4. Relevance of stochastic viability in time-variant reliability
The relevance of stochastic viability to solve design and main- tenance problems in time-variant reliability is now demonstrated. We use closed-loop feedbacks, so that the control u at date t also depends on information about the system. We will explore how this changes the general design and maintenance problem of
Section 3.2, first in the full information case in Section 4.1, and then in the more general partial information case in Section 4.2. Following that, Section 4.3 presents stochastic viability and the associated stochastic viability kernel, which it relates to the reliability kernel of the problem of Section 4.2. This relationship enables the introduction of dynamic programming to solve that problem, as detailed in Section 4.4. This method also leads to approximations of the outcrossing rate as in Section 4.5.
4.1. Closed-loop feedbacks with full information
Full information means that at date t, we have access to complete knowledge of the state xðt;πÞ, which is a realization of the randomvariable Xðt;πÞ. The existence of closed-loop feedbacks means that the choice of u at date t depends on the state xðt;πÞ. Within a closed-loop formulation, a strategy uðÞ (as introduced in Section 3.2) associates a maintenance decision uðt;xðt;πÞÞ to each date t and state y. Since we are working with realizations rather than with the random vector Xðt;πÞ itself, Eq. (11) becomes xðtþ1;πÞ¼fðt;π;xðt;πÞ;uðt;xðt;πÞÞ;wðtÞÞ ð15Þ where w(t) represents the realization of WðtÞ in Eq. (15). It represents the randomness in updating the state from date t to tþ1, and we also call scenario the sequence wðÞ¼ðwð0Þ; wð1Þ;…;wðT1ÞÞ. Let us now introduce yðtÞ¼ðxðt;πÞ;πÞ, a vector which aggre- gates the state and design vectors. The framework of so-called stochastic viability theory3 [34,1] focuses on the dynamic of y(t) instead of that of xðt;πÞ. The dynamic (15) becomes yðtþ1Þ¼fðt;yðtÞ;uðt;yðtÞÞ;wðtÞÞ ð16Þ Stochastic viability then calls y(t) the state vector, and closed-loop feedbacks are determined by y(t). Yet, if neither x nor f depend on the design, setting yðtÞ¼xðtÞ puts Eq. (15) under the form of Eq. (16) (see Section 5).
4.2. Closed-loop feedback with partial information In many cases, there is no direct access to the realization xðt;πÞ. In this paper, we assume that this partial information is a realization zðt;πÞ of a known random variable Zðxðt;πÞ;πÞ. Under this assumption, the dynamics of this realization zðt;πÞ can be deduced from that of Xðt;πÞ: zðtþ1;πÞ¼fðt;π;zðt;πÞ;uðt;zðt;πÞÞ;wðtÞÞ ð17Þ where u depends on the vector zðt;πÞ because we still are in the closed-loop feedback case. The full information case of Section 4.1 corresponds to the case P½Zðxðt;πÞ;πÞ¼xðt;πÞ¼1, Then, setting yðtÞ¼ðzðt;πÞ;πÞ yields the same equation as (16): yðtþ1Þ¼fðt;yðtÞ;uðt;yðtÞÞ;wðtÞÞ ð18Þ where y(t) is again called the state of the system from a stochastic viability perspective. Yet again, if z and f do not explicitly depend on π, then using yðtÞ¼zðtÞ turns equation (17) into (16). Working with the dynamics of Eq. (16) only makes sense if the knowledge of yðtÞ¼ðzðt;πÞ;πÞ helps in assessing the reliability of the system. Therefore, in the remainder of this article, we also assume that it is possible to compute the conditional probability PðXðt;πÞASðtÞjyðtÞÞ. This assumption holds in the full information case because then, PðXðt;πÞASðtÞjyðtÞÞ¼1 ifyðtÞ¼ðxðt;πÞ;πÞ where xðt;πÞ is the realization of Xðt;πÞ, and 0 otherwise
4. Relevance of stochastic viability in time-variant reliability
The relevance of stochastic viability to solve design and main- tenance problems in time-variant reliability is now demonstrated. We use closed-loop feedbacks, so that the control u at date t also depends on information about the system. We will explore how this changes the general design and maintenance problem of
Section 3.2, first in the full information case in Section 4.1, and then in the more general partial information case in Section 4.2. Following that, Section 4.3 presents stochastic viability and the associated stochastic viability kernel, which it relates to the reliability kernel of the problem of Section 4.2. This relationship enables the introduction of dynamic programming to solve that problem, as detailed in Section 4.4. This method also leads to approximations of the outcrossing rate as in Section 4.5.
