Knowledge-data dual-driven collapse risk decision-making for tunnel boring machine construction

Tunnel construction, particularly hard rock tunnelling using TBM, involves complex geological conditions. Traditional risk assessment methods primarily rely on engineers’ experience, which is often affected by inaccurate geological surveys, insufficient site investigations, and subjective expert knowledge, potentially leading to insufficient risk control and accidents. Moreover, purely data-driven approaches often fail to accurately assess risk severity or support effective risk decision-making. This paper proposes an innovative knowledge–data dual-driven framework that integrates machine learning with expert systems and cloud modelling. Machine learning algorithms are employed to accurately predict the likelihood of TBM collapse, while expert systems combined with cloud model theory are used to quantify accident severity. In the data-driven component, among the three algorithms tested, the Random Forest (RF) model achieved the best performance with a prediction accuracy of 98%. In the knowledge-driven component, the expert system together with cloud models quantifies the uncertainty in expert knowledge. Finally, a fuzzy-cloud probabilistic model is proposed to couple the results of data-driven and knowledge-driven approaches, quantify the probabilities of risk levels, and enable intelligent risk decision-making. The framework is supported effectively through four specific scenarios from a real-world tunnel project, and the results demonstrate its feasibility and effectiveness in supporting TBM collapse risk decision-making.