Sei sulla pagina 1di 12

Structural Safety 81 (2019) 101874

Contents lists available at ScienceDirect

Structural Safety
journal homepage: www.elsevier.com/locate/strusafe

Decision support framework for bridge condition assessments T


a,⁎ a b c
Ivar Björnsson , Oskar Larsson Ivanov , Dániel Honfi , John Leander
a
Division of Structural Engineering, Lund University, Lund, Sweden
b
Mechanics Research, RISE Research Institutes of Sweden, Sweden
c
Division of Structural Engineering and Bridges, KTH Royal Institute of Technology, Stockholm, Sweden

ARTICLE INFO ABSTRACT

Keywords: An essential aspect in the maintenance of existing bridges is the ability to adequately and accurately assess and
Bridge operation and maintenance evaluate the condition of the structure. Condition assessments, which can be carried out in any number of ways,
Condition assessment provide valuable information concerning the actual state of a bridge, including the severity of potential damages,
Decision support and form the basis for further maintenance decisions. Any decision support concerning the management of
Bayesian decision theory
existing structures thus requires attention towards the uncertainties associated with the assessment methods
Framework
when applied in practice as well as the maintenance actions these support. These uncertainties cannot be solely
described as model uncertainties but are also a result of the variation in engineering performance observed in
practice. In the current paper a rational and systematic framework is presented which provides practical decision
support concerning whether condition assessments are necessary, what assessment methods are recommended, if
invasive actions are needed, or if some other non-invasive option may be more appropriate. The framework takes
into account three main attributes of an enhanced condition assessment, namely, modelling sophistication,
considerations of uncertainties and risks, and knowledge/information content. Increasing the level of one or
more of these attributes may be advantageous only if the expected benefits or added value of information is
considered appropriate in relation to the cost of implementation in practice. A decision making model, based on
Bayesian decision theory, is adopted to evaluate this problem. Two case studies, in which the framework is
applied, are provided for illustrative purposes; the first is a generic numerical example and the second a decision
scenario related to the fatigue assessment of an existing railway bridge.

1. Introduction experience and traditions of the assessing engineers, etc. There is cur-
rently a lack of a systematic framework for identifying which methods
A substantial portion of the bridge stock in European countries is are suitable in specific cases. Without such a framework being im-
relatively old and requires regular maintenance to ensure adequate plemented, decisions are made intuitively rather than based on con-
structural safety and serviceability. In Sweden, for example, there are sistent rationale and can be seen more ad hoc than informed.
approximately 32 thousand bridges with an average age of about In the current paper, a practical decision support framework is
50 years [1]. In practice, the maintenance and operation of bridges is presented which is intended to provide decision support to bridge
carried out by a responsible authority (bridge manager) with the aid of managers concerning condition assessments of existing bridges. The
a bridge management system (BMS) [2]. In cases where doubts about framework, which is presented in more detail in later sections, is in-
the state of a bridge have been raised, a condition assessment is often a tended to provide guidance concerning whether and what additional
first step in determining whether more costly repair actions are ne- assessments may be necessary, if invasive actions are needed, or if some
cessary; see, e.g., ISO 13822 [3]. In Sweden, for instance, it is common other non-invasive option may be more appropriate. The adopted de-
for external consultants to be called upon to evaluate the condition of cision model is based on Bayesian decision theory and was purposefully
the structure and aid in determining appropriate plans of action for chosen for its simplicity and ease of use. The framework considers three
future maintenance and inspection activities. In practice, condition major attributes of a condition assessment, namely:
assessments can be carried out in any number of ways and their suit-
ability in a specific case may depend on a number of factors including 1) modelling sophistication,
bridge importance, damage severity, implementation costs, the 2) considerations of risk and/or uncertainty, and


Corresponding author at: Box 118, SE-221 00 Lund, Sweden.
E-mail address: ivar.bjornsson@kstr.lth.se (I. Björnsson).

https://doi.org/10.1016/j.strusafe.2019.101874
Received 31 January 2018; Received in revised form 25 March 2019; Accepted 3 July 2019
Available online 02 August 2019
0167-4730/ © 2019 Elsevier Ltd. All rights reserved.
I. Björnsson, et al. Structural Safety 81 (2019) 101874

a)

4th generation
Optimised inspection
and maintenance
Increased expected
3rd generation utility during service
Higher availability and life
reliability Enhanced
Greater safety and better quality robustness, resilience
2nd generation
Extended service life and sustainability
Higher availability
1st generation Greater cost effectivness
Longer life
Fix it when it Low environmental impact
Lower costs
broke
1940 1950 1960 1970 1980 1990 2000 2010 2020

b)

4th generation
Risk-based
inspection planning
Bayesian decision
3rd generation analysis
Condition monitoring VoI
Design for reliability and Cluster and cloud
maintanability computing
2nd generation Hazard studies Wireless, smart
Scheduled overhauls Small, fast computers sensors
1st generation System for planning and FMEA IoT
Fix it when it controlling work Expert systems
broke Big slow computers Multiskilling and teamwork

1940 1950 1960 1970 1980 1990 2000 2010 2020


Fig. 1. Evolution of (a) maintenance strategies and (b) techniques (extended from [10]

3) knowledge content. process also includes decisions on how to collect and analyse new in-
formation.
These attributes, which are explained more in depth in later sec- There are a number of existing research efforts concerning methods
tions, are intrinsic to any condition assessment and improving one or for improving the management and operation of bridges and other in-
more can be seen as a strategy for improved condition assessment; a frastructure. A common aim is to integrate relevant maintenance de-
strategy which may involve high direct costs in some cases (e.g. pur- cisions impacting life-cycle performance in a common framework with
chasing new equipment, investing into a structural health monitoring a probabilistic consideration of uncertainties related to loads, re-
system, involving experts, etc). Decisions concerning the management sistances, deterioration mechanisms, outcomes of the decisions etc.
of existing structures are often founded on the results of a condition Such frameworks enable the reliability- or risk-based optimisation of
assessment and the suitability of these results for informed and rational maintenance strategies considering various objectives such as e.g.
decision making is important to consider. The framework presented in minimising unavailability, maximising redundancy and minimising life-
this paper aims to address this issue in a systematic way. The novelty of cycle costs [5]. So far, most efforts in this field have been towards
the presented approach is that it integrates the evaluation of different developing or implementing more advanced computational models for
enhanced assessment choices of bridges with decisions concerning in- predicting or assessing degradation and deterioration, identifying and
terventions; thus providing practical decision support for current calculating structural-performance indicators, evaluating life-cycle
practice bridge maintenance. costs as well as optimizing inspection, maintenance and repair strate-
gies; a review of some of these approaches can be found in Sánchez-
Silva et al. [6]. Pozzi & Der Kiureghian [7] presented a unified frame-
1.1. Decisions and maintenance work where updating of information through monitoring and decision
making are treated together using VoI analysis. The concept is illu-
A general definition of maintenance is given in EN 13306:2010 [4] strated with a simple example and numerical application for long-term
as the “combination of all technical, administrative and managerial monitoring is provided by the authors. Frangopol et al. [8] present a
actions during the life cycle of an item intended to retain it in, or restore generalised framework for an integrated life-cycle management of
it to, a state in which it can perform the required function”. It is clear bridges and provide an overview of some related developments in-
that determining appropriate maintenance actions is a multi-faceted cluding computational methods for life-cycle costs, risk quantification,
decision-making problem; i.e. a number of decisions should be made on sustainability metrics, hazard and deterioration modelling, as well as
which actions are appropriate, how they should be carried out, at which probabilistic (Bayesian) updating using structrual health monitoring
point in time they should be implemented, who should be responsible, (SHM) data. However, practical implementation of these generalised
and the amount of resources that should be spent on them. To optimise frameworks is not straightforward for bridge operators. Furthermore, as
such decisions, they need to be supported by information concerning pointed out by Thoft-Christensen [9], advanced modelling and
the actual state of the given asset. Therefore, the decision-making

