Articial Neural Network Based Inverse Reliability Analysis

cartcletchΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 4 χρόνια και 8 μήνες)

94 εμφανίσεις

Articial Neural Network Based Inverse
Reliability Analysis
David Lehky,Brno University of Technology
Drahomr Novak,Brno University of Technology
The inverse reliability problem is the problem to nd design parameters
corresponding to specied reliability levels expressed by reliability index or by
theoretical failure probability.Design parameters can be treated as a determi-
nistic or as a random.In case of random,the actual design parameters could
be either mean or standard deviation of random design variable.In addition,
multiple design parameters are required,if multiple reliability constraints are
specied.The aimis to solve generally not only the single design parameter case
but also the multiple parameters problem with given constraints.The objective
of the inverse procedure is to nd,directly,the design parameters so that the
target reliability levels corresponding to constraints will be satised.Several ap-
proaches have been already established in literature to seek design parameters
of basic random variables at the desired reliability.
A new general approach of inverse reliability analysis is proposed to obtain
design parameters of a computational model in order to achieve the prescribed
reliability level.The inverse analysis is based on the coupling of a stochastic
simulation of Monte Carlo type and an articial neural network (ANN).The
design parameters (e.g.mean values or standard deviations of basic random de-
sign variables) are considered as basic random variables with a scatter re ecting
the physical range of potential values.A novelty of the approach is the utiliza-
tion of the ecient small-sample simulation method Latin Hypercube Sampling
(LHS) used for the stochastic preparation of the training set utilized in training
the articial neural network.The calculation of reliability is performed using the
rst-order reliability method (FORM).Once the neural network has been trai-
ned,it represents an approximation consequently utilized in a following way:To
provide the best possible set of design parameters corresponding to prescribed
reliability.The validity and eciency of the approach is shown using numerical
examples from civil engineering computational mechanics.
This outcome has been achieved with the nancial support of the Ministry of
Education,Youth and Sports of the Czech Republic,project No.1M0579,within
activities of the CIDEAS research centre.In this undertaking,theoretical results
gained in the project GACR No.103/07/P380 were partially exploited.