Explanations of Performance Differences in Segment Lining for Tunnel Boring Machines.

The tunnel lining process with segments is a labour-intensive task, for which the expertise of Tunnel Boring Machines' operators is crucial. For this task, human expertise can be evaluated based on the average time of building a tunnel ring. Data-driven identification of the different levels of operators' expertise can help to understand the causes of possible discrepancies. Consequently, bridging possibly existing gaps in expertise can be achieved through more training offered to less experienced operators or through support from user-assistance systems. In order to make the expertise more tangible, we trained deep learning models to classify expertise profiles of erector operators based on time series data accrued during the process. Afterwards, we investigate these with explainable artificial intelligence techniques to identify features with the highest influence on the performance prediction and derive regions of interest in ring-building sequences leading to specific performance classifications. Finally, we discuss how the observations from our study can contribute to designing assistance systems that support operators toward a more efficient ring-building process.