Challenges and Opportunities of Data-Driven Advance Classification for Hard Rock TBM excavations

Abstract Excavation with tunnel boring machines (TBMs) is a widely used method of tunneling in all ground types including soil and rock today. The paper addresses the shift from traditional subjective methods to data-driven approaches for advance classification of TBMs in hard rock tunnel excavation. By leveraging continuous TBM operational data, these methodologies offer more objective, transparent, continuous, and reproducible assessments of excavation conditions. The challenges include the need for sophisticated computational tools to interpret complex interactions between rock mass, TBM machinery, and logistics that are sensitive to the whole data processing pipeline. This contribution provides consistent, step-by-step recommendations for how to efficiently process TBM operational data. It furthermore provides the community with three open TBM operational datasets that can be used for benchmarks and educational purposes related to TBM data processing. To overcome data confidentiality issues, the datasets are synthetic and were generated with generative adversarial networks (GANs)—a method of artificial intelligence—that are trained on real TBM operational data. It is, thus, ensured that the data, on the one hand, looks like real data, but has no direct relationship to real construction sites. This study highlights the potential of data-driven techniques to improve TBM tunneling efficiency, while addressing key technical challenges.