Transfer Learning Based Tunnel Boring Machine Advance Classification

Tunnel Boring Machines (TBMs) are well established in modern tunnel construction, with monitoring and predicting TBM performance being crucial for project timelines and risk mitigation. Machine Learning (ML), offers promising avenues for analysing TBM operational data. Transfer learning, a technique of ML, enhances classification tasks by leveraging knowledge from one domain to another, making it particularly relevant for TBM tunnelling
where labelled data may be limited. This paper proposes a novel approach to TBM advance
classification using a recurrent neural network (RNN) trained via transfer learning. By training
the model on one dataset and transferring it to another site, the study aims to classify TBM
advance as regular- or exceptional advance based on TBM operational data. Transfer learning
offers advantages for classification tasks in tunnelling by reducing the need for extensive labelled data collection at each site. Results showcase the RNN's performance on datasets from the Norwegian Ulriken and Follobanen tunnels. The experiment highlights the model's effectiveness in classifying regular advance, while classification of irregular advance shows some room for improvement. Overall, this research contributes to advancing ML applications in TBM tunnelling, and underscores transfer learning's potential to streamline performance analysis and decision-making processes in tunnelling projects.