2
I. Björnsson, et al. Structural Safety 81 (2019) 101874

computational approaches cannot be useful without using expert Requests/Needs


knowledge, since it is not possible to express all previously accumulated Specification of objectives
knowledge in mathematical terms. Thus, the role of the bridge man- Definition of Scenarios
agers is crucial when making decisions concerning condition assess-
ments, at least in the current bridge management systems. Preliminary assessment
In general, the actual decision-making process concerning main-
tenance can vary in practice depending on, e.g., industry standards, the
No
operating organisation or the expectations of the public. During the last Detailed assessment?
decades, a significant development has been made on available strate-
Yes
gies and techniques to support maintenance decisions. These im-
provements can be divided into successive maintenance generations Detailed assessment
Detailed documentary search and review
according to Fig. 1. The first three generations have been previously Detailed inspection and material testing
defined by Moubray [10]: corrective (or breakdown) maintenance, Determination of actions
preventive maintenance, and condition-based (or predictive) main- Determination of propoerties of the structure
tenance. However, since the 2000s, a 4th generation of approaches Structural analysis
Verification
founded on risk and linear utility theory has started to emerge in some
industries; especially where the maintenance costs are relatively high
such as in the oil and gas or off-shore industries (see [11,12]). Further inspection?
Yes
As mentioned in Honfi et al. [13], the focus of the 4th generation
approaches is to maximize the benefits provided during the service life No
of the assets, optimise the inspection and maintenance activities to Reporting results of assessment
reduce negative impacts to the environment and the society, as well as Judgement and decision
enhance the robustness and resilience of interconnected and inter-
dependent infrastructure while considering sustainability aspects. Yes
Sufficent reliability?
Techniques utilised to achieve these goals include multi-objective op-
timisation, risk-based inspection planning, Bayesian decision analysis, No
Value of Information (VoI) theory, high performance computing infra- Intervention
structures (cluster and cloud computing), wireless and smart sensors, as
well as intelligent infrastructure as part of the Internet of Things (IoT)
(see, e.g., [14–17]). Fig. 2. General assessment procedure from ISO 13822 for existing structure [3].

1.2. Condition assessment of bridges and different solution paths to the same problem or task; i.e. the pro-
blems ill-defined [25]. Accordingly, the task of condition assessments of
To preserve the serviceability, safety and structural integrity of existing bridge structures can be carried out in any number of ways and
existing bridges, knowledge about their actual condition is needed. In there is a lack of a systematic framework for identifying which ap-
fact, it is standard practice to inspect and, in special cases, monitor proaches are suitable in specific cases. Table 1 shows some examples of
existing bridges on a regular basis as a means of obtaining indicators procedural frameworks for the condition assessment of bridges along
relating to structural performance and possible damages. In cases of with the characteristics of the various levels of the assessment. It is clear
doubt, the condition of the structure can be re-evaluated based on this from the table that several options are available to increase the level of
added information and thus provide the basis for determining further assessment; it is not clear, however, under what circumstances one
maintenance actions. In some cases, enhanced assessments may be re- should move to a higher level of the assessment or opt for intervention.
quired to determine the condition of the structure and these assess- In addition, the approaches used and results obtained by different fra-
ments, which may be carried out by third party consultants, can be meworks may not be directly comparable and the decision maker must
costly. In Sweden, e.g., the Transportation Administration has its own make some preference as to which one should be adopted without any
Bridge and Tunnel Management system for nationally owned infra- rational decision support. Any decision made concerning condition
structure which contains information concerning the condition of an assessment is thus highly dependent on the framework which is chosen.
asset based on past inspection, maintenance and repair/strengthening As such, there is a risk that sub-optimal decision could be made in the
activities [18]. While there is substantial guidance concerning the absence of further guidance on how to choose between the possible
collection of information (i.e. inspections), there is a lack of decision alternatives at the same or different levels of assessment. Therefore, it is
support concerning how this information should be utilized. desirable to develop a general framework which provides decision
Several frameworks for the general assessment procedure of existing support concerning which level of assessment is appropriate in specific
bridge structures have been developed in various research projects (e.g. cases. This framework should account for the costs and benefits of
[2,19]). These are usually based on the procedure proposed by moving to higher levels of assessment while providing various options
Schneider [20] which has been adopted by the JCSS, RILEM [21], and on how to do this in a transparent and structured manner.
more recently by ISO 13822 [3]; see Fig. 2. There are also a number of In practice it may be desirable for bridge managers to utilize lower
existing frameworks which outline a procedure for carrying out detailed levels of assessment owing to constraints in resources (costs), lack of
condition assessments of existing bridges; see, e.g., Table 1. These reliable data (e.g. based on SHM or other on-site measurement tech-
procedures are often subdivided into successive levels of assessment niques), or lack of expert knowledge. This issue is further compounded
with higher levels intended to provide improved accuracy (i.e. they are considering the relatively high variability which has been observed in
less conservative) while the resources needed for implementing them in engineering performance in practice [26–29]. Klasson et al. [30] also
practice may be substantially increased. Thus, in the context of decision observed a relatively large variation amongst experienced practicing
making, realizing an appropriate level of assessment essentially entails engineers concerning modeling decisions made during design. Deci-
a trade-off between perceived accuracy and resources spent. There are, sions concerning a suitable level of modelling may in such cases be
however, few guidelines concerning which level is appropriate in specific guided by the so-called principle of consistent crudeness in which the
instances. quality of the output of any model cannot be greater than the quality of
In engineering decision making, there are often multiple solutions

3
I. Björnsson, et al. Structural Safety 81 (2019) 101874

the crudest (or most uncertain) input [31,32]. Where there is a lack of
reliable information or an otherwise high degree of uncertainty, a

(data from tests, monitoring,


Full probabilistic assessment

Full probabilistic assessment


compromise should be made concerning the appropriate level of mod-
elling (sophistication and complexity). In such cases there is a risk that
increasing the level of modelling could yield poorly founded and mis-
leading results. A methodology for selecting appropriate methods
considering these types of uncertainties is practically valuable for the
bridge manager.
etc.).
5

2. Decision theory

3 years and simulations, special software for


Adaptation of target reliability measures
and assessment with modified structure-

In the development of the proposed framework, established con-


Modification of partial factors, material

Lifetime prediction (data from at least


properties and loads based on in-situ

cepts of Bayesian decision theory with prior, posterior and pre-posterior


decision analysis are utilized [33]. These approaches are based on the
well-established linear utility theory and the expected value criterion
for decision making first developed by von Neumann and Morgenstern
[34]. This section provides an overview of these approaches in the
context of bridge condition assessment considering discrete state vari-
decision support)
specific values.

