3. Expert System

Apart from the knowledge base for disc cutter TBMs outlined above, there exist many other factors that need to considered, and on these opinion varies widely. The construction of an expert system, compiling and summarizing these opinions and the knowledge resource, is greatly facilitated through the use of computing. A computer-based expert system has the advantage of ease and speed of use, convenience for accumulating knowledge, and the ability to clearly identify and present contradictions between experts.

The present authors have conducted a number of preliminary trials in the development of an expert system. The final system proposed in this paper consists of three stages:

Stage A. Check whether the fundamental requirements are fulfilled.

Stage B. Estimate the performance of the TBM in terms of thrust force, rolling
force, penetration, rotational speed, penetration rate, advance rate, etc.

Stage C. Evaluate the values estimated in stage B in reference to the various
models in the knowledge base.

In stage A, the fulfillment of the fundamental requirements for the applicability of TBM to a particular project is checked. The items checked include tunnel length, tunnel diameter, minimum radius of curvature, inclination/gradient, power supply, and machine and material haulage. An example of such a checklist is given in Table 3, with some conditions that are not conducive to TBM tunneling. At the end of this stage, the expert system identifies problem areas, and the user makes the decision of whether to proceed onto stage B and conduct a more detailed study.

At the outset of development of this expert system, as in expert systems in other fields, we planned to arrange the knowledge in a consistent fashion and provide a comprehensive result in the second stage. However, such a knowledge structure was found to be complex and difficult to design after we began to research previous papers and collect field data. It was apparent that the design of such a structure and subsequent compilation would be a lengthy process. After such a prolonged development period, the expert system would be out of date before completion, particularly in this field where new technology, knowledge and machine improvements are developed on almost a daily basis. For this reason, we decided to separate this second part of the expert process into two stages, B and C. Stage B is based on a relatively small number of rules or equations and has a simple structure. Further study is carried out in stage C, to which new knowledge and rules are added continuously.

In stage B, TBM performance is assessed based on the knowledge and equations
given in section 2. The detailed procedures are shown in Figure
4. Initially,
the penetration rate H is assumed to be 20 m/h, and the rotation speed of the
cutter head w is then estimated. Input data such as uniaxial compressive
strength, penetration p, thrust force F_{N} and rolling force FR are also
calculated. These results are then assessed in terms of whether the values are
within the reasonable range based on the point system given in Table
4. A total
of zero points is the strictest criterion, and this criterion was applied to the
tunnels examined in the next section. However, the user of the expert system has
the freedom to change this suitability criterion. In practice, the initial
assumption of H = 20 m/h is too high, and in most cases will not pass. A new
value of H is then assumed and the same procedures are iterated until a feasible
value of H is found.

After H is found, the expert system attempts to determine the net working
rate and advance rate. Through a detailed study of field data, it was found that
the calculated H is reasonably consistent with the maximum attainable
penetration rate. However, this is the ideal case, and in practice such speed
will not be maintained by the TBM due to a wide range of reasons, including a
shortage of haulage capacity, delays in spraying concrete (shotcrete),
precautions against ground condition changes, inexperienced operation, and
machine overheating. Therefore, the result is multiplied by an empirical
constant of 0.5 in order to obtain the mean penetration rate Hmean (= 0.5H). The
values of p, F_{N} ,F_{R} and so on corresponding to Hmean are then calculated in the
same way as before. It should be noted that this empirical constant is a very
rough estimation and may increase with improvements to total excavation systems.

The net working rate is calculated in consideration of the mean penetration rate, uniaxial compression strength, excavation diameter and tunnel length. The net working rate is estimated according to the calculated point score as shown also in Table 5. In this paper, the net working rate is defined as

r = (hours for excavation)/(total days * hours per day) (3)

Total days is the number of days from the date of commencement of excavation to the date of completion, including holidays, days for machine maintenance and repair, and days for support, but not days for machine construction prior to excavation or dismantling following excavation completion. Therefore, the values of r given in Table 5 are smaller than those calculated using the relation (hours for excavation)/(net working days L hours per day).

At the end of stage B, the data set of F_{N}, F_{R}, p, d, w, r, H, Hmean and V is
output as input data for stage C. The difference between this expert system and
the typical estimation process is that a computer program inputs proposed values
sequentially and rules are used to judge the suitable of the data. This concept
is also used in stage C, where proposed values or a data set are evaluated
according to rules or existing knowledge.

The algorithm for stage C consists of many discrete procedures, representing individual opinions. Brief summaries of the opinions are given in Table 6. For example, Saito et al. (1971) proposed one equation and one rule. Using his equation given in Table 5, the estimated cutter cost is calculated and the results are displayed to the user, along with results based on the data set input from stage B. These processes are carried out by an independent procedure named Saito. The data set is then evaluated according to the opinions proposed by Roxborough and Pillips (1975). In this manner, the data set is evaluated by all models in the knowledge base, and the results are presented to the user. As stated before, the opinions vary from author to author, but we consider that it is valuable to show the evaluation results based on many opinions even if contradictions exist between them. The procedure in stage C is similar to consulting many experts as to the feasibility of the proposal (in this case, output of stage B), a procedure that is frequently carried out in pre-feasibility studies. The compilation of models as discrete procedures in the program for stage C makes it easy to modify or add knowledge or rules, allowing the expert system to be continuously updated.