121 Downloads (Pure)

Resumé

A Bayesian approach is often applied when updating a deterioration model using observations from expected structural health monitoring or condition monitoring. Usually, observations are assumed to be independent conditioned on the damage size, but this assumption does not always hold, especially when using online monitoring systems. Frequent updating using correlated measurements can lead to an over-optimistic assessment of the value of information when the measurements are incorrectly modeled as independent. This paper presents a Bayesian network modeling approach for inclusion of the correlation between measurements through a location parameter and presents a generic monitoring model based on exceedance of thresholds for a damage indicator. Additionally, the model is implemented in a computational framework for risk-based maintenance planning, developed for maintenance planning for wind turbines. A generic example and a parameter study show that neglecting correlation in the decision model when the observations are in fact correlated, can lead to much higher costs than expected and to the selection of non-optimal strategies; much lower costs can be obtained when the correlation is properly modeled. In case of correlated observations, an advanced decision model using all past observations for decision making is needed to make monitoring feasible compared to only using inspections.
OriginalsprogEngelsk
TidsskriftReliability Engineering & System Safety
ISSN0951-8320
StatusAfsendt - 2019

Fingerprint

Structural health monitoring
Decision making
Monitoring
Planning
Condition monitoring
Bayesian networks
Wind turbines
Deterioration
Costs
Inspection

Citer dette

@article{b052657fd24243b481568e83009039d1,
title = "Bayesian updating and decision making using correlated structural health monitoring observations",
abstract = "A Bayesian approach is often applied when updating a deterioration model using observations from expected structural health monitoring or condition monitoring. Usually, observations are assumed to be independent conditioned on the damage size, but this assumption does not always hold, especially when using online monitoring systems. Frequent updating using correlated measurements can lead to an over-optimistic assessment of the value of information when the measurements are incorrectly modeled as independent. This paper presents a Bayesian network modeling approach for inclusion of the correlation between measurements through a location parameter and presents a generic monitoring model based on exceedance of thresholds for a damage indicator. Additionally, the model is implemented in a computational framework for risk-based maintenance planning, developed for maintenance planning for wind turbines. A generic example and a parameter study show that neglecting correlation in the decision model when the observations are in fact correlated, can lead to much higher costs than expected and to the selection of non-optimal strategies; much lower costs can be obtained when the correlation is properly modeled. In case of correlated observations, an advanced decision model using all past observations for decision making is needed to make monitoring feasible compared to only using inspections.",
author = "Nielsen, {Jannie S{\o}nderk{\ae}r}",
year = "2019",
language = "English",
journal = "Reliability Engineering & System Safety",
issn = "0951-8320",
publisher = "Pergamon Press",

}

TY - JOUR

T1 - Bayesian updating and decision making using correlated structural health monitoring observations

AU - Nielsen, Jannie Sønderkær

PY - 2019

Y1 - 2019

N2 - A Bayesian approach is often applied when updating a deterioration model using observations from expected structural health monitoring or condition monitoring. Usually, observations are assumed to be independent conditioned on the damage size, but this assumption does not always hold, especially when using online monitoring systems. Frequent updating using correlated measurements can lead to an over-optimistic assessment of the value of information when the measurements are incorrectly modeled as independent. This paper presents a Bayesian network modeling approach for inclusion of the correlation between measurements through a location parameter and presents a generic monitoring model based on exceedance of thresholds for a damage indicator. Additionally, the model is implemented in a computational framework for risk-based maintenance planning, developed for maintenance planning for wind turbines. A generic example and a parameter study show that neglecting correlation in the decision model when the observations are in fact correlated, can lead to much higher costs than expected and to the selection of non-optimal strategies; much lower costs can be obtained when the correlation is properly modeled. In case of correlated observations, an advanced decision model using all past observations for decision making is needed to make monitoring feasible compared to only using inspections.

AB - A Bayesian approach is often applied when updating a deterioration model using observations from expected structural health monitoring or condition monitoring. Usually, observations are assumed to be independent conditioned on the damage size, but this assumption does not always hold, especially when using online monitoring systems. Frequent updating using correlated measurements can lead to an over-optimistic assessment of the value of information when the measurements are incorrectly modeled as independent. This paper presents a Bayesian network modeling approach for inclusion of the correlation between measurements through a location parameter and presents a generic monitoring model based on exceedance of thresholds for a damage indicator. Additionally, the model is implemented in a computational framework for risk-based maintenance planning, developed for maintenance planning for wind turbines. A generic example and a parameter study show that neglecting correlation in the decision model when the observations are in fact correlated, can lead to much higher costs than expected and to the selection of non-optimal strategies; much lower costs can be obtained when the correlation is properly modeled. In case of correlated observations, an advanced decision model using all past observations for decision making is needed to make monitoring feasible compared to only using inspections.

M3 - Journal article

JO - Reliability Engineering & System Safety

JF - Reliability Engineering & System Safety

SN - 0951-8320

ER -