퍼지-신경망에 의한 사면지반의 안정성 평가방법에 관한 연구

Abstract

Since the stability of slope is determined by comprehensive examination of layer composition condition, topographical condition, presence or absence of sudden groundwater inundation, technological characteristics of soil and base rock, systematic and rational analysis processes as well as the application of reinforcement method, should be used for the long-term and economical maintenance of slope stability. Therefore, this study applies various existing evaluation methods such as limit equilibrium analysis, experimental slope stability test, stereo-graphic projection method and the test by RMR. The stability prediction methods by the reliability approach of point estimation, by the neural network as well as the fuzzy-neural network are applied to the four sections of the slope. The stability results are analysed and compared with the statistical approach such reliability and fuzzy-neural network methods. Limit equilibrium analysis shows that stability of slope may be varied greatly depending on the influence of pore water pressure and the property of soil and rock slope. It has limitation in stability test due to reliability of stereographic projection and geological survey. Since the safety value indicates that of a particullar section, it is judged that it has some uncertain part to evaluate the stability of the whole slope. Evaluation methods by experiences of soil slope can be considered to be valid in establishing causal analysis, repair and reinforcement measures of soil instability. However, since the interpretation method lacks the objectivity, there is difficulty in evaluating stability of the slope. The values of factors having influence on the process in classification of rock slope such as compressive strength and RQD may be erroneous in the process of converting it into number. Contrary, reliability interpretation by point estimation of slope stability prediction methods can draw reliability index and safety factor of the whole slope section with mean and standard deviation of probability variables. Safety factor shows more reliable value than that of limit equilibrium analysis and it is judged that slope stability can be considered carefully. In addition, since the test by fuzzy approximate reasoning can evaluate the various internal and external factors inherent in each slope quantitatively, it can evaluate slope stability interpretation by combination with the existing soil slope stability interpretation method more rationally. Systematic test of slope stability by fuzzy neural network built in this study reflects the information of human reliability interpretation properly, can objectify subjective state test, which may have ambiguity, through fuzzy logic and neural network fault tree analysis. It makes exact numerical interpretation possible by MATLAB program and does more systematic and comprehensive stability analysis through effective role of stability test prediction. In addition to the existing slope stability tests used for the stability test of soil and rock slopes, the results of the reliability method by point estimation and the fuzzy approximate reasoning are compared and analyzed in this study. When the slope stability test based on neural network and fuzzy neural network are combined in evaluating slope stability, it is expected that more safe and economical slope design can be expected. In order to fulfill this objective, establishing database of probability variable and uncertainty estimation are needed.

Keywords: Fuzzy­Neural Network / Fuzzy Approximate Reasoning / Fuzzy Fault Tree Analysis / Slope Stability / Slope Safety Factor / Reliability analysis / Point Estimation Method