Commercial projects

Data-driven soft rock interpretation with TBM operational data: Rail link Rishikesh-Karnaprayag (India)

Rail Vikas Nigam Limited (RVNL) is constructing a 125 km broad gauge rail link between Rishikesh and Karnaprayag in the Indian lower Himalayas. One construction lot was excavated with two single shield, hard rock tunnel boring machines (TBM) in challenging geological conditions, including weak rocks and an overburden of up to 859 meters. To optimise the excavation performance, the TBMs are equipped with state-of-the-art instruments to permit early detection of unfavourable ground conditions and thus ensure the timely deployment of countermeasures. Using the cutterhead as a means of ground investigation and novel “void measurement devices”, the TBM operational data is continuously monitored, and advanced parameters such as the specific excavation energy or the torque ratio are used for ground interpretation. This contribution presents the current data analysis pipeline and discusses the benefits and challenges of high-resolution TBM operational data analysis.

Publications:

• Erharter, G., Jain, S., Dammyr, Ø., & Bhasin, R. (2023a). Data driven soft rock interpretation with TBM operational data: Rail link Rishikesh-Karnaprayag (India). INDOROCK 2023: Ninth Indian Rock Conference.
• Erharter, G., Jain, S., Dammyr, Ø., & Bhasin, R. (2023b). KI-basert prediksjon av potensiell squeezing i en TBM-tunnel: Jernbaneforbindelsen Rishikesh-Karnaprayag (India). Fjellsprengningsdag / Bergmekanikkdag / Geoteknikkdag, Oslo, Norway.

Cone Penetration Test predictions for Ten noorden van de Waddeneilanden offshore wind farm

NGI, in collaboration with SAND Geophysics, has developed a workflow for the TNW offshore wind project off the Dutch coast that integrates seismic inversion and geotechnical data with a combination of geostatistical and machine learning techniques. Machine learning was used to capture complex relationships between the CPT and seismic data, enhancing the site characterisation and informing the risk assessment. Senior Geoscientist Guillaume Sauvin presented this methodology at EAGE GET 2022, and DIGEX 2023, detailing the benefits and the challenges associated with implementing data-driven techniques in offshore wind site assessments.

Publications:

• Klinkvort, R. T., Sauvin, G., Dujardin, J., Griffiths, L., Vardy, M. E., & Vanneste, M. (2024). Cone Penetration Testing Prediction Using Seismo-Acoustic Data. 2024(1), 1–5. https://doi.org/10.3997/2214-4609.2024101434
• Sauvin, G., Vardy, M., Klinkvort, R. T., Vanneste, M., Forsberg, C., & Kort, A. (2022). State-of-the-Art Ground Model Development for Offshore Renewables – TNW Case Study (p. 5). https://doi.org/10.3997/2214-4609.202221109

Graph Neural Networks for InSAR Time Series Analysis

Interpreting large-scale InSAR time series data is challenging due to the vast number of data points and the complexity of displacement patterns. Traditional machine learning methods typically rely on either temporal or spatial features, limiting their ability to capture complex ground movement trends. In a study funded by the Norwegian Space Agency, NGI explored the use of Graph Neural Networks (GNNs) to integrate both spatial and temporal relationships for improved displacement trend detection in and around Oslo. Preliminary results suggest that the GNN approach shows promise in enhancing classification accuracy, particularly for subtle but critical movement anomalies. This study highlights the potential of applying GNNs in remote sensing and discusses both the opportunities and challenges in large-scale automated InSAR interpretation.