ables; for an overview of decision making in the continuous domain see,


observations

e.g., Faber [35]. For illustrative purposes decision trees and influence
diagrams are also presented.
There are three to five different aspects considered in the afore-
4

mentioned decision analysis approaches, see Fig. 3 for an overview


Detail assessment and rating (includes
Assessment using refined model based

probabilistic analysis, 3-D frame, FEM

using a decision tree representation [33]. A prior analysis involves


methods (data from tests, monitoring,
properties and loads based on in-situ

determining an optimal action, a, based on information concerning


analytical model representing the
Partial factor method, material

possible bridge states, θ, and considering the utilities, u, associated with


each action and state pair. The action, a, is a decision to intervene and
a) Probabilistic approach

b) + Bayesian updating

adjust the state of the structure (e.g. repair, strengthening or replace-


ment) while the state, θ, refers to the condition of the bridge (e.g. da-
maged or unserviceable). Generally, the ‘true state’ of the structure is
observations

uncertain; i.e. it is not known with certainty. The purpose of the deci-
assessment:

structure)
Enhanced

sion model is therefore to provide guidance which both accounts for


uncertainties as well as the expected benefits/costs associated with
etc.)

various decision alternatives or actions. The state of a structure could


3

be discretely determined as belonging to the category safe or in some


e.g. partial safety factor format, 2 or 3-D

than the previous level, more indicators)


using simple model based methods (data

a) Non-linear analysis with global safety

predefined state of damage ranging from limited or localized damage or


b) + updated loads from measurements
Performance assessment (more detailed
resistances from records and standards,

Assessment of safety and serviceability

functional impairment to extensive failure. In practice, condition


Partial factors method, loads and

classes are typically used to guide bridge inspections. In Sweden, for


example, a condition class (CC) from 0 to 3 is used to describe the
physical and functional condition of bridge elements; the CC, which is
Intermediate assessment:
refined structural model

determined by a bridge inspector, indicates in which time frame a de-


fective function is expected and thus forms the basis for follow-up
from documents)

maintenance actions [18].


An important aspect of decision analysis is the utility function, u,
which reflects the decision makers’ preferences over the possible out-
factor
frame

comes (actions and states), where a higher utility reflects preference for
Some existing frameworks for the condition assessment of bridges.

one possible outcome over another with lower utility. The value of the
from records and standards, simple structural

Condition assessment (simple instrumentation


Partial factors method, loads and resistances

utility function for different outcomes may include benefits or rewards


from measurements (no structural analysis)
Direct assessment of serviceability values

as well as costs or drawbacks. In the context of bridge management, it


b) + updated loads from measurements
e.g. deterministic or semi-probabilistic

may be difficult to directly determine the benefit of a structure while


the consequences of various damage states can be more readily de-
termined and compared (see, e.g., [36,37]). As the decision problem is
and simple decision support)

uncertain, the expected value, E(.), for the utility is a common measure
for determining a decision maker’s preference between different deci-
formats, 2-D frame

sion alternatives [33,34]; see, e.g., Eqs. (1) and (2).


a) Linear analysis
Assessment level

In a posterior decision analysis, additional information concerning


assessment:

the ‘true state’ of the structure has been obtained using some method e;
the outcome of this method is denoted z. It should be noted that e
model

Initial

traditionally refers to an “experiment”; however, the term “method” (of


1

obtaining improved knowledge/information) to express that it should


be understood in a more generic way. The probabilities associated with
Sustainable Bridges

Skokandic et al.

the states of the bridge are then updated considering this result and the
SAMCO [22]

Wenzel [23]

decision problem re-evaluated based on the new information; i.e. using


BRIME [2]

[19]

[24]

posterior probabilities for θ. This requires that the uncertainties asso-


Table 1

ciated with the outcomes of the methods are accounted for (i.e. the so-
Ref.

called sample likelihoods). In a pre-posterior analysis, the decision to

4
I. Björnsson, et al. Structural Safety 81 (2019) 101874

...

PREPOSTERIOR (choice of methods for updating knowledge)


u(e0,z0,a2,θ1)
a1 , θ1
th ing, safe
o no ge

PRIOR
D Brid
Repair/strengthen, a 2 Bridge unserviceable, θ2
u(e0,z0,a2,θ2)
Rep Brid
la ce, a ge fa
ilure
3 , θ3

u(e0,z0,a2,θ3)
0

...
,e
ing

u(e2,z2,a2,θ1)
th

z1
ted,
no

e1 a1
od etec ing, afe,
θ1
Do

th mage d noth ge s
Me o da Do Brid

POSTERIOR (new information)


N
Damage type a, z 2 Repair/strengthen, a 2 Bridge unserviceable, θ2 u(e2,z2,a2,θ2)
d e2
Metho Dam Rep Brid
a ge t lace ge fa
ype , a3 ilure
b, z , θ3
3
u(e2,z2,a2,θ3)
M
et

...
ho
d
en

θi u(en,zk,aj,θi)
Action a State
m e zk
Outco j

...

Total Utilities
Method Method outcome Action State of nature
(consequences)
Examples: Examples:
Examples:
visual inspection, on- Examples: Bridge safe,
Examples: Cost of repair,
site testing or Do nothing, repair/ unserviceable, local
No crack detected, consequences of
monitoring, lab-testing, strengthen, replace, failure, global failure,
damage detected, failure, etc.
enhanced assessment demolish etc.
etc.

Fig. 3. Decision tree depicting prior, posterior and pre-posterior decision making for maintenance.

collect new information is a decision in itself. The method, e, in this


case refers to an active approach for obtaining new knowledge and can Utilities*,
range from visual inspections to laboratory testing of samples taken on- Action, a
u(a,θ,z)
site and may also include the choice of using more sophisticated
modeling techniques to obtain less conservative results. As the method
has not been carried out, all different possible outcomes, z, should be
considered. Posterior probabilities are determined in an analogous way
Method
as for the posterior analysis and the expected utilities associated with Method, e
outcome, z
each experiment determines the optimal choice. To reflect the trade-off
between added information and the efforts required for obtaining it, the
utility functions in a pre-posterior analysis should also include the costs
of each experiment (although this is not strictly necessary).
Another way of expressing the same decision problems as earlier is
through an influence diagram [38]; see Fig. 4. An influence diagram is a Method
State of nature, θ
costs, u(e)
Bayesian probabilistic network augmented with decision variables and
utility functions which can be used for solving sequential decision * excluding method cost
problems [39]. Traditional Bayesian networks, on the other hand, are Fig. 4. Influence diagram.
only used for reasoning under uncertainty, representing and processing
probabilistic knowledge without explicitly incorporating any decision
making. In an influence diagram, it is possible to have a graphical re- • Value nodes which contain the utilities (diamonds)
presentation of a decision problem involving a sequence of interleaved
decisions and observations, thus providing a decision model that is Directed paths between each node reflect dependence and (usually)
more accessible for the decision maker [40]. causality between nodes. An influence diagram can provide a solution
An influence diagram contains nodes related to the different aspects to all three of the classical Bayesian decision problems mentioned
of the decision problem (see Fig. 4) where the node types are: earlier. It is obvious from Fig. 4 that the influence diagram version of
the decision problem is a more compact representation compared to the

