Normal Wear Prediction of Disc Tool Based on Clustering and Transformer Algorithm for Tunneling in Upper Soft and Lower Hard Strata
Shield tunneling is the main method for tunnel construction,widely used in soft strata and composite strata.How to accurately predict the tool wear when the shield machine is tunneling in the upper soft and lower hard strata is an important issue to ensure construction safety and efficiency.Based on the unsupervised clustering algorithm,supervised Transformer algorithm and genetic algorithm in machine learning,a method is proposed to predict tool wear based on the ground conditions,tool placement and construction parameters.The kmeans clustering algorithm is used to analyze the relationship between the shield construction parameters,so as to classify different construction states in order to assign different wear correction coefficients to different construction states.The initial values of wear correction coefficients are determined by the ground conditions and construction parameters.Genetic algorithm is used to optimized the values of wear correction coefficients.In Transformer algorithm,the ground conditions,construction parameters,tool cutting radius,and wear correction coefficients from kmeans clustering are used as input parameters,and tool wear is used as output parameter,and the hyperparameters of the model are optimized by the genetic algorithm.The data used are collected from a large-diameter slurry balance shield machine in Shenzhen,and the ground conditions contain full-section soft soil,upper soft and lower hard strata,and full-section hard rock.The dataset contains 217samples.The construction states revealed by the clustering results have clear correspondence with the tool wear,and the average prediction error of tool wear of the prediction model is 3.6mm,accounting for 9%of the max tool wear of 40.0mm.
万方数据
维普
