報告書番号

U94033

タイトル（和文）

ニューラルネットワークを用いた非線形構造解析手法の検討（その４）非対称履歴則のモデリングと剛体－ばねモデルへの応用

タイトル（英文）

APPLCATION OF NEURAL NETWORKS TO NONLINEAR STRUCTURAL ANALYSIS(PART4) MODELING OF ASYMMETRIC HYSTERETIC BEHAVIOR WITH NEURAL NETWORKS AND APPLICATION TO RIGID-BODY-SPRING-MODEL

概要 （図表や脚注は「報告書全文」に掲載しております）

本研究報告は，ニューラルットワークを用いた非対称歴則のモデリングと簡易構造解析への応用について論じたものてある。ニューラルネットワークによる認識対象は，曲り配管の弾塑性大変形有限要素解析から得られたエルボ部の曲げモーメント回転角関係である。履歴挙動の経路依存性を適切にとらえることができるよう曲げモーメント－回転角関係図上の二つの座標点と現在の回転角を入力に与え，曲げモーメントを出力するネットワークを作成した。非対称履歴則は，回転角の増分値を正負区別し，二つのネットワークに分けることにより取り扱った。学習後，収束精度の高いネットワークを剛体－ばねモデルに組み込んだ曲り配管の非線形解析を行った結果，有限要素解析結果との良好な一致が見られ，ニューラルネットワークを用いた履歴モデリングの有効性が確認された。

概要 （英文）

NEURAL NETWORKS(NNS) ARE COMPUTING SYSTEMS MODELED AFTER THE STRUCTURE OF THE HUMAN BRAIN. THERE HAVE BEEN A NUMBEROF ATTEMPTS TO APPLY NNS TO STRUCTURAL ENGINEERING. ALTHOUGH NNS CAN BE TRAINED TO SOLVE MOST PROBLEMS WITH SUFFICIENT TRAINING EXAMPLES,THERE ARE FEW CASES WHERE NNS HAVE BEEN APPLIED TO PRACTICAL NONLINEAR STRUCTURAL ANALYSIS. THIS REPORT DESCRIBES ASYMMETRIC HYSTERETIC MODELING WITH NNS AND ITS APPLICATION TO SIMPLIFIED NONLINEAR STRUCTURAL ANALYSIS. THE RECOGNITION TARGETS ARE THE BENDING MOMENT-ANGLE OF ROTATION RELATIONSHIP OF ELBOW ELEMENTS OBTAINED FROM ELASTIC-PLASTIC FINITE ELEMENTANALYSES OF THIN-WALLED CURVED PIPES. THE NN SHOWED THAT THE PATH-DEPENDENCY OF HYSTERETIC BEHAVIOR COULD BE ADEQUATELY REPRESENTED BY INPUTTING THE CURRENT ANGLE AND TWO PREVIOUS COORDINATE POINTS ON THE MOMENT-ANGLE RELATIONSHIP. TO REPRESENT THE ASYMMETRIC BEHAVIOR,TWO NNS DIVIDED INTO POSITIVE AND NEGATIVE ANGLE INCREMENTS WERE USED. AFTER TRAINING,THE NNS WERE APPLIED TO A RIGID-BODY-SPRING-MODEL(RBSM) DEVELOPED FOR ANALYZING THE LIMIT STATE OF STRUCTURES. THE RESULTS WERE ACCEPTABLE SHOWING THAT MODELING WITH NNS COULD BE INCORPORATED INTO NONLINEAR STRUCTURAL ANALYSIS. THE DIFFERENCE BETWEEN CURRENT MATHEMATICAL MODELING AND NNS IS THE NECESSITY FOR EXPLICIT FORMULATION. IN NN MODELING,WHERE IT IS UNNECESSARY TO FORMULATE DIRECTLY,THERE ARE NO RESTRICTIONS AND SIMPLIFYING ASSUMPTIONS IN THE EXPRESSION,SO IT IS POSSIBLE TO IMPROVE ACCURACY. IMPROVEMENT OF ACCURACY OF NONLINEAR STRUCTURAL ANALYSES BY NN MODELING WILL CONTRIBUTE TO ESTABLISHMENT OF LIMIT STATE STRUCTURAL DESIGN.

報告者

キーワード