آلان شوکتی

Predicting penetration rate of tunnel boring machines (TBM) is a decisive factor in scheduling and budgeting of tunnelling projects. This paper is aimed at predicting the TBM penetration rate (ROP) in the southern lot of Kerman water conveyance tunnel by means of Group method of Data Handling (GMDH) neural network. Having compiled the database using the geomechanical data of rock mass and machine performance data, correlations between ten various parameters were explored and two linear and nonlinear multivariate regression equations for the penetration rate were set up. Multi-objective Genetic Algorithm in the form of bi-objective optimization was applied for designing the optimal structure of the network and the dataset was randomly divided into training subset (70% of the total data) and test subset (the remaining 30%) as two objective functions. A multi-layered polynomial penetration rate function in terms of the parameters having the strongest correlation with ROP, i.e. compressive strength of the rock mass, quartz content, the angle between plane of weakness and TBM-driven direction, and the average force acting on the single cutter was obtained. Application of the rock mass compressive strength led to reducing the number of involving parameters and making the prediction model simpler. The comparison of the observed and predicted values showed high determination coefficient (R2) of 0.81 (R2=0.6 for nonlinear multivariate regression) which reveals high prediction capability of the proposed GMDH model. Unlike other neural network prediction models which produce their outputs as a “black box”, the suggested ROP GMDH model was expressed as a recurrent polynomial function in terms of the inputs. This outstanding feature of the GMDH model enables the proposed prediction model to be used in other projects as well as future research.