ID 31880
広兼 道幸
In Japan, cases of slope failure are occurring in every year, each bringing about a considerable loss. In particular, those due to heavy rains by typhoons, built, or seasonal rain, fronts, etc. are bound to be disastrous. For instance, a slope disaster in the west region of Shimane Prefecture in July 1983 took a heavy tool of 107 lives and its material damage amounted to over \360 billion. Out of the 107 killed or missing, 91, or 85% were victims of slope failure and debris flows. Cases of slope failure involving such disasters are reported in various regions all over the country in almost every year. Besides, as of 1986, national roads under the administration of Ministry of Construction were exposed to the danger of rock-fall and slope failure at 2,619 points in mountain areas, and municipal roads at numerous points throughout the country. Countermeasures in both hardware and software are required to prevent such disasters. The countermeasure in hardware is to take material measures to slopes judged dangerous beforehand. For instance, we will choose a proper slope protection structure in accordance with each slope's condition and execute the slope protection structure to it. The countermeasure in software is to take fundamental measures for the countermeasure in hardware to function effectively. Involved in it are to diagnose vulnerable slopes and determine among them the priority order for slope protection structures to be executed, to build systems for predicting the danger of slope failure beforehand and give early and precise warning to the population for refuge, and so on.

Slope failure is caused by complex working among various factors such as geology, topography, ground, vegetation, rainfall, etc. As a method of diagnosing the danger levels of slope failure, there is an approach to making clear all such factors, elucidating the mechanism of slope failure, and building a prediction model. This approach has often been used in the fields of civil engineering, erosion control, etc. and several prediction models have been proposed to date. Most of these models, however, incorporate only a few factors. Some of the factors are difficult to quantify. Besides, these models are built on the assumption that accurate information about their built-in factors are obtainable. To determine parameters for evaluating the ground properties, however, we have still often to rely on our subjective judgement. Even if all parameters could be made clear, it is reckoned difficult to find their values in situ. Thus, in many cases accurate information on all necessary factors is not available, and results of prediction often turn out to be of considerably poor reliability when these models are used.

In addition, factors such as spring water, vegetation, and slope failure histories are involved in diagnosing the danger levels of slope failure, and the prediction models proposed at present can not properly evaluate these factors. In municipal disaster prevention programs, on the other hand, engineers in change and well familiar with their areas, but without using such models, have been performing the diagnoses of danger levels based on the facts and experiences in the past, and refuse programs have been prepared based on their diagnoses. Similar phenomena can be observed on the hardware side for instance, in the work of choosing proper slope protection structures. However, past facts and experiences, based on which an engineer in charge is performing the diagnoses of danger levels and choosing proper slope protection structures, belong to himself, and at present his knowledge of the facts and his experiences are not shared with other engineers.

This paper presents, the subject being slopes alongside roads, a method of sharing with others the experiential knowledge that an engineer with a career of ten years or more in choosing slope protection structures and diagnosing the danger levels of failure acquired from the facts and experiences in his service.

Chapter 1, "INTRODUCTION," outlines the ES, that is drawing attention as a method of sharing knowledge of experts, and describes a method of representing knowledge and the ambiguity in it, and so on. This chapter also mentions how the ES is applied in the field of civil engineering. Further described is a diagnostic case of the danger level of slope failure by an expert, that is used in Chapter 3, Chapter 4 and Chapter 5.

In Chapter 2, "EXPERT SYSTEM FOR SELECTING OF SLOPE PROTECTION STRUCTURES," the application of the ES is attempted for sharing of past facts and experiences knowledge of experts. Here, the subject is focused on the choice of slope protection structures, points to be heeded in actually applying the ES are described, a method of representing ambiguity in their experiential knowledge that experts unconsciously use in making their judegment is proposed, a method of determining the degree of the effects that the factors for judegment have upon the results is proposed, and a process in which a constructed ES is evaluated by using actual cases of slope protection structures and the ES is improved toward the practicability is described.

Chapter 3, "ANALYSIS OF OBSERVATION DATA FOR DIAGNOSING SLOPE-FAILURE DANGER LEVEL WITH S-P SCORE TABLES," presents a method of acquiring knowledge which is considered most difficult and important in sharing knowledge with the ES. Described here are a method of determining the weights of judgement factors with S-P score tables and a method of evaluating the results of diagnoses performed by engineers other than experts, both based on a diagnostic case of the slope failure danger level by an expert. Because, in the application of S-P score tables, a case has to be transformed into "0's" and "l's," an expert's case can not be used in its original form. To address this problem, presented is a method of using an expert's case in its original form with a generalized S-P score table, without transforming it into "0's" and "1's."

