Online multi-objective optimization for real-time TBM attitude control with spatio-temporal deep learning model

This paper developed a novel online optimization deep learning framework for accurate estimation and control of tunnel boring machine's (TBM) attitude and position. The proposed spatio-temporal deep learning model integrates graph convolutional network (GCN) and long short-term memorial (LSTM) neural network which can provide an accurate estimation of TBM's trajectory deviations. This model is later utilized as the meta-model and integrated with an online non-dominated sorting genetic algorithm II (NSGA-II) method to optimize and minimize the trajectory deviations in real-time. A tunnel project from Singapore is utilized as a case study for demonstration purposes. TBM's key operational parameters, articulation displacements, and TBM's trajectory deviations are processed and utilized for real-time prediction. The results indicate that (1) The developed deep learning models can predict TBM's trajectory deviations with high accuracy. (2) TBM's trajectory deviations can be effectively reduced by the proposed optimization method. (3) The proposed spatio-temporal deep learning methods outperform the state-of-the-art methods. (4) The proposed online optimization method is more practical and creates better overall improvement than the conventional multi-objective optimization (MOO) approach. The novelty of this paper includes (a) developing a hybrid framework for accurate trajectory deviation estimation and TBM operating optimization, and (b) investigating the influence of time and distance lag between the TBM operation and position control by considering an online optimization.