BBT, Lot Mules 2–3. Application of machine learning on TBM parameters for risk prediction tools

The risk of TBM jamming because of squeezing rock mass or excessive caving (instability of the cavity) conditions is always a focal point while boring long and deep tunnels with shielded TBMs. Mechanized excavation provides a large amount of real-time data, for each machine’s parameters. Several analyses have been executed to assess the correlations between different sets of TBM parameters for the exploratory tunnel. Instead of analysing the full dataset, a different approach is proposed, with data processed sequentially to simulate the real excavation situation. The probability distributions are recalculated together with their statistical parameters while advancing. This approach is extended also to the training of a Machine Learning model able to foresee the behavior at future rings based on past rings (using a Recurrent Neural Network). The predictions show a promising first step toward creating practical tools that can assist contractors and designers to correlate and predict TBM excavation data. These tools will assist to predict and quantify high risk situations that could arise during excavation.