Health Assessment And Prediction Of Cutter Head Of Tunnel Boring Machine For High-dimensional Spatiotemporal Data Analysis

With the continuous development and utilization of urban underground space in my country,tunnel boring machine has become the key equipment for underground tunnel construction due to its advantages of safety,efficiency and economy.As one of the core components of the tunnel boring machine,the health status of the cutter head of the tunnel boring machine not only affects the construction efficiency,but also affects the safety during the construction process.Therefore,this thesis takes the cutter head of tunnel boring machine as the research object for high-dimensional spatiotemporal data,and focuses on the research on the health assessment and trend prediction of the cutter head.Firstly,the health assessment study of the cutter head of the tunnel boring machine based on unsupervised learning is carried out.Aiming at the characteristics that the health data of the cutter head of the tunnel boring machine is easy to obtain,but the amount of failure data is small,and the health state can not be directly determined,the LSTM-ED model is established based on unsupervised learning.The health data is used as the input of the model,and the model can learn the internal law of the health data,so that the trained model can quantify the difference between the data to be tested and the health data,and achieve the purpose of the health assessment of the cutter head of tunnel boring machine,and set up multiple sets of comparative experiments to verify the superiority of the model.Secondly,the health assessment study of the cutter head of the tunnel boring machine based on supervised learning is carried out.A complete degradation data of the cutter head is selected to calculate the excavation time,i.e.useful life,of the cutter head.Based on the useful life,the health index of each time point of the cutter head is artificially labelled as a sample label for supervised learning.Comprehensively using the advantages of CNN network and GRU network,the CNN-GRU model is established for health assessment of the cutter head,and a simple CNN model and a simple GRU model are built as comparative experiments,and the results show that the CNN-GRU model has the best effect.Finally,the research on the health trend prediction of the cutter head of the tunnel boring machine is carried out.Taking cutter head health trend prediction as the research goal,singlestep prediction and multi-step prediction research are carried out respectively.The features related to the cutter head health state are selected as covariates,and different models are established for trend prediction with HI curve as the target prediction sequence,and the prediction results of each model are compared and analyzed,and the results showed that the LSTM model has the best effect for single-step prediction and the Seq2 Seq model with attention mechanism for multi-step prediction.The research methods and ideas of the thesis can provide a certain reference for the detection and maintenance of cutting tools in the construction process of tunnel boring machines,and have strong engineering significance.