Shield Machine Tool Life Analytical Modeling And Research

Subway construction is an important part of urban three-dimensional traffic.However,because the high cost of building subways,the development of subways is hindered,shield machine blades account for a large part of the cost of consumables,effectivly using of blade savings is an effective way control over construction costs.There are mainly two ways to save money by optimizing tools: firstly,predicting the time of blade damage,so we can arrange scheduling in advance and save the construction period;secondly,predicting the cause of damage in order to reasonably control the machine and avoid non-wear damage.For the two kinds of forecasting situations,they have the characteristics of complex time series and nonlinearity,which leads to the traditional fitting method in the problems to falure.In this paper,we build a series of optimization models based on the actual mining data of the Subway Line 2 in Qingdao and optimize tooling from different angles.Starting from the theoretical direction,a relation model between tool wear and dynamic influencing factors of a shield machine based on an improved CSM model was established,and the accuracy rate was about 85%.The accuracy of the theoretical model is not high because the factors that affect the damage have static factors and dynamic factors.Once the tool is manufactured,static factors such as shape and position cann't be changed,so a more convenient application of the model need to be established and the accuracy should be improved.Therefoce,we start from the data-driven direction,basing on the actual measurement data in the project,selecting a variety of intelligent methods to try,finally,finally multiple forecasting models is established.When aiming at predicting time of tool failure,we consider normal damage and abnormal damage as a unified whole,the wear quantity and life are selected respectively as dependent variables,and the ant colony-neural network algorithm is used to predict the number 1-41 tool.Principal component analysis and gray correlation are applied firstly so that factors under different targets are selected.In order to raise the efficiency of large-scale ant colony iterations,the first part is we use the ant colony to do a preliminary screening of weights and thresholds,the second part is select the final weight,threshold and hidden layer nodes through feedback fine-tuning selection.After comparing the two models,we found accuracy of the ant colony-neural network prediction model based on the target of life is 90.23%.In order to improve the accuracy of predicting damage time and adapt to the characteristics of small samples and complex linearity,we study the use of support vector machine regression algorithm.Based on the advantages of SVM,such as small dimension and transduction reasoning,the information matrix is composed of seven driving factors such as daily driving mileage,rock grade,numbers of tool failuer,cutter head thrust,cutter head rotation speed,and tunneling speed.Meanwhile,the k-cross check and grid search are used to find the optimal parameters,and the optimal kernel function is finally determined as the Gaussian kernel function,the accuracy of the final data inspection lifetime reached 94.52%.In order to optimize the maneuvering strategy and effectively fit the relationship between various influencing factors and abnormal damage patterns which can provide theoretical guidance for operations in different environments.We choose a variety of research methods,such as BP neural network and recurrent nerves as a comparative study,due to the recurrent neural network can learn and train based on the previous geological conditions and operating conditions based on the data of the input layer within a very small time interval,and finally realizes the identification of the causes of the damage,so a complex mapping relationship is established to meet the timing characteristics.The final prediction result is that the accuracy of the prediction model based on the recurrent neural network is about 90%,which show the operation is feasible.