Levent-Hisarüstü metro tünelinde kullanılan darbeli kırıcı için performans tahmini modellerinin önerilmesi

Hydraulic hammer excavation can serve as a fast method requiring lower capital cost. Compared to drilling and blasting method, mechanical excavations can also offer more control on strata and a safer working environment. To choose the best method of excavation, one should consider feasibility, installation problems, ability of negotiation with adverse geological conditions, total cost and advance rate. Therefore, determining the advance rate (performance) of an excavator is an indispensible task even from the very early phase of feasibility studies. Impact hammers have been widely used in mining and in the field of construction since 1960. Lower capital cost compared to Tunnel Boring Machines (TBM) and Roadheader, makes impact hammer a desirable choice when the conditions are favorable. Being able to operate flexibly is another advantage of impact hammer, which makes it irreplaceable in terms of mining engineering. The flexibility makes the operator able to either follow irregular ore bodies through bulk material or use foliations or beddings to ease excavation process. Although, there had been studies on performance prediction of impact hammers in the past years, most of them are solely focused on the properties of the rock mass. Moreover, the previously proposed models suffer from either limited range of application or shortage of accuracy and reliability. . The aim of this study is to assess multiple regression analysis (MRA) and artificial neural network (ANN) using a new and ample set of data (60 rock samples) gathered from Levent-Hisarüstü metro project in Istanbul, to suggest more accurate and reliable performance prediction models for impact hammer using a simpler combination of the most commonly used physical and mechanical properties of rocks. The optimum prediction model will be the one that provides the highest reliability and accuracy (compared to the other models developed during this study and previously developed prediction models by the other researchers) while requiring the most comparatively few, common, and easy to access set of predictors. To fulfil such a goal, physical and mechanical properties of excavated rocks in Levent-Hisarüstü metro tunnel were investigated using the most accepted standards in test procedures. The unshaped rock samples were collected from the tunnel face and sharpened using the preliminary instruments such as small diameter saw and core drill machine. Physical and mechanical property tests include uniaxial compressive strength, Brazilian tensile strength, Shore scleroscope hardness, Schmidt hammer rebound values, Cerchar abrasivity index, point load test, and density. In addition, performance of the impact hammer used in Levent-Hisarüstü metro tunnel was carefully monitored and recorded. Moreover, RQD values were calculated using Brown's method and volumetric joint counts for each individual tunnel ring and the types of rock along the tunnel alignment were determined. Sandstone is the main rock (60%) encountered with compressive strength values ranging from 68.6 to 145 MPa and RQD values from 40 to 60%. Siltstone is met in 20% of the tunnel with compressive strength from 45.8 to 123.3 MPa and RQD 38-44%. In the tunnel route, 10% of the rock formations are composed of diabase with compressive strength ranging from 158.7 to 195.6 MPa and RQD 45-60%. Mudstone and shale are met in 10% of the tunnel with compressive strength from 8.9 to 40.1 MPa and RQD of 10-40%. Next, multiple regression analysis (MRA) methods were used to investigate the relationship between the physical and mechanical properties and recorded IBR values. The multiple regression analysis yielded a performance prediction model for impact hammers. Finally, the analysis of variance technique, F-test and student's t-test statistics, which are based on sum of squares of errors, were applied to the data in order to understand the reliability of the suggested models. Analysis indicated that the proposed models are statistically meaningful with overwhelmingly very high confidence level and/or very low significance value in two-tail analysis. Among all the recorded data, collinear pairs of parameters were identified. Knowing linearly parallel parameters can lead into better test design and significant cost reduction for the future studies. An ANN model was developed to increase the prediction accuracy even furher. It provided extremely high correlation between predicted and target data. Meanwhile, the MSE is signifantly low. The associated final weights and biases of the network were reported in order to make the network more handy and transcend the technical limitations. The developed ANN and MLR models provide better correlations of all the available data along with considerably lower MSE values. For the test data, although the proposed model by Bilgin et al. (1997) shows surprisingly good correlation, it carries a considerably high MSE compared to the developed ANN and MLR models. The reason can either be the steep technological achievements during the last two decades that can affect performance of the machine or the fact that the data that were used to develop this model did not include RQD values between 38 to 78. The developed prediction models contribute to reliable and accurate prediction of instantaneous breaking rate of impact hammer requiring the least number of the most commonly used and versatile rock properties as predictors. The developed MLR model shows that an increase in RQD and UCS causes the IBR to decrease. As a result, one may say that higher rates of breaking can be achieved when the mentioned factors have lower values. This study also showed that the relationship between IBR as predicted value and CAI and SSH as predictors is extremely weak. This indicates that CAI and SSH do not contribute to prediction of IBR for impact hammers. This research reapproved the belief that increasing the number of the samples will result in an increase in accuracy of the model. Since the effect of machine's power on the performance of the impact hammer is not investigated in this study, it is highly recommended to perform future studies on this subject. In addition, by gathering more data the range of application for the future models will be extended. Finally, incorporating the operator's effect will increase accuracy and reliability of the future models.