of 94.3%, and predict the representative grain size distributions with an average R2 of 0.85. Moreover, the feature importance analyses revealed that the average bulkhead soil pressure is an important indicator of mixed soil face condition and grain size distribution, whereas, for clogging potential, it is the average foam flow rate. The third study utilized tunnel in-situ monitoring data from the Shanghai, China, Metro Line 1 to forecast long-term tunnel longitudinal settlements and horizontal convergences. The challenging geological conditions of Shanghai subsoil with unquantifiable human-influenced factors contributed to the sustained long-term tunnel longitudinal settlements and convergences. These factors are mostly intractable in building prediction models. However, the initial measures of settlements capture the inherent uncertainties that can affect future tunnel settlements and convergences. This study used the initial 2.4 years of settlement records to forecast the upcoming tunnel settlements and convergences as later as 14 years apart. Machine learning algorithms (MLP, RF, and k-NN) were trained to build models, where RF gained the highest R2 (0.94-0.98) for predicting long-term settlements, and MLP gained the highest R2 (0.76) for horizontal convergences prediction.","abstract_html":"Uncertainties associated with difficult ground conditions pose major threats to tunnel engineers to ensure safe tunnel construction and maintenance. The existing tunneling practices mostly depend on limited subsoil investigation and monitoring data to deal with such uncertainties. In recent years, highly equipped tunnel boring machines (TBMs) have frequently collected machine-ground interaction data during excavation. The collected data can improve ground characterization from the limited sampling and testing of the site’s soils and rocks. However, these tunneling data need to be optimally utilized to identify their interactions with the ground, which could help deal with various difficult ground conditions for tunnels. Data-driven modeling approach, particularly machine learning (ML), has the potential to identify such interactions hidden inside the complex non-linear tunneling data. To this end, this research explores the development, use, and validation of machine learning for three distinct tunneling-related problems with difficult ground conditions utilizing field data and proposed predictive models developed by machine learning techniques. The first study used TBM data from a water conveyance tunneling project in China to predict upcoming tunnel ground collapses. The ground under TBM excavation was prone to collapse due to the natural occurrences of fault zones and karstic caves, unlocated by prior geological surveys. Three machine learning classifiers, 1) multilayer perceptron (MLP), 2) support vector machine (SVM), and 3) random forest (RF), were trained on TBM data to predict these collapse incidents. The prediction accuracy reached 98% for training data and 96% for validation data. Furthermore, the developed models can identify an “Influence Zone” associated with each collapse incident. The research demonstrated that proposed ML models could predict impending tunnel collapse and the extent of that collapsing ground ahead of the tunnel excavation face. The second study utilized EPBM (earth pressure balance machine) operational data to identify three as-encountered ground properties; 1) clay-sand mixed face conditions, 2) grain size distributions, and 3) TBM clogging potential. The EPBM operational data and geotechnical data collected from Northgate Link extension tunneling project in Seattle were used for this study. Machine learning classifiers and regressors were trained to develop prediction models using ten EPBM operational parameters as input features. The developed models can identify clay-sand mixed face conditions with a balanced accuracy of 87.42%, clogging potentials with a balanced accuracy of 94.3%, and predict the representative grain size distributions with an average R2 of 0.85. Moreover, the feature importance analyses revealed that the average bulkhead soil pressure is an important indicator of mixed soil face condition and grain size distribution, whereas, for clogging potential, it is the average foam flow rate. The third study utilized tunnel in-situ monitoring data from the Shanghai, China, Metro Line 1 to forecast long-term tunnel longitudinal settlements and horizontal convergences. The challenging geological conditions of Shanghai subsoil with unquantifiable human-influenced factors contributed to the sustained long-term tunnel longitudinal settlements and convergences. These factors are mostly intractable in building prediction models. However, the initial measures of settlements capture the inherent uncertainties that can affect future tunnel settlements and convergences. This study used the initial 2.4 years of settlement records to forecast the upcoming tunnel settlements and convergences as later as 14 years apart. Machine learning algorithms (MLP, RF, and k-NN) were trained to build models, where RF gained the highest R2 (0.94-0.98) for predicting long-term settlements, and MLP gained the highest R2 (0.76) for horizontal convergences prediction.","abstract_has_math":false,"creators":["Sarna, Sharmin"],"institution":"Colorado School of Mines. Arthur Lakes Library","degree_name":"Doctor of Philosophy (Ph.D.)","degree_level":"Doctoral","degree_discipline":"Civil and Environmental Engineering","degree_department":null,"school":null,"contributors":[],"advisors":["Gutierrez, Marte S."],"committee_chairs":[],"committee_members":["Zhang, Xiaoli","Pei, Shiling","Tunstall, Lori E."],"year":2023,"date_issued":"2023","date_published":"2023","updated_at":"2026-07-24T01:43:08Z","subjects":["adverse geology","ground characterization","long-term tunnel settlement","machine learning","TBM parameters","tunnel collapse"],"languages":["eng","English"],"rights":["Copyright of the original work is retained by the author."],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["T 9518"],"render_values":[{"text":"T 9518","href":null,"code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/11124/178456","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Gutierrez, Marte S."]