Computerization of Rock Engineering Systems Using Neural Networks with an Expert System

Summary With the Rock Engineering Systems (RES) methodology, rock mechanics and rock engineering problems are studied systematically using a total systems approach, incorporating rock mass properties, interrelated parameters, complex interaction mechanisms and dynamic behavioral modes. In this paper, a method of implementation and computerization of RES is considered, using neural networks together with an expert system. The computerized RES starts with the data processing of rock mass properties and boundary conditions and data base management of rock engineering case records. This step is followed by building and operating parameter interaction matrices with the combined use of backpropagation networks and an expert system. Finally, a simulator for modelling the dynamic process of rock engineering systems using the Hopfield network is incorporated. With the aid of neural networks' learning capability and expert system's symbol-reasoning capability, the RES approach is implemented in an “intelligent” mode.