Performance Prediction Of Tunnel Boring Machine By Using Machine Learning Techniques

Prediction of machine performance and cutter consumption are considered as a nonlinear and multivariable complex issues.This study is undertaken to: i)establish an artificial intelligence(AI)framework to determine the performance of tunnel boring machine(TBM),ii)predict the machine performance(i.e.,advance rate and penetration rate)during tunneling construction,iii)establish a statistical and intelligent models to predict the disc-cutter life,iv)analyze the performance,characteristics,and influence factors of each parameter during tunneling.In order to investigate these issues,statistical analyses,machine learning techniques,intelligent analysis,and verification from the field observation are conducted.Firstly,the most effective parameters in the prediction of machine performance are determined.New prediction models for advance rate and penetration rate of shield machine are proposed.Secondly,new model is established to predict the performance of disc-cutter life.To achieve a better performance,an intelligent approach is proposed based on both geological and operational parameters.Then,a discussion is conducted to analyse the performance of the most significant parameters in the prediction of disc-cutter life.The innovative results of this study are summarised as follow:(1)New machine learning model is proposed to predict the advance rate of TBM.The proposed model integrates an improved particle swarm optimization(PSO)with adaptive neuro-fuzzy inference system(ANFIS).The improved model depended on a combination of fuzzy rule-based system and PSO algorithm,which simultaneously adjust both of antecedent and consequent variables.This model demonstrates better accuracy in predicting the advance rate of shield performance than the current models,e.g.neural network,fuzzy logic,ANFIS,and empirical models.(2)Multi-objective optimization model is proposed to predict the penetration rate.The optimization model integrates the adaptive neuro-fuzzy inference system(ANFIS)with genetic algorithm(GA).GA is employed to enhance the accuracy of ANFIS for runtime parameters tuning by multi-objective fitness function.It is found that the optimized model has higher accuracy in the prediction of penetration rate of shield machine than ANFIS model.(3)A prediction model for the disc-cutter life is proposedAn empirical model is proposed to predict the disc-cutter life.With proposed model,the effect of shield construction parameters and mechanical properties of rock and soil on the machine performance are investigated.The proposed empirical model can provide a reasonable means for quick evaluation of construction parameters in predicting disc-cutter life with an acceptable range of accuracy in comparison with the previous models.(4)An intelligent approach is established to predict the disc-cutter life.The intelligent model integrates group method of data handling polynomial neural network(GMDH)with genetic algorithm(GA).GA is employed to optimize the most appropriate network structure of GMDH,enabling each neuron to search for its optimal set of connections with the preceding layer.Although both empirical and intelligent models are applicable for the estimation of disc-cutter life,the intelligent GMDH-GA model is able to provide a higher degree of accuracy.Furthermore,an analysis is performed to investigate the most significant parameter in the prediction of disc-cutter life during tunneling process.It is found that the penetration rate of the shield machine is the significantly influenced parameter in the prediction of disc-cutter life model.(5)Research achievements are verified with field observed data.Two field cases,Ma-Lian section in Guangzhou metro line 9 tunneling project and the Bao'an section in Guangzhou-Shenzhen inter-cities railway project are used to verify proposed AI models.To obtain this aim,the data including information from the field,laboratory,and literature are collected along the tunnel alignment and applied to the proposed models.The effectiveness of prediction models for TBM performance and disccutter life is confirmed.Finally,this research shows the feasibility of using different novel machine learning techniques for predicting TBM performance,disc-cutter life,and subsequent works to increase the opportunity of successful projects using TBM field data and the aforementioned models to achieve more comprehensive prediction methods.