Georg H. Erharter , Paul Johannes Unterlaß, Nedim Radončić, Thomas Marcher, Jamal Rostami

Tunnel Boring Machines (TBMs) have revolutionized tunneling industry and are currently the dominant method of tunneling in all ground types including soil and rock. Traditional approaches to TBM advance classification, however, rely heavily on subjective assessments by onsite personnel, which are often hampered by limited access to the excavation face and discontinuous observation intervals. This paper elaborates on the shift towards data-driven methodologies for TBM advance classification, which leverage continuously recorded TBM operational data to provide a more objective, transparent, and reproducible assessment of excavation conditions. We discuss the challenges associated with TBM data analysis, including the need for computational tools capable of disentangling complex influences such as rock mass conditions, TBM machinery, and operational logistics. To support the advancement of this field, we provide synthetic TBM datasets generated by generative adversarial networks (GANs), which can be utilized to circumvent issues related to data confidentiality and Python tools to facilitate adoption by practitioners. GAN based data sets used in this analysis have been developed based on the actual field measurements made on TBMs and hence are authentic and reliable for such applications. We also provide practical recommendations for implementing data-driven TBM advance classification, using the torque ratio as a reliable representative of machine penetration rate. This work underscores the potential of data-driven approaches to enhance TBM tunneling efficiency while addressing the technical challenges that accompany their implementation.