TBM penetration rate prediction using neural networks and neuro-fuzzy systems

TBM penetration rate (PR) prediction is one of the most crucial issues for project cost and time estimation, however, its prediction has remained an important challenge for engineers and investors. Results of former investigations show that there are different methods for PR prediction, including theoretical and statistical models as classic methods and neural networks, fuzzy logic systems, and neuro-fuzzy models as intelligent and new methods. Modern methods are more capable of analyzing complex and non-linear relationships in comparison with classic methods. Accordingly, the implementation of modern methods for PR prediction will lead to a more precise outcome. In this paper, the information of 14 tunnels around the world is compiled within a database. Studied parameters in this database are a combination of machine and rock mass specifications, including, normal mean thrust force (Fn), cutterhead revolution per minute (Rpm), tunnel diameter (TD), rock mass rating (RMR), rock quality designation (RQD), and uniaxial compressive strength (UCS). By analyzing and reviewing relevant results, it was determined that omission or failure to use appropriate parameters causes a poor PR prediction. These results also show that UCS and RQD are amongst the most effective parameters. Furthermore, it has been concluded that using neuro-fuzzy networks (RMSE = 0.13 m/h) provides more accurate results than neural networks (RMSE = 0.38 m/h).