4.1. Closed-loop feedbacks with full information
Full information means that at date t, we have access to complete knowledge of the state xðt;πÞ, which is a realization of the randomvariable Xðt;πÞ. The existence of closed-loop feedbacks means that the choice of u at date t depends on the state xðt;πÞ. Within a closed-loop formulation, a strategy uðÞ (as introduced in Section 3.2) associates a maintenance decision uðt;xðt;πÞÞ to each date t and state y. Since we are working with realizations rather than with the random vector Xðt;πÞ itself, Eq. (11) becomes xðtþ1;πÞ¼fðt;π;xðt;πÞ;uðt;xðt;πÞÞ;wðtÞÞ ð15Þ where w(t) represents the realization of WðtÞ in Eq. (15). It represents the randomness in updating the state from date t to tþ1, and we also call scenario the sequence wðÞ¼ðwð0Þ; wð1Þ;…;wðT1ÞÞ. Let us now introduce yðtÞ¼ðxðt;πÞ;πÞ, a vector which aggre- gates the state and design vectors. The framework of so-called stochastic viability theory3 [34,1] focuses on the dynamic of y(t) instead of that of xðt;πÞ. The dynamic (15) becomes yðtþ1Þ¼fðt;yðtÞ;uðt;yðtÞÞ;wðtÞÞ ð16Þ Stochastic viability then calls y(t) the state vector, and closed-loop feedbacks are determined by y(t). Yet, if neither x nor f depend on the design, setting yðtÞ¼xðtÞ puts Eq. (15) under the form of Eq. (16) (see Section 5).
4.2. Closed-loop feedback with partial information In many cases, there is no direct access to the realization xðt;πÞ. In this paper, we assume that this partial information is a realization zðt;πÞ of a known random variable Zðxðt;πÞ;πÞ. Under this assumption, the dynamics of this realization zðt;πÞ can be deduced from that of Xðt;πÞ: zðtþ1;πÞ¼fðt;π;zðt;πÞ;uðt;zðt;πÞÞ;wðtÞÞ ð17Þ where u depends on the vector zðt;πÞ because we still are in the closed-loop feedback case. The full information case of Section 4.1 corresponds to the case P½Zðxðt;πÞ;πÞ¼xðt;πÞ¼1, Then, setting yðtÞ¼ðzðt;πÞ;πÞ yields the same equation as (16): yðtþ1Þ¼fðt;yðtÞ;uðt;yðtÞÞ;wðtÞÞ ð18Þ where y(t) is again called the state of the system from a stochastic viability perspective. Yet again, if z and f do not explicitly depend on π, then using yðtÞ¼zðtÞ turns equation (17) into (16). Working with the dynamics of Eq. (16) only makes sense if the knowledge of yðtÞ¼ðzðt;πÞ;πÞ helps in assessing the reliability of the system. Therefore, in the remainder of this article, we also assume that it is possible to compute the conditional probability PðXðt;πÞASðtÞjyðtÞÞ. This assumption holds in the full information case because then, PðXðt;πÞASðtÞjyðtÞÞ¼1 ifyðtÞ¼ðxðt;πÞ;πÞ where xðt;πÞ is the realization of Xðt;πÞ, and 0 otherwise
การแปล กรุณารอสักครู่..
