Expert System for Applicability of Tunnel Boring Machines in Japan
4. Discussion

As shown in Table 7, the expert system was applied to 18 tunnels in Japan. Tunnels 1 to 11 are old tunnels (JSCE, 1976; Okubo, 1990) constructed prior to 1977, and 12 through 18 are new tunnels (JTA, 2000; Fukui et al, 2000) constructed after 1992. This expert system was developed mainly for use in pre-feasibility studies in which available information is limited. Therefore, only the tunnel length, tunnel diameter, uniaxial compression strength and minimum curvature of radius were used. The diameter of the disc cutters was assumed as follows (JTA, 2000):

300 mm for tunnel diameter less than 3 m
400 mm for tunnel diameter between 3 - 5 m
430 mm for tunnel diameter larger than 5 m

In all cases, 16 working hours per day were assumed, although in reality this would vary considerably from tunnel to tunnel.

In stage A, the fundamental conditions for each tunnel were evaluated in order to assess the appropriateness of the use of a TBM in their construction according to Table 3. Only 3 of the old tunnels, 8 through 10, were identified as being unsuitable, due to their short length. All of the other tunnels were then passed to the next stage.

In stage B, the daily advance rate was estimated for each tunnel, as shown in Figure 5. The results indicate that the mean advance rates of new tunnels was 6.56 m/d, and that of old tunnels was 5.55 m/d. These results may indicate the technical advance of TBM excavation in the past 20 to 30 years. The ratios of actual/expected advance rate are 0.7 and 0.75 for old and new tunnels, respectively. These low values are attributed to unexpected difficulties. For example, in the Saijo tunnel, the specification of the disc cutters was changed, and modifications were made to the TBM. In the Hamada tunnel, the tunnel length was only 155 m, and high excavation speed was not important. Among the new tunnels, Nikengoya, Doshi and Shinyuyama tunnels experienced considerable difficulties, with significant down-time due to unexpectedly bad ground conditions in fracture zones. In general, ground conditions in Japan are difficult to forecast due to rapidly changing ground water levels and the prevalence of fracture zones. This remains a topic for future research.

Equation (2) was proposed more than 20 years ago for calculating the rotation speed of the TBM. Present-day TBMs are typically much more powerful and have higher rotation speeds. However, as shown in Figure 6, the actual rotation speed in most cases is far below the expected value, except for the Takisato tunnel. For old tunnels, low power was the main reason for the low rotation speed, however, for new tunnels, low rotation speed is selected by the TBM operator. This is attributed to cautious operation preempting abrupt changes in ground conditions, and in some cases, a shortage of haulage capacity. Very recently, haulage capacity was increased by the introduction of extension belt conveyers as widely used in coal mines.

Penetration p is also shown in Figure 6. The mean cutter spacing was about 80 mm, making at least 10 mm of penetration theoretically desirable for efficient cutting. However, only Takisato exceeded this value and actual values in other tunnels were lower than expected. The rock mass at Takisato was weaker than anticipated, and a TBM with sufficient power was used. It can also be considered that low penetration may be partly attributed to inexperienced operation.

For FN, actual values were lower than expected. The difference between actual and expected values is thought to be partly attributable to the fact that the equation used did not consider the increase in FN with disc diameter. For example, in tunnels 17 and 18, relatively large disc cutters with a diameter of 432 mm were used, for which a higher FN is necessary to achieve a given penetration. It should also be noted that, the measurements of FN include friction between the cutter head and the rock mass, which was significant in some cases. For FR, the difference between the actual and expected values was slightly smaller.

In the case of the tunnel excavated in Chicago (Illinois, USA), the tunnel diameter was 5.7 m and the average uniaxial compression strength was about 135 MPa, with a maximum of 260 MPa and minimum of 65 MPa (Okubo, 1990). We studied this 1970s example using the proposed expert system, and estimated the advance rate at 5.2 m per day. The expert system did not point out any problems other than a high maximum strength. The actual advance rate of 13.4 m per day is far beyond the predicted result. This tunnel is very similar to the 10th section of the Sakawagawa water tunnel in Japan, also constructed in the 1970s. The diameter of the Sakawagawa tunnel was 4.8 m, and the average uniaxial compression strength was 120 MPa. The actual advance rate was 3.3 m per day, which is considerably less than the value estimated by the expert system (Okubo, 1990).

The Chai Wan tunnel in Hong Kong constructed in 1993 was studied as another example of tunneling in countries other than Japan. The information is shown in Table 7 and Figures 5 and 6 as tunnel 19. The actual advance rate was more than 14 m per day and far beyond the expected value. The actual rotation speed, FN and FR were also higher than expected. Penetration alone was lower than expected. In the other cases we studied, advance rates of greater than 10 m per day are relatively common in foreign countries. The primary reasons for the low speed exhibited in Japan are thought to be complex ground conditions, inexperience of the operators, and a shortage of haulage capacity.

In stage C, related knowledge is shown to the user. In this system, the information presented is likely to be contradictory, however, we believe that this is valuable to users in the majority of cases. Stage C is rather complicated; here we will provide a simplified explanation. As the output of stage B, information such as the uniaxial compressive strength and penetration are provided to stage C. In the first part of stage C, the values of FN are calculated according to the equations given in Table 1 and compared with the estimated values in stage B. If the difference exceeds 20%, the user will be notified of the equation giving the erroneous result. Advanced rates are then calculated according to the equations given in Table 2, and again the user will be alerted to any significant discrepancies.

The values obtained in stage B are then checked against a set of rules. A subset of these rules is given in Table 5. The expert system asks the user for any additional information necessary to examine the applicability of a TBM to the site with more precision. For example, if necessary, a user is asked to input the maximum and minimum uniaxial compression strengths expected. In Japan especially, the sonic velocity of the target rock mass is commonly measured in advance. Then the maximum, mean and minimum velocities expected can be input. In this case, a mean velocity of less than 2.5 km/s and a difference between maximum and minimum of greater than 3 km/s results in warning message. Information regarding expected fracture zones is also requested, however, it is often a difficult question to answer.

These results suggest that there is indeed room for improvement. However, the system presents itself as simple, easy to use, and highly expandable for including new models and developments.

This system was written in Pascal and Basic. The initial trials were conducted in Prolog, however this language was abandoned when the system was divided into the three discrete stages. Using current personal computers, outputs even from Basis were produced in less than one second.