Numerical methods commonly used in rock mechanics, such as finite element, displacement discontinuity, and boundary integral equation methods, offer solutions to problems which can be described by differential equations subject to boundary conditions. These numerical methods use approximations leading to solutions in terms of matrix algebra and numerical integration. Many rock mechanics problems which are beyond the scope of such approximations can be treated effectively by application of techniques developed in the field of artificial intelligence (AI). Among the possible applications are: site characterization (Duda, 1975), geological interpretation (Smith and Baker, 1983), solution of problems involving rock Joint networks, and underground construction. Each of these applications utilize particular features of AI. Here we shall present results of AI applications which utilize primarily search and to lesser extent learning or pattern recognition. Specific examples presented are in the field of rock fracture network problems and underground construction.
Artificial intelligence techniques attempt to achieve human-like reasoning and learning capabilities. This reasoning capability is valuable in applications where repetition, complexity, or tediousness makes human-like intervention impractical. In rock mechanics this occurs, for example, in stochastic rock mess modeling applications where some complex decision-making is required for each of a large number of iterations. The strength of artificial intelligence techniques is especially marked in applications where the n-tuber of possible states which must be considered in decision making becomes very large. For simple problems, with limited numbers of states to be considered, traditional nested "if-then-else" constructs (see below) can be sufficient, and artificial intelligence is unnecessaL7. However, as the complexity of the decision states increases it becomes increasingly difficult and cumbersome to attempt to enumerate all possible decisions, and a unified AI approach becomes attractive. Rock mechanics is a discipline in which the number and complexity of variables controlling decision- making can be large, as illustrated by rock mass classification methods (e.g., Bieniawski, 1976, Barton et al., 1976). The handling of complex relations is even more problematic when "learning" is involved e.g. the incorporation of experience in exploration planning, design decisions and construction planning. As a relatively new technique in rock mechanics applications, there is still considerable confusion about the difference between artifical intelligence and techniques which have been widely used for some time. Much of AI can be viewed as an extension of techniques already in common use: if-then-else logic, relational data bases for representative of information, and integration of logical and computational techniques (algorithms). Below we will discuss each of these, and how AI extends and applies them. Further information can be found in Winston (1979) and Nielson (1971).
"If-then-else" and AI logic Whenever decisions must be made within standard rock mechanics computer programs, the decisions are "hardwired" into the program with if-then-else branches, illustrated in Figure 1. If-then-else branches are extremely powerful; in efficient programs they are nested. The number of situations that can be handled increases geometrically with the depth of nesting.
FIGURE 1: NESTED "IF-THEN-ELSE' LOGIC (available in full paper)
Within if-then-else logic, FORTRAN programming language allows one to use the Boolean operators .AND., .OR., and