Identifying probable failure modes for underground openings using a neural network

Abstract

Abstract The design of underground openings in complex geological environments is restrained by our limitations in defining the geological environment in which the opening will be created and in modelling the response of the geological environment to the excavation created and support procedures used. This paper describes the use of a neural network to identify probable failure modes for underground openings from prior case history information. This step in geotechnical design is a critical step in identifying the dominant geologic parameters to be included in a simplified geologic and engineering model. The structure of the neural network adopted and the “learning” algorithm by which the neural network obtains its knowledge from case histories are described. The results of “learning” are then used to examine the operational characteristics of the neural network. Four experiments are designed to test its abilities in inferring possible failure modes, retrieving patterns from partial cues (content-addressability) and in resistance to faulty input data. Limitations of and possible improvements on the neural network are also described. Use of the knowledge obtained by the neural network learning in a geotechnical design context is demonstrated by a tunnel design assistance system. This neural network approach differs from conventional rule-based expert systems in the manner of knowledge representation and the problem solving process. Instead of applying rules and facts to end up with conclusions, the approach solves problems by pattern matching and allows input information to be incomplete and vague.