Stochastic Modeling of Deterioration Processes through Dynamic Bayesian Networks
read more
Citations
Bayesian Updating with Structural Reliability Methods
Application of Bayesian Networks in Reliability Evaluation
Some Challenges and Opportunities in Reliability Engineering
Bayesian Network Enhanced with Structural Reliability Methods: Methodology
Bayesian Methods for Updating Dynamic Models
References
Artificial Intelligence: A Modern Approach
Markov Chain Monte Carlo in Practice
Bayesian networks and decision graphs
Dynamic bayesian networks: representation, inference and learning
Structural Reliability Methods
Related Papers (5)
Frequently Asked Questions (15)
Q2. What are the future works mentioned in the paper "Stochastic modeling of deterioration processes through dynamic bayesian networks" ?
Future work should aim at extending the presented framework to consider multiple structural elements. Depending on the type of dependence among the system elements, different models can be envisaged and should be investigated.
Q3. How many cycles are required to perform inspections of the structural element?
Inspections of the structural element are carried out in intervals of 106 cycles and it is assumed that all inspections result in no-indication (i.e., no defect is found).
Q4. What is the reliability index after n cycles?
For the case including inspection results, the authors apply the filtering inference algorithm, i.e., the reliability index after n cycles is computed by consideration of all inspection results up to n cycles, but neglecting later inspection results.
Q5. How long does it take to compute the results?
Straub (2009): Stochastic modeling of deterioration processes through DBN 21/32With the DBN established, the computation of the results presented here takes in the order of 10 CPU seconds on a standard PC with a 2.0 GHz processor with a Matlab-based program.
Q6. What is the common type of deterioration model?
Straub (2009): Stochastic modeling of deterioration processes through DBN 10/32Continuous versus finite state space modelsMost variables in deterioration models are defined in a continuous space.
Q7. What is the PMF of a set of random variables X?
BN are probabilistic models based on directed acyclic graphs that represent ( )p x , the joint probability mass function (PMF) of a set of random variables X .
Q8. what is the probability function for tp dz z?
The likelihood function is 1( ,..., | , , )t T t t tp dz z θ ω , the result of the backward operation, because of independence of 1,...,t Tz z from 1,..., tz z for given , ,t t tdθ ω , as prescribed by the DBN structure.
Q9. what is the crack size after n cycles?
Straub (2009): Stochastic modeling of deterioration processes through DBN 22/32 d ( ) , d cg a n Y n g a n n θ (19)( )a n is the crack size after n cycles and [ ]cgg can be any deterministic crack growth law as a function of ( )a n and of time-invariant parameters θ .
Q10. What is the conditional PMF of X?
The conditional PMF of X̂ isStraub (2009): Stochastic modeling of deterioration processes through DBN 13/32 ( ) ( ) ( ) ( )ˆ k l l lP X k P X k Pp x F x F x x x x (9)wherein XF is the cumulative distribution function (CDF) of X , which is conditional on ( )lP PX x , and kx and kx are the lower and upper boundaries of the interval corresponding to state k .
Q11. what is the dbn function for predicting and smoothing?
As demonstrated in the Appendix, the computation time for filtering is 2 2[( ) ]d dO m m m m m tω ω θ , whereas it is 2 2[( ) ]d dO m m m m m Tω ω θ for predicting and smoothing.
Q12. What is the probability of a variable being outside the range of values?
The probable range of values is defined so that the a-priori probability of a variable being outside that range is smaller than p for all time slices t .
Q13. What is the difference between the DBN results for small numbers of cycles?
The deviation of the DBN result from the second-order results for small numbers of cycles is mainly due to the approximation in the discretization of tY , as found from additional numerical investigations, which are not reported here for brevity.
Q14. How many cycles can be used to learn about the correlation length?
it is conceivable that learning about the correlation length is possible by collecting measurement data from a larger number of specimens, in particular in combination with other information, e.g., on the initial crack depth.
Q15. what is the function of the boundary condition 0( 0)a n a?
With the boundary condition 0( 0)a n a , this differential equation can be solved for the crack size as a function of the number of cycles n , (Ditlevsen and Madsen 1996): 11 21 22 01 2m mm mma n C S n a (15)The event of failure is described by the limit state function g as a function of ( )a n and the critical crack length ca :Straub (2009): Stochastic modeling of deterioration processes through DBN 16/32 cg a a n (16)The performance of the structural element is represented through the binary variable E .