Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique

A selective Monte Carlo simulation procedure
 for the evaluation of the reliability of nonlinear
 structures subjected to dynamic loading is presented.
 The proposed “Russian Roulette and
 Splitting” procedure allows generation of important
 low-probability sa...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Проблемы прочности
Datum:2003
1. Verfasser: Labou, M.
Format: Artikel
Sprache:Englisch
Veröffentlicht: Інститут проблем міцності ім. Г.С. Писаренко НАН України 2003
Schlagworte:
Online Zugang:https://nasplib.isofts.kiev.ua/handle/123456789/47022
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Zitieren:Solution of the First-Passage Problem by Advanced Monte Carlo
 Simulation Technique / M. Labou // Проблемы прочности. — 2003. — № 6. — С. 67-74. — Бібліогр.: 5 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
_version_ 1860016509117530112
author Labou, M.
author_facet Labou, M.
citation_txt Solution of the First-Passage Problem by Advanced Monte Carlo
 Simulation Technique / M. Labou // Проблемы прочности. — 2003. — № 6. — С. 67-74. — Бібліогр.: 5 назв. — англ.
collection DSpace DC
container_title Проблемы прочности
description A selective Monte Carlo simulation procedure
 for the evaluation of the reliability of nonlinear
 structures subjected to dynamic loading is presented.
 The proposed “Russian Roulette and
 Splitting” procedure allows generation of important
 low-probability samples in nonlinear dynamics.
 This procedure is quite efficient in
 comparison to the straightforward Monte Carlo
 simulation method and allows evaluation of the
 nonlinear stochastic response in a very low
 probability domain. Предложен модифицированный численный метод Монте-Карло для оценки надежности
 нелинейных конструкций, подвергнутых динамическому нагружению. Использование метода
 позволяет обобщить важные низковероятностные выборки для нелинейных динамических
 задач. Эффективность данного метода по сравнению с прямом методом Монте-Карло
 состоит в возможности определения нелинейного стохастического отклика в области
 очень низких вероятностей. Запропоновано модифікований числовий метод Монте-Карло для оцінки
 надійності нелінійних конструкцій під дією динамічного навантаження.
 Метод дозволяє узагальнити важливі низькоімовірнісні вибірки для нелінійних
 динамічних задач. Ефективність даного методу в порівнянні з прямим
 методом Монте-Карло полягає в можливості визначення нелінійного сто-
 хастичного відгуку в області дуже низьких імовірностей.
first_indexed 2025-12-07T16:45:01Z
format Article
fulltext UDC 539.4 Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique M. Labou Institute of Structural Mechanics, Kiev State University of Construction and Architecture, Kiev, Ukraine УДК 539.4 Решение задачи “первого вы броса” модифицированным численным методом Монте-Карло М. Лабу Институт строительной механики при Киевском национальном университете строительства и архитектуры, Киев, Украина Предложен модифицированный численный метод Монте-Карло для оценки надежности нелинейных конструкций, подвергнутых динамическому нагружению. Использование метода позволяет обобщить важные низковероятностные выборки для нелинейных динамических задач. Эффективность данного метода по сравнению с прямом методом Монте-Карло состоит в возможности определения нелинейного стохастического отклика в области очень низких вероятностей. Клю чевы е слова : динамическое нагружение, нелинейная динамическая задача, метод Монте-Карло. In tro d u c tio n . In a recent review o f methods currently available to analyze multidegree-of-freedom systems under stochastic loading, their merits, limitations and potential to be used in engineering practice have been discussed [1]. From this study it becomes clear that only num erical procedures such as Monte Carlo simulation (MCS) techniques (both in their direct and variance reduction form) and the Response Surface M ethod (RSM) are capable o f treating nonlinearities without restriction as well as systems o f higher dimension including reliable information on the response distribution tails. As direct MCS procedures are quite lim ited with respect to providing accurate information on the response distribution tails, the variance reduction techniques such as importance, directional, adaptive sampling, etc. are used for static structural problems in order to guide the simulation procedures to the region o f interest and, hence, to reduce the numerical effort. For dynamic problems, the selection o f this region is more involved than for static problems, since the importance measure is generally time variant. Based on a suitable criterion for indicating which realization will m ost likely lead to failure, the samples in the region o f interest m ay be split, whereas in the region o f less interest they m ay be discarded or m ay survive. This procedure has been expanded