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Zhang, Xiaoli","Pei, Shiling","Tunstall, Lori E."]},{"key":"dc:creator","label":"Author","values":["Sarna, Sharmin"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2023-10-23T22:40:38Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2023-10-23T22:40:38Z"]},{"key":"dc:date.issued","label":"Date","values":["2023"]},{"key":"dc:publisher","label":"Institution","values":["Colorado School of Mines. Arthur Lakes Library"]},{"key":"dc:type","label":"Dc Type","values":["Text"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Civil and Environmental Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Doctoral"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy (Ph.D.)"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Colorado School of Mines"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["adverse geology","ground characterization","long-term tunnel settlement","machine learning","TBM parameters","tunnel collapse"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["English"]},{"key":"dc:language.iso","label":"Language (ISO)","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Copyright of the original work is retained by the author."]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["Sarna_mines_0052E_12588.pdf","T 9518"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/11124/178456"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Includes bibliographical references.","2023 Spring."]},{"key":"dc:description.abstract","label":"Abstract","values":["Uncertainties associated with difficult ground conditions pose major threats to tunnel engineers to ensure safe tunnel construction and maintenance. The existing tunneling practices mostly depend on limited subsoil investigation and monitoring data to deal with such uncertainties. In recent years, highly equipped tunnel boring machines (TBMs) have frequently collected machine-ground interaction data during excavation. The collected data can improve ground characterization from the limited sampling and testing of the site’s soils and rocks. However, these tunneling data need to be optimally utilized to identify their interactions with the ground, which could help deal with various difficult ground conditions for tunnels. Data-driven modeling approach, particularly machine learning (ML), has the potential to identify such interactions hidden inside the complex non-linear tunneling data. To this end, this research explores the development, use, and validation of machine learning for three distinct tunneling-related problems with difficult ground conditions utilizing field data and proposed predictive models developed by machine learning techniques. The first study used TBM data from a water conveyance tunneling project in China to predict upcoming tunnel ground collapses. The ground under TBM excavation was prone to collapse due to the natural occurrences of fault zones and karstic caves, unlocated by prior geological surveys. Three machine learning classifiers, 1) multilayer perceptron (MLP), 2) support vector machine (SVM), and 3) random forest (RF), were trained on TBM data to predict these collapse incidents. The prediction accuracy reached 98% for training data and 96% for validation data. Furthermore, the developed models can identify an “Influence Zone” associated with each collapse incident. The research demonstrated that proposed ML models could predict impending tunnel collapse and the extent of that collapsing ground ahead of the tunnel excavation face. The second study utilized EPBM (earth pressure balance machine) operational data to identify three as-encountered ground properties; 1) clay-sand mixed face conditions, 2) grain size distributions, and 3) TBM clogging potential. The EPBM operational data and geotechnical data collected from Northgate Link extension tunneling project in Seattle were used for this study. Machine learning classifiers and regressors were trained to develop prediction models using ten EPBM operational parameters as input features. The developed models can identify clay-sand mixed face conditions with a balanced accuracy of 87.42%, clogging potentials with a balanced accuracy of 94.3%, and predict the representative grain size distributions with an average R2 of 0.85. Moreover, the feature importance analyses revealed that the average bulkhead soil pressure is an important indicator of mixed soil face condition and grain size distribution, whereas, for clogging potential, it is the average foam flow rate. The third study utilized tunnel in-situ monitoring data from the Shanghai, China, Metro Line 1 to forecast long-term tunnel longitudinal settlements and horizontal convergences. The challenging geological conditions of Shanghai subsoil with unquantifiable human-influenced factors contributed to the sustained long-term tunnel longitudinal settlements and convergences. These factors are mostly intractable in building prediction models. However, the initial measures of settlements capture the inherent uncertainties that can affect future tunnel settlements and convergences. This study used the initial 2.4 years of settlement records to forecast the upcoming tunnel settlements and convergences as later as 14 years apart. Machine learning algorithms (MLP, RF, and k-NN) were trained to build models, where RF gained the highest R2 (0.94-0.98) for predicting long-term settlements, and MLP gained the highest R2 (0.76) for horizontal convergences prediction."]