• Decision nodes corresponding to actions (rectangles) decision tree alternative from Fig. 3; although the number of inputs as

• Uncertainty nodes corresponding to uncertain states (ovals) well as the computational effort will remain relatively unchanged. One

5
I. Björnsson, et al. Structural Safety 81 (2019) 101874

main difference from the ID in Fig. 3 and the decision tree in Fig. 4 is Uncertainty and/or risk
considerations, U
that the utilities are divided such that the method costs, u(e), are se- Increase model sophistication
parate from the other utilities/consequences, u(a,θ,z). For both decision
trees and influence diagrams, the underlying expressions for calculating

Increased consideration of risks/


the resulting utilities are analogous and can be found in, e.g., Benjamin
and Cornell [33] and Kjærulff & Madsen [39]; see also Eqs. (1)–(3).

uncertainties
An influence diagram can also be used for decision problems where
more than one person’s decision affects the final utility [41]. This can
be useful related to bridge management in situations where multiple
stakeholders are involved in the process and where a decision taken by
one stakeholder may affect the subsequent decisions of others involved.
Modeling
3. Decision framework sophistication, M

In the current paper, a practical decision framework is presented for

e
dg
determining whether and which type of enhanced condition assessment

le
ow
should be adopted for evaluating the state of an existing bridge, if in-

kn
te
da
vasive (repair) actions are needed, or if some other non-invasive option

Up
may be more appropriate. A practical decision model based on Bayesian Knowledge
content, K
decision theory is used for evaluating the decision problem while un-
certainties associated with the application of different levels of condi- Fig. 5. Three attributes of an enhanced condition assessment and basic strate-
tion assessment methods in practice are discussed and evaluated using a gies for obtaining improved results.
graphical representation; both are described in more detail in the sec-
tions that follow. and/or risks are explicitly considered in the condition assessment and
can range from a deterministic calculation to a full probabilistic risk
3.1. Levels of condition assessment assessment; examples of three different levels are shown in Table 2. The
lowest level generally involves deterministic calculations in which case
In general, decisions relating to the management of existing struc- representative values for the variables are used and the results com-
tures are based on the result of some type of condition assessment; pared with prescribed criteria usually in the form of code specified
whether qualitative or quantitative. Assessing the condition of a safety targets. These targets, and the nominal values used in the cal-
structure as a basis for decision making can be done in a number of culations, may depend on the modeling sophistication; e.g. safety ver-
different ways and determining an appropriate assessment approach is ification formats for linear elastic first order analysis of components
in itself an important decision. From the perspective of a bridge man- may be unsuitable for non-linear analyses [43]. Uncertainties are,
ager, an approach maximizing the expected utility considering safety however, not directly considered and the safety of the structure is not
and societal cost should be sought. It is thus convenient to differentiate expressed explicitly, rather its compliance to a somewhat arbitrary
between these approaches by considering specific attributes associated criterion. Going from a deterministic calculation to, e.g., a reliability
with them and their application in practice. The following three attri- based one will require a stochastic modeling approach involving an
butes, which can be seen as characterizing the methods (e) from Figs. 3 explicit consideration of epistemic, aleatory and model uncertainties.
& 4, are considered in this paper: Usually those parameters which are considered most significant are
modeled as random variables, with certain statistical moments and
1. Modeling sophistication (performance model) distributions, while the remaining variables are modeled as being de-
2. Uncertainty and/or risk consideration terministic. The amount of resources (in terms of time and effort)
3. Knowledge/information content needed to carry out these investigations will be increased compared to a
deterministic calculation. The application of a higher-level approach,
In the current framework it was considered suitable to represent the such as risk-based approaches, will require consideration of the costs
MUK (Modeling sophistication, Uncertainty consideration, Knowledge and consequences associated with identified damage and/or failure
content) triplet as a cube in order to provide a visual representation for scenarios; or, more generally, all the costs and benefits caused by the
the relation between the three attributes of a condition assessment, see dysfunction and functioning of the bridge.
Fig. 5. Furthermore, this representation, which was first introduced by The third axis in Fig. 5, the knowledge or information content,
Honfi et al. [42] and is further developed in this paper, also helps to prescribes the degree to which additional (updated) knowledge is in-
visualize strategies for improving the enhanced assessment for an ex- cluded in the assessment. The added information could be related to,
isting structure; i.e. by moving along one or more of the axes. An as- e.g., material properties, degradation phenomena, dimensions, re-
sessment aiming for the highest accuracy will require advancement inforcement content, etc.; but may also come from the knowledge of the
along all axes and utilize models that can represent all attributes. experts called on to guide the assessment. This type of information will
The first attribute, the modeling sophistication, is a measure of how generally provide a more accurate depiction of the actual state of the
encompassing the performance model is and could generally be related structure and thus do away with potentially unneeded conservative
to the model complexity; i.e. more sophisticated models usually contain modeling assumptions. The exact manner with which this additional
more variables/factors. The performance model is a model for which information can affect the assessment may depend on the level of risk/
quantitative results can be determined pertaining to the condition uncertainty considerations as well as the modeling sophistication. For
(often structural) of the bridge. Less sophisticated models may also be example, in a deterministic assessment it may alter the value of some of
more conservative while more sophisticated models are more refined the modeling parameters while for reliability-based assessments the
but will require more input. One factor that is important to consider is information may be directly integrated using Bayesian updating; e.g.
whether a reduced modeling uncertainty could be offset by com- posterior probabilities or distribution functions. Moving along this axis
pounded parameter uncertainty. Some examples for the different levels should provide added accuracy but may not always be cost-effective.
of modeling sophistication are shown in Table 2. Table 2 shows three possible levels of assessment associated with the
The second attribute in the MUK cube relates to how uncertainties

6
I. Björnsson, et al. Structural Safety 81 (2019) 101874

Table 2
Example showing assessment levels for different attributes according to the MUK cube.
Model sophistication
M1: Rule of thumb formulas, linear elastic 1st order M2: Elasto-plastic 1st order analysis, linear elastic 2nd M3: Non-linear geometrical and material analysis.
analysis. order analysis.
Uncertainty consideration
U1: Deterministic code-based verification. U2: Reliability-based assessment. U3: Risk-based assessment.
Knowledge content
K1: Existing technical documents. K2: Collect data (low value) on material properties and/ K3: Collect data (high value) on material properties and/or
or actual loading. actual loading.