and applied successfully to M DOF systems. © M. LABOU, 2003 ISSN 0556-171X. Проблемы прочности, 2003, № 6 67 M. Labou 1. F irst-P assage D istribu tions. A general performance measure for structures under stochastic dynamic loading is characterized by first-passage probabilities, where the distribution o f the exit times T into a state o f failure is considered. The boundary separating the safe state from the unsafe one m ight be specified by a so-called limit state function g (x ) = 0. The lifetime o f the structure is then defined as the first exit time T : T = in f [t: g ( X ( t )) < 0, t > 0], (1) and the probability o f the first excursion P f ( t ) within the time interval t is defined as Pf ( t ) = P[T < t ], (2) specifying the evolution o f the failure probability w ith respect to time. Presently, the available solutions to the above first probability are quite rare. There is a complete absence o f analytical exact solutions in closed form, and numerical procedures based on the Fokker-Planck equation are confined to problems o f low dimension (< 4 in the state space). The only solution technique at hand seems to be the MCS, which is well suited to simulate nonlinear stochastic structural responses as required in the reliability assessment o f structures. 2. M eth o d of A nalysis. For the purpose o f assessing the reliability o f a structure, the domain o f the response is divided into the safe and unsafe parts, respectively. The so-called limit state function (LSF) serves as a criterion of separation. The domain is in general m ultidim ensional containing parameters such as displacements, velocities, plastic deformations, etc. I f x is defined as the state vector o f all the response components and g (x ) as the LSF, it is necessary to determine the probability P ( t ) that the response vector x( t ) exceeds the LSF at least once within a given time period [0, t], where x(0) lies initially in the safe domain, i.e., P (0) = 0. In other words, the first-passage probability defines the evolution o f the failure probability w ith respect to time. It is well known, that the direct MCS is well suited to obtain, w ith a relatively small num ber o f realizations, reasonably close estimates for the m ean vector and the covariance matrix. It is not suited, however, to assess the low probability domain o f the stochastic response and for reliability analysis where_o _4 failure probabilities P f are o f the order 10 _ 10 . I n order to access such ranges, it is obvious that a sample size (number o f realizations) o f the order 105 _ 1 0 9 will be required. Such sample sizes are possible for small systems using a m assively-parallel supercomputer [2]. However, the MCS with a sample size o f the order > 105 are expensive and certainly not efficient in view o f the fact that only a small fraction o f the num ber o f realizations falls into the domain o f interest for the reliability analysis. This drawback o f the direct MCS has been commonly recognized, and the so-called variance reduction techniques have been developed and utilized successfully. All these procedures increase the density o f realizations in the region o f interest, i.e., in the region that contributes m ost to the failure 68 ISSN 0556-171X. n p o 6n.eubi npounocmu, 2003, N 6 Solution o f the First-Passage Problem probability. These variance reduction methods are used m ainly for nonlinear static reliability systems, but can be also used for nonlinear dynamic problems. The main difficulty with these approaches is the fast growing num erical effort w ith the num ber o f random variables involved. In case loading is represented by a stochastic process, it m ight be possible in some cases to represent the stochastic process by a small num ber o f random variables using the Karhunen Loeve expansion. M onte Carlo simulation, however, is applicable to stochastic loading involving more than ju st a small set o f random variables. In such cases, the traditional variance reduction procedures are no longer applicable. Advanced M CS methods have been developed for the class o f systems with M arkovian properties. Before describing the details and im plementation o f such procedures, we shall briefly discuss the m ain features o f the advanced MCS. Consider first the M CS and the resulting approximation for the cumulative distribution function CDF (x ): N CDF (x , t ) = 2 1 [ x n ( t ),x ] w n ( t ), (3) n=1 where x is the state vector, t is time, x n is the state vector o f the nth realization, w n ( t ) is the w eight and discrete probability o f the nth realization at time t, and N is the sample size. The indicator function I [X n ( t ),x ] takes the value 1 i f the components X n k > x k , k = 1,..., M , for all components M o f the state vector. Otherwise, the indicator function assumes the value 0. In case the direct MCS is applied, all weights are constant and assume the value