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["born digital","doctoral dissertations"]},{"key":"dc:title","label":"Title","values":["Predicting adverse geology and tunnel responses utilizing tunnel-ground interaction data with machine learning techniques"]}]}],"canonical_facts":{"dc:contributor.advisor":["Gutierrez, Marte S."],"dc:contributor.committeemember":["Zhang, Xiaoli","Pei, Shiling","Tunstall, Lori E."],"dc:creator":["Sarna, Sharmin"],"dc:date.accessioned":["2023-10-23T22:40:38Z"],"dc:date.available":["2023-10-23T22:40:38Z"],"dc:date.issued":["2023"],"dc:description":["Includes bibliographical references.","2023 Spring."],"dc:description.abstract":["Uncertainties associated with difficult ground conditions pose major threats to tunnel engineers to ensure safe tunnel construction and maintenance. The existing tunneling practices mostly depend on limited subsoil investigation and monitoring data to deal with such uncertainties. In recent years, highly equipped tunnel boring machines (TBMs) have frequently collected machine-ground interaction data during excavation. The collected data can improve ground characterization from the limited sampling and testing of the site’s soils and rocks. However, these tunneling data need to be optimally utilized to identify their interactions with the ground, which could help deal with various difficult ground conditions for tunnels. Data-driven modeling approach, particularly machine learning (ML), has the potential to identify such interactions hidden inside the complex non-linear tunneling data. To this end, this research explores the development, use, and validation of machine learning for three distinct tunneling-related problems with difficult ground conditions utilizing field data and proposed predictive models developed by machine learning techniques. The first study used TBM data from a water conveyance tunneling project in China to predict upcoming tunnel ground collapses. The ground under TBM excavation was prone to collapse due to the natural occurrences of fault zones and karstic caves, unlocated by prior geological surveys. Three machine learning classifiers, 1) multilayer perceptron (MLP), 2) support vector machine (SVM), and 3) random forest (RF), were trained on TBM data to predict these collapse incidents. The prediction accuracy reached 98% for training data and 96% for validation data. Furthermore, the developed models can identify an “Influence Zone” associated with each collapse incident. The research demonstrated that proposed ML models could predict impending tunnel collapse and the extent of that collapsing ground ahead of the tunnel excavation face. The second study utilized EPBM (earth pressure balance machine) operational data to identify three as-encountered ground properties; 1) clay-sand mixed face conditions, 2) grain size distributions, and 3) TBM clogging potential. The EPBM operational data and geotechnical data collected from Northgate Link extension tunneling project in Seattle were used for this study. Machine learning classifiers and regressors were trained to develop prediction models using ten EPBM operational parameters as input features. The developed models can identify clay-sand mixed face conditions with a balanced accuracy of 87.42%, clogging potentials with a balanced accuracy of 94.3%, and predict the representative grain size distributions with an average R2 of 0.85. Moreover, the feature importance analyses revealed that the average bulkhead soil pressure is an important indicator of mixed soil face condition and grain size distribution, whereas, for clogging potential, it is the average foam flow rate. The third study utilized tunnel in-situ monitoring data from the Shanghai, China, Metro Line 1 to forecast long-term tunnel longitudinal settlements and horizontal convergences. The challenging geological conditions of Shanghai subsoil with unquantifiable human-influenced factors contributed to the sustained long-term tunnel longitudinal settlements and convergences. These factors are mostly intractable in building prediction models. However, the initial measures of settlements capture the inherent uncertainties that can affect future tunnel settlements and convergences. This study used the initial 2.4 years of settlement records to forecast the upcoming tunnel settlements and convergences as later as 14 years apart. Machine learning algorithms (MLP, RF, and k-NN) were trained to build models, where RF gained the highest R2 (0.94-0.98) for predicting long-term settlements, and MLP gained the highest R2 (0.76) for horizontal convergences prediction."],"dc:format.medium":["born digital","doctoral dissertations"],"dc:identifier":["Sarna_mines_0052E_12588.pdf","T 9518"],"dc:identifier.uri":["https://hdl.handle.net/11124/178456"],"dc:language":["English"],"dc:language.iso":["eng"],"dc:publisher":["Colorado School of Mines. Arthur Lakes Library"],"dc:rights":["Copyright of the original work is retained by the author."],"dc:subject":["adverse geology","ground characterization","long-term tunnel settlement","machine learning","TBM parameters","tunnel collapse"],"dc:title":["Predicting adverse geology and tunnel responses utilizing tunnel-ground interaction data with machine learning techniques"],"dc:type":["Text"],"thesis:degree_discipline":["Civil and Environmental Engineering"],"thesis:degree_level":["Doctoral"],"thesis:degree_name":["Doctor of Philosophy (Ph.D.)"],"thesis:institution_name":["Colorado School of Mines"]},"updated_at":"2026-07-24T01:43:08Z"}