knowledge/information content axis. viewed as the origin in Fig. 5. Increasing the level of any one of these
The advantage with the MUK cube representation shown in Fig. 5 is factors, either individually or in combination, is a strategy for improved
that any method for condition assessment can be incorporated along its and more informed decision making. In other words, an enhanced as-
axes. Consider, for example, the frameworks referenced in Table 1; the sessment essentially involves moving further away from the origin.
assessment levels for these approaches represent navigation along one Utilizing the MUK model in a decision-making context then requires
or more of the axes in Fig. 5. In practice, the type of assessment used specific attentions to one and/or both of the following issues: a) eval-
might be determined by the experience, competence and traditions of uating the suitability of certain methods in relation to others; b) de-
those performing the assessment. The aim with the decision support termining an optimal way of navigating the MUK cube in the case of
framework in this paper is to provide a systematic approach for de- successive assessments.
ciding upon a suitable navigation through the cube; i.e. providing a Generally speaking, moving further away from the origin of the
rational and transparent basis for deciding on which approach to apply. MUK cube will require resources. Optimum decision making concerns
One possible issue with adopting this approach in favor of traditional or determining a decision which achieves the greatest expected benefit (in
routine procedures is that the investment in terms of resources required terms of improved performance) in relation to the resources required
for adopting a new method may be higher the first time owing to, e.g., (considering that these are limited). In the context of a specific level of
limitations in terms of competence or experience. This is something that assessment according to the MUK cube the following decision are
is difficult to avoid but could be mitigated through appropriate edu- possible:
cation, quality assurance procedures, and will become less of an issue
when adequate experience has been achieved. 1. Do nothing (for now);
Ultimately, navigation in the cube should somehow consider the 2. Adopt higher level of assessment; i.e. navigate within the cube by
amount of resources required to adopt a higher level of assessment o Increasing modelling sophistication;
while accounting for the potential benefits in terms of improved results. o Increasing uncertainty/risk considerations;
It may be assumed that moving away from the origin of the cube is o Collecting data/improving knowledge; or
synonymous with improved accuracy of results. In other word, the o Combination of two or all of these.
aforementioned improvement relates to an increase of confidence and 3. Implement invasive actions; i.e. repair, strengthening, etc.
erosion of conservatism that may be inherent with lower levels of as-
sessment. However, this assumption may not always be accurate in The first decision is taken if the initial condition assessment de-
practice and as such it is important to consider to what degree the re- termines that the structure does not, within some set reference time
sults actually improve. A prominent issue to consider is the variation in frame (e.g. inspection intervals), require any more attention (i.e. ac-
engineering performance which has been observed in practice; i.e. the ceptable performance level is determined based on predefined criteria
so-called engineering modeling uncertainty discussed by Fröderberg & such as a code-specified target reliability) and that no immediate action
Thelandersson [28]. Another source of uncertainty may be related to is required. Following this decision, an update of the inspection/
the fact that higher levels of assessment are often more complex, re- maintenance plan may be appropriate. The second and third decisions
quiring a greater number of variables/parameters, consequently in- are taken if the initial condition assessment determines that the level of
creasing the parameter uncertainty in relation to the model uncertainty. damage in the structure is not acceptable (i.e. unacceptable perfor-
In cases where there is limited knowledge or information regarding the mance level based on current level of assessment) and that more action
value of these parameters, the decision maker is forced to make as- is needed. The second decision, to adopt a higher level of assessment,
sumptions in order to carry out the analysis and obtain some results. essentially involves acquiring ‘more acceptable results’, where accep-
The risk is that the results obtained from an assessment can be mis- table may be interpreted as ‘accurate’, and may be more appropriate,
leading if they are based on unfounded or inaccurate values or as- e.g., if:
sumptions. Klasson et al. [30], for instance, observed a relatively high
degree of variation in design and modeling assumptions made by ex- • The margin for underperformance is low (e.g. the load carrying
perienced engineers. Considering these issues, moving along either of capacity is slightly less than the prescribed demand);
the first two axes – modeling sophistication or risk/uncertainty con- • The initial degree of conservatism is high – this can be remedied by
sideration – should ideally be accompanied by a shift along the third more explicit consideration of the underlying uncertainties or by
axis, knowledge content. hiring experts (e.g. for older structures with limited data);
• The amount of resources needed for invasive action is deemed un-
3.2. Generic decision model acceptably high in relation to the cost of improving the assessment
method.
A generic decision model is presented which integrates the MUK
model from the previous section with decisions concerning the main- The decision to implement invasive actions is taken only if an im-
tenance of an existing bridge. A decision maker could consider that proved level of enhanced assessment methods is assumed to yield
there are a number of possible methods for evaluating the condition of minimal improvements and/or if the consequences of doing nothing
part or all of the structure and determining which method is suitable in given that damage is present are significantly high. If both assessment
a specific case can be guided by the MUK cube. The initial, often code- methods and invasive actions are considered too costly, then less costly
based, assessment of a structure whose condition is in doubt can be measures, such as limiting vehicle access, may be more suitable; this is

7
I. Björnsson, et al. Structural Safety 81 (2019) 101874

Table 3
Example showing generic sample likelihood for enhanced max(E [u (aj |en, z k )]) = max u (en, z k , aj , i )·P ( i)
condition assessment.
j j
i (2)