w n ( t ) = = 1/N . For example, for importance sampling, the original probability density function f (x ), which reflects the direct MCS, is m odified by using the sampling distribution h(x ). This can be done without violating the original density function f (x ) by m odify ing the w eight w n o f all realizations X n ( t ) by w n = = f (x ) / ( Nh(x )). Since the advanced MCS is designed to assess low probability _8 _4 ranges o f the order 10 P f < 10 as well, it is obvious that m anipulation of the weights is indispensable. A selective M onte Carlo simulation technique, namely, “Russian Roulette & Splitting” simulation technique (RR&S) has been applied successfully in nuclear physics [3] to solve neutron transport problems. The basic features o f the algorithm are described and discussed. The RR&S procedure involves subdivision o f the safe region by several borders o f splitting (or sub-barriers). Then the process o f splitting can be introduced as follows. W hen the response sample function x ( t ) upcrosses the ith sub-barrier, the state vector is split into mt same vectors, where m is the integer num ber and i is the num ber o f the sub-barrier. The weight o f each new vector is equal to w new = w ĉid /m i . Due to the m i stochastically independent sample functions produced by splitting, one simply increases the chances to upcross the next (more highly placed) sub-barrier. Thus the splitting technique makes it possible to increase significantly the num ber o f samples which are capable o f crossing the limit state function. ISSN 0556-171X. npoôëeMbi npounocmu, 2003, N2 6 69 M. Labou The Russian Roulette is a well-known game o f chance where the ‘probability o f death’ is high (i.e., 1/6, originally). This game can be played to reduce the sample size to a required amount. The “Russian Roulette” technique is applied in an inverse situation, i.e., when the response sample function downcrosses the ith sub-barrier. In this case with a probability o f 1 — 1 / , the simulation process is discarded (killed) and w ith a probability o f 1/mt the simulation is continued. The weight o f such surviving sample is m ultiplied by mi . Hence, the main aim o f the “Russian Roulette” technique is to diminish drastically the num ber o f sample functions which are o f “low interest” in the first passage problem. From the algorithm described above it becomes obvious that between the ith and ( i + 1)th sub-barriers the weights o f the samples are equal to each other and are expressed as - 1 n -Wi - - n — , (4)n .m , w ] - 1 ] where n is the num ber o f initially simulated samples and —] is the num ber of the splitting on the j th sub-barrier. It is clear that the weights o f the samples Wb after outcrossing the barrier will be equal to k w b - 1 n (5) j -1 ] where k is the total num ber o f sub-barriers. It is noteworthy that the direct Monte Carlo algorithm is a subset o f the RR&S technique. In particular, the direct MCS can be obtained from the RR&S for mt - 1 [4]. 3. D e te rm ination of the Effectiveness o f the S im ulation Technique. According to Rubinstein [5], the effectiveness E o f each simulation technique can be expressed in the following form: 1 E - v ar[P ]T (6) where var [P ] is the variance o f the estimate and T is the computation time of the estimate P * . The Bernoulli scheme o f independent trials in the case o f straightforward MCS allows one to obtain the following well-known formula for the variance of the estimate: * P(1 — P ) var [p ] - \ ' , (7) where N is the sample size. 70 ISSN 0556-171X. n p o 6n.eubi npounocmu, 2003, N2 6 Solution o f the First-Passage Problem Unfortunately, due to the correlation between samples in the RR&S technique, there is a difficulty in obtaining an expression for the estimate variance similar to (7). Based on the sample statistics, the variance o f estimate for the RR&S technique can be calculated by the following formula: * 1 ^ ( li * v a r [P ] = —-------t > | — - P n( n — 1) “ V M \2 / (8) where * 1 P = - >n M 1=1 (9) k M = П mj , j =1 n is the sample size, lt is the num ber o f barrier outcrossings in the ith independent trial, and mj is the number o f the splitting on the j th sub-barrier. As long as variance (7) has already been minimized for a given sample size N by the straightforward Monte Carlo simulation, a more effective solution can be obtained using another simulation scheme (8) with a higher variance o f the estimate and m uch shorter computation time. 4. N um erical Exam ples. Nonlinear vibrations o f a cylindrical panel under the action o f a uniformly distributed load are analyzed; the time history is described by a truncated white noise. It is considered that at the initial instant of time the cylindrical panel is in a quiescent mode. The action o f the applied load triggers the excitation o f a nonstationary stochastic process. The following parameters o f the panel