本文ここから

A Development of Fuzzy Pavement Condition Assessment and Radial Basis Function Neural Networks based Pavement Deterioration Model

(ファジー理論を応用した路面性状評価指標と動径基底関数ニューラルネットワークに基づいた舗装劣化モデルの開発に関する研究)

氏名 Joni Arliansyah
学位の種類 博士(工学)
学位記番号 博甲第285号
学位授与の日付 平成15年12月31日
学位論文題目 A Development of Fuzzy Pavement Condition Assessment and Radial Basis Function Neural Networks based Pavement Deterioration Model (ファジー理論を応用した路面性状評価指標と動径基底関数ニューラルネットワークに基づいた舗装劣化モデルの開発に関する研究)
論文審査委員
 主査 教授 原田 秀樹
 副査 助教授 山田 良平
 副査 助教授 大橋 昌良
 副査 助教授 小松 俊哉
 副査 東北大学大学院情報科学研究科 助教授 武山 泰

平成15(2003)年度博士論文題名一覧] [博士論文題名一覧]に戻る.

CONTENTS

Acknowledgements p.i
Abstract p.ii
Contents p.iv
List of Figures p.vi
Llst of Tables p.ix

1.INTRODUCTION p.1
1.1 BACKGROUND AND OBJECTIVES OF RESEARCH p.2
1.2 SUMMARY OF THESIS p.4

2.A DEVELOPMENT OF FUZZY PAVEMNET CONDITON ASSESMENT p.7
2.1 LITERATURE REVIEW p.8
 2.1.1 Concept of Fuzzy Set Theory p.8
 2.1.2 Fuzzy Weighted Average Operation p.10
 2.1.3 Interpretation of Fuzzy Output p.13
2.2 METHODOLGY p.14
 2.2.1 Fuzzy Pavement Condition Assessment p.14
 2.2.2 Members Functions Determination p.16
 2.2.3 Date Used p.19
 2.2.4 Example Application p.20
 2.3 ANALYSIS ON THE RESULTS OF PAVEMENT CONDITION ASSESSMNET p.21
 2.3.1 Reliability of The Proposed Method p.21
 2.3.2 The Effects of The Inclusion or Omission of Pavement Prameter p.24
 2.3.3 The Effect of Wight Changes of Pavement Prameter p.25
 2.3.4 The Sensitivity of The Linguistic Rating Terms'Range Values of Pavement Prameter p.26
2.4 THE COMPARUSON OF FUZZY PAVEMENT CONDITION INDEX (FPCI) AND MCI p.29
2.5 CONCLUSIONS p.32
FEFERENCES p.32

3. A PAVEMENT DETERIORATION MODEL USEING RADIAL BASIS FUCTION NEURAL NETWAORKS p.34
3.1 LITERATURE REVIEW p.34
 3.1.1 Radial Basis Functions Neural Networks p.34
 3.1.2 Orthogonal Least Squares Learning Algorithm p.36
3.2 METHPDLOGY p.40
 3.2.1 Data Used and Family of Pavements p.41
 3.2.2 The Architectures of Pavement Deterioration Model Using RBFNN p.43
3.3 RESULTS AND DISCUSSION p.45
 3.3.1 Pavement Deterioration Model Based on The Long Histories Conduction Data p.45
 3.3.2 Pavement Deterioration Model Based on The Database that has Two Points History Condition Data p.52
3.4 CONCLUSIONS p.57
FEFERENCES p.57

4.A NEURAL-FUZZY PAVEMENT DETERRIORATION MODEL p.59
4.1 METHODLOGY p.60
 4.1.1 Data Used and Family of Pavement p.61
 4.1.2 FuzzyPavement Condition Index Computation p.61
 4.1.3 Application of FPCI in Pavement Deterioration Model Using RBFNN p.62
4.2 RESULTS AND DISCUSSION p.62
 4.2.1 RBFNN Optimization p.65
 4.2.2 The Comparison between The RBFNN Model Results with Actual Rating and Markow Model Results p.66
4.3 CONCLUSIONS p.69
FEFERENCES p.69

5.CONCLUSIOS AND FUTURE RESEARCH p.70
5.1 CONCLUSIONS p.71
5.2 FUTURE RESEARCH p.72

Pavement Management Systems (PMS) are widely used in the world to assist administrators of pavement networks in making consistent and cost effective decision about public investment in highways. The ability both to represent the condition of pavement and to predict the future condition of pavements are the fundamental parts of PMS. These parts have great influence on the reliability of the final results of PMS.
In this thesis, a development fuzzy pavement condition assessment and new pavement deterioration model using radial basis function neural networks (RBFNN) are presented. The pavement data from "the database of Hokuriku Region pavement management support system" and "the report of the structural design of asphalt pavement, technical memorandum of Japan Public Work Research Institute, 1991" were used. This thesis is divided into five chapters and organized as follows.
Chapter 1 presents the background, objectives and summary of the study.
Chapter 2 discusses a development of fuzzy pavement condition assessment. The fuzzy pavement condition index (FPCI) model that designed to be used in Japan and a method to determine membership functions of linguistic terms used in pavement condition assessment based on experts' opinion about the linguistic rating term range RBFNN) with orthogonal least squares (OLS) learning algorithm. The architectures of proposed pavement deterioration model are designed as sequential pavement deterioration model where the future pavement condition of a pavement can be predicted using only information about present MCI value and age of a pavement. This model can be used to develop pavement deterioration model based on the database that has at least 2 points history condition data, and the model has fast convergence time and guarantees convergence to global optimum parameter. The comparison between the results of the RBFNN model with actual MCI and other existing methods' results are discussed.
In Chapter 4, the development of a neural fuzzy pavement deterioration model is presented. This model uses the FPCI as one of the input variables in RBFNN based pavement deterioration model. The comparisons of the model's results with the actual rating and Markov model results are analyzed and discussed.
In Chapter 5, the conclusions of the study are summarized, and the future work, which emerges from this study, is given.
The results indicate that the membership functions of the proposed membership function determination method give more reliable results than existing membership functions, and FPCI model seems to be able to give more reliable results than MCI model. The proposed RBFNN based pavement deterioration model is able to give satisfactory pavement condition prediction results and its application is very flexible. A neural fuzzy pavement deterioration model gets the advantages from both FPCI model and RBFNN model where the model can account for human subjectivity and impression associated with the evaluation of pavement parameters and has good capability to predict the future pavement condition.

非公開論文

お気に入り

マイメニューの機能は、JavaScriptが無効なため使用できません。ご利用になるには、JavaScriptを有効にしてください。

ページの先頭へ戻る