Research On Front Disc Cutter Wear Data Of Shield Machine In Hard Rock Stratum Based On Algorithm Combination Optimization BP Neural Network

The wear rate of the shield disc cutter under hard rock stratum is fast,and it is necessary to shut down and open the cabin frequently to check and replace the disc cutter.At present,the position to open cabin mainly depends on the engineering experience to make human judgments.At the same time,there are many types of disc cutter wear affecting factors,and the existing theoretical models and empirical formulas for disc cutter wear prediction at home and abroad do not have good tolerance and universality.Therefore,it is imperative to find a popularized method to effectively predict disc cutter wear,in order to achieve the effect of reducing project risks,speeding up construction progress,and reducing project costs.In response to the above problems,this paper uses the BP neural network based on the Levenberg-Marquardt(LM)algorithm as the core data analysis tool to establish a prediction model for the wear of the front disc cutter and uses the Sequential Model-Based Global Optimization(SMBO)algorithm and Genetic Algorithm(GA)to optimize the hyperparameters of the neural network.Use a combination of algorithm to improve the prediction accuracy of the model,and finally carry out the sensitivity analysis of the parameters to verify the correctness and rationality of the input parameter selection.The main research work and results of this paper are as follows:(1)Select 14 factors"earth pressure,thrust of cutterhead,torque of cutterhead,tunneling speed,penetration rate,shield depth,cutter size,installation radius,old or new knives,HRC,blade width,blade yield strength,UCS and RQD"as input parameters,divide the shield interval as a whole to divide the data set,and establish a BP neural network based on the LM algorithm on MATLAB and Python to predict the amount of front disc cutter wear.The three parameters of mean square error MSE,coefficient of determination R~2 and prediction error are selected to evaluate the network model training effect,and Python with better open source is selected as the building platform of the neural network model after comparing the prediction results of the two models.(2)Using Python to build a BP neural network,after summarizing and processing nearly three million original data,conduct research on the wear data of multiple disc cutters in a single shield section and the wear data of multiple shield sections of a single disc cutter,and the results show that the BP neural network has high accuracy in predicting the wear amount in the single shield section with multiple disc cutters and single disc cutter with multiple shield sections,and the following conclusions are obtained:The larger the disc cutter installation radius,the more obvious the influence of geological conditions;when the input parameters change rapidly,the lack of a enough data set to train the neural network model will reduce the prediction accuracy of the model;the focus of actual engineering should be the cumulative wear of multiple tunneling rings,and when evaluating the prediction effect of the neural network model,it is necessary to start from multiple perspectives such as theory and practice.(3)The wear data of multi-disc cutters with the multi-shield sections is studied,and the hyperparameters of the BP neural network are optimized by using the SMBO Algorithm and the Genetic Algorithm.The results show that the SMBO Algorithm has a better optimization effect.The comparison of multiple data set classification methods shows that the network model of multi-disc cutters with the multi-shield sections has better generalization and practicability,and from the overall point of view,the absolute value of the forecast error is basically kept within 20%,which has good controllability and engineering practicability.Through parameter sensitivity analysis,it is proved that the input parameters selected in this paper have a strong correlation with the amount of disc cutter wear.