Research On Tool Wear Prediction Of Shield Tunneling Based On Support Vector Machine

Due to the advantages of high efficiency of shield construction,small impact on existing traffic,safety and reliability,the shield method has developed rapidly and has accumulated rich construction experience.Shield cutter is the key hardware that affects the efficiency of shield construction,and needs to be replaced according to factors such as driving distance.Replacement of the shield cutter untimely will cause excessive wear,which will seriously affect the construction efficiency.The tunnellingling parameters can reflect the tunnellinging situation of the shield machine,and the tunnellinging parameters can be used to detect the wear state of the shield cutter.Support Vector Machine(SVM)is a machine learning algorithm that has developed rapidly in the past ten years.It has achieved good application results in signal processing,portrait recognition,text recognition and other fields.In recent years,in the field of shield construction,such as the prediction of ground settlement,SVM has also shown good performance and advantages.At present,the application of SVM in the prediction of shield cutter wear is relatively few.Therefore,this paper will establish a shield cutter wear prediction model based on SVM.It provides a new detection method for shield construction in sandy cobble stratum in Chengdu,and provides a reference for shield construction in similar stratum in the future.This paper is based on the tunnellinging data recorded by the shield construction information real-time monitoring system of Chengdu Metro Line 7.Analyzed the components of total thrust and cutter head torque.According to the calculation model of the tunnellinging parameters based on the engineering application and the optimized CSM model of the Colorado School of Mines,a new formula about the cutter head torque,total thrust,penetration,and earth pressure is derived.It reflects the relationship between shield cutter wear and changes of tunnellinging parameters.Analyze the state of shield construction in the normal construction section,the results show that the tunnellinging parameters can reflect the changes in shield construction.The selection criteria of effective tunnellinging data are proposed.Using Wavelet Transform method to reduce the noise of the tunnellinging data,comparative analysis of the 2-dimensional and 3-dimensional feature distribution of the data before and after cutter change,to complete the judgment of the critical wear state of the shield cutter.The two parameters of thrust displacement and thrust pressure were added.Based on the six feature parameters,the kfold cross-validation algorithm and grid-search algorithm are used to optimize the parameters of SVM.Three hyperparameter combinations were selected by the heat map.When C is taken as 1,10,100 and gamma is taken as 100 respectively,the training accuracy of the model is0.8850,0.8765 and 0.8750,and the prediction accuracy of the validation set is 0.9236,0.9120 and 0.9051.The Recall Rate and F1-Measure are used to evaluate the performance of the model.The Recall Rates were 0.9397,0.9347 and 0.9246,and the F1-Measure were 0.9189,0.9073 and 0.8998.The results show that the model has good predictive classification performance.Applying the model to actual projects,the model outputs the prediction results and gives a guidance suggestion on whether to change the shield cutter.