The authors discuss the development of an expert system called Classex which classifies ground according to the RMR and Q methods and predicts various parameters of stability and support. C1asscx was developed as a convenient aid for rock mechanics personnel and as a learning tool for students. It uses the VPEXPERT development tool and a personal computer.
Correlations between alternative input parameters for both Q and RMR give the user latitude in pre-design data collection. Once a consultation is performed to obtain either Q or RMR, previously entered input can be applied to calculate the other. Thus, comparisons can be made of recommended support, maximum spans, support pressures, etc., predicted from both systems. Much of the subjective nature of classification is eliminated and automatic checks ensure both consistent input and output. Difficulties and ambiguities in both Q and RMR were encountered during the development of Classex; these are discussed.
Expert systems show considerable promise in rock engineering. particularly for use in ground classification and empirical design. Computer-based expert systems combine judgement and experience with analysis in a hybrid approach that closely simulates the decision-making processes of real life. Seldom in engineering or mining are decisions made solely on the basis of numerical modelling or any other single design tool. The factors to be considered are many and interrelated. often in a complex manner. Expert systems provide a very useful aide memoir under such conditions.
Examples of their use in geomechanics are given by Star-field et al. (1983). Dershowitz and Einstein (1984). Fairhurst and Lin (1985), Nguyen and Ashworth (1985), Bandopadhyay and Venkatsubramanian (1988). Paul and Gershon (1988). and Zhang et al. (1988).
An expert system can work as an empirical design tool to predict ground behaviour directly from index characteristics by storing and appropriately applying "rule-of-thumb" Correlations. It can also provide constitutive equations for closed form or numerical modelling of rock mass behavior, by predicting properties such as bulk shear strength or deformability starting with knowledge only of the index characteristics of the rock mass. It can suggest to the user Which analytical tools are most appropriate and can even encompass and supervise the modelling process itself.
Perhaps the main advantage is that an expert system pro- Vides less objective, less biased. thorough, pre-planned and painstaking predictions. Judgemental errors caused by stress or carelessness are minimised. With careful questioning and exhaustive yet rapid consideration of many possible scenarios. the user can be guided towards a pre-evaluated solution.
A further important characteristic of an expert system is the ability to contain a much larger store of rules and facts than can be remembered by any single person. Experts are expensive and not always available. The expert engineer may leave the company or retire before his or her knowledge can be passed on to a junior. Expert systems allow experience not just of one but of several experts to be encoded for future use by anybody not unlike how rock mass classifications permit the use of previous tunnelling experience by today's tunnellers.