Assessment result, z Action, a Since the outcome of the assessments is uncertain, a set of decision
Indicates damage Repair rules d(zk|en,aj) ∈ d(z,a) is required to map the observations into ap-
Indicates no damage Do nothing propriate actions. In general, the decision rule applied in this analysis,
can be stated as always choosing the decision with yields the lowest
expected cost (i.e. the highest expected utility), assuming that if addi-
another variation of the first decision alternative (do nothing).
tional information becomes available, the decision maker will continue
To solve the previous decision problem, the generic influence dia-
act optimally (according to the current beliefs). This general definition
gram from Fig. 4 is adopted. In this case, the method decision node now
of the decision rule is sufficient if the extensive form (backward in-
concerns possible assessment methods (i.e. moving along the MUK cube)
duction) of the decision analysis is used. If the number of decision al-
while the action decision node concerns possible invasive actions (i.e.
ternatives is large, the extensive form might be impractical and the
repair, strengthening, etc.). The state spaces for these decision nodes
normal form with a simple decision rule can be adopted as presented in
are {e} and {a} respectively and include the null decision alternative e0
Table 3.
and a0 (i.e. to make no enhanced assessment and to take no invasive
The probability of obtaining a specific result for any given assess-
action). The two chance nodes, state of nature and experiment outcome,
ment method P(zk|en) is calculated from the sample likelihoods and
now reflect the condition, or state of the bridge, and the assessment re-
prior probabilities for the bridge state:
sults; the state spaces for these nodes are {θ} and {z} respectively. The
actual state of the bridge is independent of the other decisions and P (z k |en) = P ( i )·P (z k |en, i) = P ( i)· Lk, i
outcomes and thus only the prior probabilities are needed as input. The i i (3)
assessment results, on the other hand, are dependent on which assess- where the sample likelihoods, P(zk|en,θi) = Lk,i, reflect the uncertainty
ment method, en, is chosen as well as on the actual bridge state, θi. of the result provided by a given assessment method when applied in
Finally, the utility nodes, method costs and utilities, reflect the costs practice. An example is shown in Table 4 for the simplified case of
(negative utilities) associated with the various assessment methods, {u binary outcomes and bridge states. In Table 4, damage and no damage
(e)}, as well the costs and/or benefits associated with a certain state of could be taken to mean non-compliance or compliance with the ac-
the bridge θi and whether or not an invasive action aj has been taken, {u ceptance criteria specific to the assessment method being considered.
(θ,a)}. For example, the direct costs of a0 (doing nothing) is obviously
zero in contrast with the non-zero costs of some enhanced assessment 3.2.2. Prior probabilities and sample likelihoods
which would require resources in terms of man hours, administration Assigning appropriate values for the prior probabilities P’(θ) for the
costs, collecting information, etc. If, for example, a bridge or bridge bridge states and the sample likelihoods (Lk,i) associated with a specific
component has some degree of damage (reflected by a specific bridge assessment method is an important aspect of the decision modeling. The
state θj) then the potential costs will depend on whether the bridge or former should reflect the degree of doubt cast on the condition of the
the detail is repaired, strengthened or if nothing is done; this is reflected bridge and could be determined based on some reliability or prob-
using the utilities for each outcome pair (θ,a). The case of strengthening abilistic approach. However, this is in itself an enhanced assessment
could be considered by applying a positive utility (benefit); this ap- and as such a more simplified approach could be utilized during the
proach could, however, be questioned as it seems logical that it is ac- initial stages of applying the decision framework. Successive applica-
tually the bridge state probabilities P(θ) which are altered but this need tion of the framework should ensure that the prior probabilities are
not be the case. Consider, e.g., a substandard bridge which cannot fulfill updated based on new information collected during the bridges life-
its function through simple repair as a result of a substantial increase in time. An example for prior probabilities is provided in the calculation
traffic loads on the bridge since it was built. example in the next section.
The sample likelihoods can be determined with the aid of the MUK
3.2.1. Expected utilities model which codifies possible condition assessment methods for a
Initial doubts concerning the condition of the bridge (or one of its specific bridge case according to the cube in Fig. 5. Moving along the
components) may be based on a simplified assessment and the decision different axes of the cube could be considered as a way of altering the
to be made in light of this information is whether to take some invasive sample likelihoods (Lk,i) associated with a specific assessment method.
action, e.g. to repair the bridge, or to perform an enhanced assessment Navigating within the cube away from the origin should improve the
in hopes of obtaining more favorable results. There are thus two suc- results; the question is how reliable these results are in relation to their
cessive decisions to be considered in this case: costs. More specifically, how can the sample likelihoods (see Table 4)
be determined for a specific assessment method based on the position
1. Decision concerning assessment method (i.e. navigating the cube); within the cube? The sample likelihoods are independent of the initial
2. Decision concerning invasive action (i.e. maintenance, rehabilita- (or any previous assessments) which have been made; as they merely
tion or repair actions). reflect the decision makers’ confidence in the results of a specified as-
sessment method. A lack of confidence may, e.g., be due to poor or
The optimal decision concerning the assessment method is de- overly simplified modeling or uncertainty assumptions as well as a lack
termined by maximizing the expected utility as follows: of reliable data, knowledge or information to support the underlying
assumptions.
max[E [u (en )]] = max P (z k |en )·max u (en, z k , aj , i )·P ( i)
n n j Table 4
k i
Example showing generic sample likelihood for enhanced condition assessment.
(1)
Assessment result, z Bridge (or component) state, θ
where P’’(θi) is the posterior probability of state θi considering the
possible outcomes of the assessment methods {e, z}. The optimal de- Damage No damage
cision concerning which invasive action should be chosen will also
Indicates damage L1,1 L1,2
depend on this outcome (i.e. the pair {e, z}) and is contained with Eq. Indicates no damage L2,1 L2,2
(1):

8
I. Björnsson, et al. Structural Safety 81 (2019) 101874

Considering the MUK cube from Fig. 5, some assumptions can be Table 5
made regarding the effect that moving along the axes of the cube has on Overview of nodes of the influence diagram for numerical example; see also
the sample likelihoods. To start, it is reasonable to assume that navi- Fig. 4.
gating along any of the axes will increase the decision makers’ con- Name Type States Input
fidence in the results; i.e. L1,1 and L2,2 from Table 4 will increase. It
should, however, be mentioned that this adjustment may not necessa- Assessment*, e Decision Assessment 1
Assessment 2
rily improve the overall results using the decision model (in terms of the
Intervention, a Decision Do nothing
final expected utilities). The degree of increase along each axis reflects Repair
the reliability of the method chosen and could be assessed based on Component state, θ Chance No damage Prior probabilities
previous experience with using this method or with the aid of experts in Damage P’(θ i)
Assessment results, z Chance Indicates damage Sample likelihoods
the field. This requires that each method first be categorized to reflect
Indicates no P(zk|en, θ i)
their position within the cube. damage
Generally speaking, it is reasonable to assume that moving along Assessment costs Utility – Cost of assessment
one axis (M, U or K) of the MUK cube improves the sample likelihoods method (e)
more than moving along another (e.g. dL1,1/dK > dL1,1/dU) but this U(en)
Utilities Utility – Cost of action, state pair
would need to be assessed in more detail and is likely to be case de-
U(aj, θ i)
pendent (i.e. considering the damage scenario types and available as-
sessment methods). On the other hand, moving along more than one * The decision not to do an assessment is also considered (i.e. a prior ana-
axis is likely a more effective approach than simply moving along one lysis).
(e.g. d3L1,1/dMdUdK > d2L1,1/dMdU > dL1,1/dM). Ultimately, a sys-
tematic approach is required to effectively determine the sample like- considered; one more accurate, and costly, than the other. In practice,
lihoods for different methods; this is an area intended for future re- these methods could, for instance, represent two difference approaches
search by the authors of this paper but is outside the scope of the for estimating the impact of corrosion on an edge beam for a reinforced
current paper. concrete bridge. The purpose of the case study is to illustrate, in a
general sense, how the current framework provides a structured way of
3.2.3. Utility assignment dealing with issues arising at making decisions about selecting the most
The most favourable path within the MUK cube is predicted by suitable assessment option in bridge condition assessments.
maximizing the expected utility as described by Eq. (1). The value of An overview of the decision nodes, including state spaces and re-
the utility function for different outcomes, including possible benefits/ quired input in terms of probabilities and utilities is provided in
rewards as well as costs/drawbacks, need to be assigned with care. Table 5.
These could be based on linear utility theory as proposed initially by
von Neumann & Morgenstern [34] or, as is a common approach in civil
engineering application, based on some form of consequence model- 4.1.1. Prior probabilities and sample likelihoods
ling. Relevant consequences could include human casualties, structural As mentioned earlier, the prior probabilities for the state of the
damages, disruptions in the transport network, etc (see, e.g., bridge component, θ, can be determined using some reliability or
[36,37,44]). In practice monetary values are often used; however, this probability based assessment. However, an initial assessment may not
is not strictly necessary. It may be sufficient if the utility function de- yield these types of figures and as a result some assumption must be
scribes the relative consequences (and possible benefits) correctly be- made concerning the criticality of the damage for the specified re-
tween the different bridge states, the invasive actions, and the costs of ference period. In this example all non-conditional probabilities will be
the assessments. Relative costs can be estimated based on, e.g., time expressed for a one-year reference period (although a reference period
required for assessment and repair, and loss of time due to traffic dis- corresponding to the inspection interval may also be used). In this ex-
ruption. A more complete option would be an elaborated utility func- ample the prior probabilities are determined consider that the initial
tion based on actual monetary costs, including the costs of the different assessment is likely to be a deterministic and normative calculation in
actions and socioeconomic costs for the outcomes. in that case, the which a performance indicator σS, based on specified loads (or more
references mentioned earlier could be useful. generally demands) to the structure, is compared with a specified limit
resistance (or capacity) value σR. From this the prior probabilities may
4. Case studies determined for three illustrative cases in which the margin between the
demand and capacity (i.e. σR - σS) are relatively small, moderate or
The following are two case studies in which the framework has been large. It is reasonable to assume that there is a correlation between this
applied. The first is a generic example demonstrating possible decision margin and the annual probabilities of there being damage; see Table 6.
scenarios in a general manner. The second is a recapitulation of a de- The damage severity could also be assessed qualitatively by considering
tailed case for fatigue assessment of a steel railway bridge in Stockholm, the severity of the damage developed over time. Thus, the age of the
Sweden, presented by Leander et al. [45]. bridge being considered may also be considered a relevant factor in
determining prior probabilities. The numbers in the table may seem to
4.1. Generic example be large compared to safety targets in structural codes; however, they
represent the probability of damage in a bridge component based on the
In this section a generic example is presented in which the decision result of a deterministic safety assessment and it would not be any
framework from this paper is applied. The example considers a bridge purpose to proceed with more enhanced actions if the prior failure
component which may be damaged and the bridge manager must de-
cide whether to repair the component or not before the next inspection Table 6
Prior probabilities for Bridge state, θ (reference period of 1 year).
takes place. It is assumed that a simplified assessment using the relevant
structural codes yielded a non-compliant result. However, prior to Damage severity P(No Damage) P(Damage)
taking any immediate and costly invasive action the bridge manager
Small 0.999 0.001
may choose to employ some enhanced condition assessment to de-
Moderate 0.99 0.01
termine whether the damage level is in fact critical. In this example, Large 0.9 0.1
two possible choices for an enhanced condition assessment are