are adopted: the dimensions in plan a = 0.2 m, b = 0.3 m, the height h = 0.002 m, the modulus o f elasticity 3 3E = 70.63 GPa, the density o f the material p = 2.7-10 kg/m , Poisson’s ratio v = 1/3, and radius o f curvature R = 0.8363 m. In the problem under study, the frequency band o f the spectral density o f the loading covers the first, fifth, seventh, and eleventh symmetrical vibration modes. The equation o f motion is governed by the following differential equation: x i + C tx i + K ix i + K f x j x k + K f lx j x kxi = ц i Q i ( t) ( 10) ( i, j , k , l = 1 ,..., 4), where C j are factors o f the damping matrix, p t are the oscillation modes, Q ( t ) is the Gaussian white noise with intensity I = 1000 m 2/s3, and K ijk and K ijkl ( i, j , k , l = 1 ,..., 4) are, respectively, the coefficients o f matrices o f square and cubic nonlinearity. ISSN 0556-171X. Проблеми прочности, 2003, № 6 71 M. Labou 2 2The following input data are considered: K 1 = m 1 = 1.00, K 2 = m 2 = 7.64, K 3 = m 2 = 11.2, K 4 = m 2 = 22.00, C 1 = 0.0159, C 2 = 0.044, C 3 = 0.0532, C 4 = 0.0747, / i 1 = - 0.0297, ^ 2 = 0.0095, ^ 3 = 0.00979, ^ 4 = 0.00342. The flexures in the center o f the cylindrical panel are determined by the formula 4 x c ( 0 = h 2 x i ( t ) . (11) Z=1 The first passage probabilities o f the system response x c ( t ) were determined during time [0, t] for double-sided symmetrical barriers located at the levels ± 3h , ± 4h, and ± 5h and shown in Fig. 1. xc (t), mm -1500 t, s t, s Fig. 1. Sample white noise function Q (t) - a and the system response variation xc (t) - b. A t the beginning, the first passage probability for symmetrical barriers of level ± 3h was determined. Figure 1a shows one realization o f the stochastic process Q ( t ). The corresponding variation o f the system response x c( t) is presented in Fig. 1b. This trajectory reaches the limiting barrier at t = 4 s. In the same figure, the arrangement o f borders o f splitting and the numbers o f the splitting are shown. In this case, 6 borders o f splitting were located at ±0.15 cm, ±0.3 cm, and ±0.45 cm; the num ber o f the splitting at each o f the borders are mi = m2 = m3 = 7; the sample size n = 1000. Figure 2 (curve a ) presents in a logarithmic scale the cumulative distribution function o f the first passage probabilities o f the system response x c ( t ) for barriers located at the level ± 3h. From this figure one can see that at t < 4 s the probability o f reaching the barrier sharply decreases. The appropriate events are rather rare. We also analyzed the problems on the first passage probability o f the system response x c ( t ) for barriers located at the levels ± 4h and ± 5h. The appropriate cumulative distribution functions are shown in Fig. 2 (curves b and c, respectively). 72 ISSN 0556-171X. npoôëeubi npounocmu, 2003, N2 6 Solution o f the First-Passage Problem T a b l e 1 The Effectiveness and Calculation Time Ratios - MCS to RR&S Time (s) err &s /EMCS TMCS TRR &S 2.22 169.00 169.0 2.50 4.20 149.0 3.00 1.10 83.0 4.00 0.40 38.0 5.00 0.17 22.4 t, s t, s Fig. 2 Fig. 3 Fig. 2. Estimates of the first passage probabilities versus time for the double-sided barriers located at a = ±3h, b =±h, c = ±5h. Fig. 3. Estimates of the first-passage probabilities and variances of estimates for barriers located at ± 5 h. In the second problem (barriers at the level ± 4h) 6 borders o f splitting were located at distances ±0.2 cm, ±0.4 cm, and ±0.6 cm, the numbers o f the splitting at each o f the borders are mi = m2 = m3 = 7; the cumulative distribution function for this problem is shown in Fig. 2 (curve b). In the third problem (barriers at the level ± 5h) 6 borders o f splitting were located at distances ±0.25 cm, ±0.5 cm, and ±0.75 cm, the num ber o f the splitting at each o f the borders are mi = m2 = m3 = 7, the appropriate cumulative distribution function is shown in Fig. 2 (curve c). The sample size n in the second and third problem equals to 1 0 0 0 . * For the barriers located at ± 5h, the variance o f estimates P was calculated for different techniques. Figure 3 (curve a) represents cumulative distribution functions obtained successively by the straightforward MCS (dotted lines), and the RR&S procedure (solid lines) (see Fig. 2, curve c). Figure 3 (curve b) represents the variance o f estimates P * , obtained w ith the help o f formula (8); and curve (c) obtained using formula (7). The ratio o f the effectiveness Err&s / E MCs calculated for different times t in accordance with (6 ) is presented in Table 1 together w ith the ratio o f the calculation times tmcs / trr &s ■ ISSN 0556-171X. Проблемы прочности, 2003, № 6 73 M. Labou From Table 1 it is seen that the RR&S simulation technique offers a considerable gain in computation time and a higher effectiveness (in the first 3 s) as compared to the direct MCS technique. Conclusion. A n advanced M onte Carlo simulation procedure has been applied to M DOF systems showing its basic applicability to higher dimensional problems and the ability to assess the low probability range in terms o f the first-passage probabilities. Р е з ю м е Запропоновано модифікований числовий метод М онте-Карло для оцінки надійності нелінійних конструкцій під дією динамічного навантаження. М етод дозволяє узагальнити важливі низькоімовірнісні вибірки для неліній­ них динамічних задач. Ефективність даного методу в порівнянні з прямим методом М онте-Карло полягає в можливості визначення нелінійного сто- хастичного відгуку в області дуже низьких імовірностей. 