Chapter 4, "ANALYSIS OF OBSERVATION DATA FOR DIAGNOSING SLOPE-FAILURE DANGER LEVEL WITI I ROUGH SETS" presents a method of acquiring know leddge which is considered most difficult and important in sharing knowledge with the ES. In case that we observe slope, etc. and review the judgement factors for diagnoses of failure danger levels based on the observation results, they may contain factors not determined and those determined erroneously If expert's knowledge is sought for from such cases, shortage and errors of information contained in their observations themselves may be reflected into acquired knowledge. Thus, presented is a method of removing inconsistencies from such cases and acquiring the minimum knowledge to deduce results. Then following is a method of constructing a knowledge base for the ES, using acquired knowledge. Further following is a method of determining the values of correctness and reliability of the knowledge and evaluating the knowledge base.

Chapter 5, "ANALYSIS OF OBSERVATION DATA FOR DIAGNOSING SLOPE-FAILURE DANGER LEVEL WITH MULTIVARIATE ANALYSIS." using the quantification class-II of multivariate analysis, discriminant analysis is performed on the basis of a diagnostic case of the slope failure danger level by a expert. Multivariate analysis enables to elucidate factors or phenomena connecting wirh each other complicatedly, and it has, since long time ago, been used in analyses of data for diagnoses of danger levels of slope failure. Analyses through these traditional methods were performed mainly to compare with the methods proposed in Chapter 3 and 4. Conceivable as the results of multivariate analyses are alinement discriminants, or the weights to judgement factors indicated by coefficients in alinement discriminants. In the comparison with the analysis results obtained in Chapter 3 and 4, the weights to the judgement factors indicated by coefficients in alinement discriminants were used.

In Chapter 6, "EVALUATION OF ANALYSIS METHODS USING S-P SCORE TABLES, ROUGH SETS, AND MULTIVARIATE ANALYSIS," described are the features of the methods presented in Chapter 3, Chapter 4 and Chapter 5. An actual diagnostic case of the slope failure danger level by an expert was analyzed by using the three methods of acquiring knowledge for building the ES, the results obtained were compared with one another, and the significance of each method was verified.

Finally, the results of the present study are summarized and subjects for the future are mentioned in Chapter 7.

The features of this paper in the above-mentioned composition are that an ES was constructed to address the problem of selecting slope protection structures where rationalization and standardization with computers, etc. have been lagging to date, and that described are the applicability and a building method of the ES against the problem of handling experiential knowledge of experts which has been considered difficult to systematize. Besides, proposed in Chapter 3 and Chapter 4 are, although already being used in other fields, methods using S-P score tables and rough sets to handle the problem of acquiring experts' experiential knowledge which is considered very difficult and important and of which the method has yet to be established.

In the appendix, points and procedure for building an ES are described on the basis of the ES actually built up Chapter 2 and a part of knowledge base actually constructed is summarized.
第1章 序論 / p1
 1.1 本論文の目的 / p1
 1.2 本論文の構成と内容 / p3
 1.3 エキスパートシステム / p5
 1.4 既往の研究 / p22
 1.5 本論文で使用する斜面の崩壊危険疫診断の事例 / p34
 参考文献 / p44
第2章 斜面保護工の選定に関するエキスパートシステム / p49
 2.1 緒言 / p49
 2.2 知識ベース / p50
 2.3 エキスパートシステムの構築 / p69
 2.4 エキスパートシステムの評価 / p73
 2.5 結語 / p82
 参考文献 / p84
第3章 S-P表を用いた斜面の崩壊危険度診断事例の分析 / p87
 3.1 緒言 / p87
 3.2 S-P表分析 / p88
 3.3 判定要因の評価 / p93
 3.4 診断結果の評価 / p99
 3.5 一般化したS-P表分析 / p109
 3.6 一般解したS-P表による判定要因の評価 / p110
 3.7 一般化したS-P表による診断結果の評価 / p115
 3.8 結語 / p126
 参考文献 / p128
 4.1 緒言 / p131
 4.2 ラフ集合 / p132
 4.3 斜面の崩壊危険度診断への適用 / p144
 4.4 結語 / p155
 参考文献 / p157
第5章 多変量解析を用いた斜面の崩壊危険度診断事例の分析 / p159
 5.1 緒言 / p159
 5.2 多変量解析 / p160
 5.3 数量化II類 / p161
 5.4 斜面の崩壊危険度への適用 / p170
 5.5 結語 / p180
 参考文献 / p182
第6章 S-P表,ラフ集合,および多変量解析による分析方法の評価 / p183
 6.1 緒言 / p183
 6.2 S-P表による分析方法の特徴 / p184
 6.3 ラフ集合による分析方法の特徴 / p186
 6.4 多変量解析による分析方法の特徴 / p187
 6.5 結語 / p188
 参考文献 / p192
第7章 結論 / p193
謝辞 / p197
付録 / p199
 1 エキスパートシステムの構築方法 / p199
  1.1 題材の検討 / p199
  1.2 問題の検討 / p200
  1.3 知識の獲得 / p203
  1.4 システムの構築 / p208
  1.5 システムの評価 / p209
  1.6 システムの維持 / p215
 2 エキスパートシステムの構築例 / p216
参考文献 / p240
Copyright(c) by Author
広島大学(Hiroshima University)