9
I. Björnsson, et al. Structural Safety 81 (2019) 101874

Table 7 Table 9
Sample likelihoods for enhanced assessments. Optimal decision concerning enhanced assessments for different scenarios.
Assessment result, z Assessment 1 Assessment 2 Damage severity
Bridge state, θ Bridge state, θ
Small Moderate Large
No damage Damage No Damage Damage
Low criticality No Assessment* No assessment* Assessment 1
Indicates damage 0.3 0.7 0.1 0.9 Moderate criticality No Assessment* Assessment 1 No Assessment**
Indicates no damage 0.7 0.3 0.9 0.1 High criticality Assessment 1 No Assessment** No Assessment**

* No repair needed (repair costs too high compared with expected damage
probabilities would be acceptable. In practice, initial assessments can costs)
sometimes show very high failure probabilities and the major objective ** Repair bridge (expected damage costs exceed repair costs)
can then be to reach as low probabilities as possible, not necessarily
trying to reach as low as the target levels for the design of new struc- Table 10
tures. Expected utilities based on pre-posterior analysis for moderate bridge criti-
cality.
The sample likelihoods for the enhanced assessment methods pro-
vided in Table 7 reflect the accuracy of these methods applied in Assessment method Damage severity
practice. From this table, it can be seen that the decision maker is fairly
Small Moderate Large
confident that the results obtained from the first approach do in fact
accurately reflect the state of the bridge although their confidence in No Assessment −0.1 −1 −1
the second, more expensive, approach is higher. In this example it is Assessment 1 −0.2 −0.70 −1.1
assumed that the likelihoods of a ‘true positive’ and a ‘true negative’ are Assessment 2 −0.6 −0.71 −1.5
the same although this may not necessarily be the case. The two as-
sessment methods represent different locations in the MUK cube, where
combined with less severe damage levels the bridge manager should
Enhanced assessment 1 can be seen as moving along one axis and En-
make no additional assessment nor make any repairs. This is since the
hanced assessment 2 as moving along two axes. For example, in the case
expected damage costs are lower than the costs of repair. This type of
of a corroded edge beam it may be that the first method is a more
scenario illustrates cases in which a risk (decision) based criteria yields
advanced deterministic calculation (based on the initial code-based
more favorable results than the design codes. In these instances the
assessment) while the second extends this to a probabilistic calculation.
bridge manager should consider whether some less invasive action
should be taken; e.g., reducing the speed limit or the allowable weight
4.1.2. Utilities
of the trucks passing the bridge. Adopting these types of actions may
In this example, the cost of carrying out repairs is assumed to be −1
ensure that the safety level for the bridge is adequate. While invasive
unit of utility while the remaining utilities are provided relative to this
actions usually involve making adjustments to the bridge structure,
value. The costs of carrying out the enhanced assessments 1 and 2 are
non-invasive actions may reduce the impact of external perturbations
assumed to be −0.1 and −0.5 respectively. The cost of damage to the
on the bridge; i.e. level of loading. Evaluating the exact nature of the
bridge in case no repair is carried out is considered for three different
non-invasive action may, however, require a new assessment. Such a
scenarios to reflect the criticality of the bridge in terms of the indirect
decision could also be made more explicit by incorporating them di-
consequences of local failure. Thus, a bridge with low criticality is ex-
rectly into the decision problem.
pected to have lower costs of local failure in relation to the repair costs
In cases where the damage severity and criticality levels are both
while for a bridge with high criticality the opposite is true. The utilities
high, then no enhanced assessment should be made in favor of making
used in this example are provided in Table 8. For reference, Wong et al.
immediate repairs. In these cases, the expected costs of damage exceed
[36], estimated the ratio of bridge (system) failure to component (ele-
the repair cost and additional assessments are unlikely to improve the
ment) failure costs for eight roadway bridges in the UK; this ratio
outcome. This type of scenario concerns, e.g., bridges which are critical
ranged from approximately 20 to 120.
for the surrounding transportation network. It is reasonable that these
types of bridges should be allotted more resources to maintain in
4.1.3. Results comparison with less critical infrastructures. The framework can thus
The decision problem was evaluated to determine what the optimal potentially help with prioritizing a group of structures to optimize re-
decision would be both regarding the choice of enhanced condition source expenditure by the managing authorities.
assessments as well as the invasive action (i.e. to repair or not) using the In between the two aforementioned extremes, the decision con-
expected utility decision criterion. The results were calculated for dif- cerning which enhanced assessment is more appropriate can be made.
ferent damage severities and bridge/component criticalities according The results showed that the cheaper assessment method (Assessment 1)
to Tables 7 and 8. Table 9 shows the optimal decision concerning the is preferable, although the difference between two assessment methods
enhanced assessment for the different scenarios considered. To show is not significant in all cases; see Table 10. This means that, in this
how these results were obtained, the expected utilities for each method specific case, there is no reason for adopting a method which is placed
are provided in Table 10 for the case of moderate criticality. further along the axes of the MUK cube since the benefit such an
The results from Table 9 indicate that for lower levels of criticality

Table 8
Utilities for different end scenarios/states.
Action, A Low criticality Moderate criticality High criticality
Component state, θ Component state, θ Component state, θ

Damage No Damage Damage No Damage Damage No Damage

Do nothing −10 0 −100 0 −1000 0


Repair −1 −1 −1 −1 −1 −1

10
I. Björnsson, et al. Structural Safety 81 (2019) 101874

investment will not be large enough. If the prior probabilities in Table 6


had been altered or if the cost of the second assessment method was
decreased, then Assessment 2 may have been more optimal; e.g. As-
sessment 2 would have been preferred to Assessment 1 in Table 9 if its
cost had decreased from −0.5 to −0.4. The second decision in the
decision model, concerning repairs of the structure, should be evaluated
based on the outcome of the enhanced assessment once it has been
carried out. If the assessment, in this case Assessment 1, reveals the
condition is still not adequate then repair should be chosen (assuming
no additional assessment is carried out). In that case, the same decision
model can be applied using updated prior probabilities; i.e. posterior
probabilities based on outcome of the assessment.