1. G. I. Schuller, H. J. Pradlwarter, and M. D. Pandey, “M ethods for reliability o f n o n lin ear system s u n d er stochastic load ing - a rev iew ,” Proc. EUROD Y N ’93, Balkema (1993) pp. 751-759. 2. E. A Johnson E. A. and L. A. Bergmann, “M onte Carlo simulation of dynamic systems o f engineering interest in a m assively parallel computing environment,” in: Proc. IUTAM Symp. “Advances in Nonlinear Stochastic Mechanics,” (Trondheim, Norway, July 3 -7 , 1995), Kluwer Academic Publishers, Dordrecht, the Netherlands (1996), pp. 225-234. 3. J. Spanier and E. M. Gelbard, Monte Carlo Principles and Neutron Transport Problems, Addison W esley Publishing Company (1969). 4. P. G. M el’nik-M el’nikov, E. S. Dekhtyaruk, and M. Labou, “Application of the “Russian Roulette and Splitting” simulation technique for the reliability assessment o f m echanical system s,” Probl. Prochn., No. 3, 131-135 (1997). 5. R. Y. Rubinstein, Simulation and the Monte Carlo M ethod , W iley, N ew Y ork (1981). Received 25. 09. 2003 74 ISSN 0556-171X. Проблеми прочности, 2003, № 6
id nasplib_isofts_kiev_ua-123456789-47022
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
issn 0556-171X
language English
last_indexed 2025-12-07T16:45:01Z
publishDate 2003
publisher Інститут проблем міцності ім. Г.С. Писаренко НАН України
record_format dspace
spelling Labou, M.
2013-07-08T16:13:55Z
2013-07-08T16:13:55Z
2003
Solution of the First-Passage Problem by Advanced Monte Carlo&#xd; Simulation Technique / M. Labou // Проблемы прочности. — 2003. — № 6. — С. 67-74. — Бібліогр.: 5 назв. — англ.
0556-171X
https://nasplib.isofts.kiev.ua/handle/123456789/47022
539.4
A selective Monte Carlo simulation procedure&#xd; for the evaluation of the reliability of nonlinear&#xd; structures subjected to dynamic loading is presented.&#xd; The proposed “Russian Roulette and&#xd; Splitting” procedure allows generation of important&#xd; low-probability samples in nonlinear dynamics.&#xd; This procedure is quite efficient in&#xd; comparison to the straightforward Monte Carlo&#xd; simulation method and allows evaluation of the&#xd; nonlinear stochastic response in a very low&#xd; probability domain.
Предложен модифицированный численный метод Монте-Карло для оценки надежности&#xd; нелинейных конструкций, подвергнутых динамическому нагружению. Использование метода&#xd; позволяет обобщить важные низковероятностные выборки для нелинейных динамических&#xd; задач. Эффективность данного метода по сравнению с прямом методом Монте-Карло&#xd; состоит в возможности определения нелинейного стохастического отклика в области&#xd; очень низких вероятностей.
Запропоновано модифікований числовий метод Монте-Карло для оцінки&#xd; надійності нелінійних конструкцій під дією динамічного навантаження.&#xd; Метод дозволяє узагальнити важливі низькоімовірнісні вибірки для нелінійних&#xd; динамічних задач. Ефективність даного методу в порівнянні з прямим&#xd; методом Монте-Карло полягає в можливості визначення нелінійного сто-&#xd; хастичного відгуку в області дуже низьких імовірностей.
en
Інститут проблем міцності ім. Г.С. Писаренко НАН України
Проблемы прочности
Научно-технический раздел
Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique
Решение задачи “первого выброса” модифицированным численным методом Монте-Карло
Article
published earlier
spellingShingle Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique
Labou, M.
Научно-технический раздел
title Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique
title_alt Решение задачи “первого выброса” модифицированным численным методом Монте-Карло
title_full Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique
title_fullStr Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique
title_full_unstemmed Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique
title_short Solution of the First-Passage Problem by Advanced Monte Carlo Simulation Technique
title_sort solution of the first-passage problem by advanced monte carlo simulation technique
topic Научно-технический раздел
topic_facet Научно-технический раздел
url https://nasplib.isofts.kiev.ua/handle/123456789/47022
work_keys_str_mv AT laboum solutionofthefirstpassageproblembyadvancedmontecarlosimulationtechnique
AT laboum rešeniezadačipervogovybrosamodificirovannymčislennymmetodommontekarlo