Wear Detection Of Shield Hob Based On Deep Learning

Based on the projects of Guangzhou rail transit line 18 and 22,this paper studies the disc hob,which is the key component affecting the efficiency of shield construction,and analyzes the feasibility and practicability of in-depth learning method for quantitative detection of hob wear.The main research contents are as follows:This paper discusses the background and research significance of the subject,expounds the research status of shield tool wear detection at home and abroad,summarizes the composition and working principle of shield machine,and discusses the classification of hobs: single edge hob,double edge hob and three edge hob.This paper introduces the wear characteristics of hobs and several widely used mechanical models of the interaction between hobs and rocks.During shield construction,in order to ensure construction efficiency,the wear amount of hob can be set as follows: front hob 25 mm,edge hob 15 mm and center hob 25 mm.If it exceeds this range,the hob can be changed.Because the data of disc hob wear is very limited,the Holm wear calculation formula is used to distribute the wear to each ring,and the data set used to establish the neural network model is expanded.Using the gray analysis method,it is analyzed that the wear of hob is closely related to the variables of cutter head speed,cutter head thrust,installation radius and cutter head torque,and takes it as the input feature of neural network.Using the method of data standardization,the five-dimensional data set is constructed by using the pre measurement of cutter head speed,cutter head thrust,installation radius,cutter head torque and hob single ring wear,which reduces the calculation error.The BP neural network is built.By discussing the learning rate of the network,the number of hidden nodes and optimizing the network,the hob wear is quantitatively predicted,and the feasibility of BP neural network in hob wear detection is verified.A wear detection method based on adaptive Cauchy mutation particle swarm optimization and short-term memory neural network(acmpso-lstm)is proposed.The number of hidden layer neurons,learning rate and training iteration times of LSTM network are optimized by ACMPSO algorithm.Experiments verify the effectiveness and superiority of this method in the quantitative detection of hob wear,and the average detection error is controlled at about 7%.