4.2. A fatigue assessment scenario Fig. 7. Expected utilities for the decision options. Reproduced after Leander
et al. [45].
The application of the MUK model and decision framework for a
decision scenario related to the fatigue assessment of a steel bridge has
evaluation of the expected utilities are shown in Fig. 6. For details about
been demonstrated in detail by Leander et al. [45]. This case is briefly
the numbers used in the example (priors, sample likelihoods, costs), the
recapitulated in this section. The purpose of the inclusion of this case is
reader is referred to the paper by Leander et al. [45].
to highlight that the approaches can be used in real cases even with a
Fig. 7 shows that the more costly options e2 and e3 involving in-
specific scope such as the assessment of a fatigue critical detail. The
spections provide the highest utilities and are therefore the most ap-
Söderström Bridge is a high priority railway bridge in the city center of
propriate decisions alternatives for this case. These options are located
Stockholm along the line with the highest traffic intensity in Sweden.
away from the origin in all three directions of the MUK cube and are
Based on inspections and preliminary theoretical assessments, its fa-
expected to provide more accurate assessments. For this example, the
tigue life has been judged exhausted. To secure the structural safety of
gain in knowledge overrides the costs of these more advanced methods.
the bridge until a replacement is in place the following alternative
See the paper by Leander et al. [45] for more details about the case
enhanced condition assessments have been suggested:
study bridge and the example.

e0 – Prior reliability-based assessment using measured stresses


5. Conclusions
e1 – Assessment based on LEFM without inspection
e2 – Assessment based on LEFM with visual inspection
In the present paper a rational and systematic framework has been
e3 – Assessment based on LEFM with magnetic particle testing
presented to aid decisions concerning practical bridge condition as-
sessments. This framework is based on Bayesian decision theory and
A preliminary assessment using a deterministic verification format
takes into account three main attributes of enhanced condition assess-
as suggested in the governing regulations has been performed and is
ments, namely modelling sophistication, consideration of uncertainties
therefore not included as an option for further actions. It represents the
and risks, and knowledge/information content. Improving the level of
base state or the origin in the MUK model (Fig. 5). The relative position
one or more of these attributes is a strategy for improved condition
of each of the considered assessment methods (e0 – e3) within the cube
assessment. However, the suitability of adopting an enhanced condition
is provided in Fig. 6. The first option e0 involves a reliability-based
assessment in practice will depend on the expected benefits of obtaining
assessment which is an advancement along the U axis, and the use of
favourable results in relation to the implementation costs. The frame-
measured response which provides an increase of knowledge and is,
work presented in the paper provides guidance in this regard and aids
thereby, an advancement along the K axis. The option e1 involves the
with decisions concerning whether and what additional assessments are
use of linear elastic fracture mechanics (LEFM) which increases the
necessary, if invasive actions are needed, or if some other non-invasive
level of modeling sophistication in comparison to the conventional
option may be more appropriate. It allows for the consideration of
linear damage accumulation. The options e2 and e3 provide further
uncertainties associated with the implementation of condition assess-
advancements along the K axis with increased knowledge by in-
ments in practice; an aspect which is not often explicitly considered in
corporating inspection outcomes in the assessment. The costs increase
existing approaches. In addition, the framework is flexible in the sense
concurrent with the decision options due to more advanced assessment
that the sequence of enhancement steps is not predefined. This can help
models and more accurate inspection methods. The results of the
the decision maker select the most suitable assessment method and/or
action; i.e. the method or action that is likely to save the most money
U
for the bridge operator.
A limitation of the presented decision framework is that it currently
does not include the possibility of successive assessments; i.e. that after
selecting a given assessment method, it is possible to carry out further
e0 e1
assessments. The framework could, however, be re-applied in practice
e2 to aid in selecting successive assessments in a sequential manner.
e3 M
However, this may result to a suboptimal sequence of assessment ac-
tions. It would thus be more appropriate to include the possibility of
successive actions within the same framework; one possible solution
would be to apply a dynamic decision model.

K Acknowledgments

Fig. 6. Position of enhanced assessments for fatigue verification using MUK The authors acknowledge the financial support provided by the
model. Swedish Transport Administration (Trafikverket) and the strategic

11
I. Björnsson, et al. Structural Safety 81 (2019) 101874

innovation programme InfraSweden2030, a joint effort of Sweden's engineering documents. Zürich: International Association for Bridge and Structural
Innovation Agency (Vinnova), the Swedish Research Council (Formas) Engineering.
[21] JCSS (2001) Probabilistic Assessment of Existing Structures, A publication of the
and the Swedish Energy Agency (Energimyndigheten). The grant Joint Committee on Structural Safety (JCSS), RILEM Publications.
numbers are TRV 2016/35662 and 2016-03320 respectively. Finally, [22] SAMCO (2006) Final Report, F08a Guideline for the Assessment of Existing
we are grateful to the members of COST Action TU1402 (http://www. Structures.
[23] Wenzel H. Health Monitoring of Bridges. John Wiley & Sons; 2009.
cost-tu1402.eu/). The discussions at various meetings provided a great [24] Skokandic, D., Mandić Ivanković, A., Dzeba, I. (2016) Multi-level road bridge as-
deal of help and motivation to our work. sessment. In: 19th IABSE Congress Stockholm, 21-23 September 2016.
[25] Jonassen D, Strobel J, Lee CB. Everyday problem solving in engineering: lessons for
engineering educators. J Eng Educ 2006;95(2):139–51.
References [26] Stewart MG, Melchers RE. Simulation of human error in a design loading task.
Struct Saf 1988;5:285–97.
[1] Trafikverket (2018). BaTMan RAPPORT – Broars ålderfördelning. (https://webapp. [27] Bürge M, Schneider J. Variability in professional design. Struct Eng Int
trafikverket.se/VNapps/vv/webhybris/30.Fakta/32.03 Broars åldersfördelning. 1994;4:247–50.
vns) Accessed 2018-08-14. [28] Fröderberg M, Thelandersson S. Uncertainty caused variability in preliminary
[2] BRIME (2001) Bridge Management in Europe, Deliverable 14, Final report. structural design of buildings. Struct Saf 2015;52(B):183–93.
[3] ISO. ISO 13822:2010, Bases for design of structures – Assessment of existing [29] Fröderberg M. Conceptual Design Strategy: Appraisal of Practitioner Approaches.
structures. Geneva: International Organization for Standardization; 2010. Struct Eng Int 2015;25(2):151–8.
[4] CEN (2010) EN 13306:2010, Maintenance – Maintenance terminology. European [30] Klasson A, Björnsson I, Crocetti R, Hansson EF. Slender roof structures – failure
Committee for Standardization, Brussels. reviews and a qualitative survey of experienced structural engineers. Structures
[5] Frangopol DM. Life-cycle performance, management, and optimisation of structural 2018;15:174–83.
systems under uncertainty: accomplishments and challenges. Struct Infrastructure [31] Elms, D.G. (1985) The principle of consistent crudeness. In: Proc. Workshop on Civil
Eng 2011;7(6):389–413. Engineering Application of Fuzzy Sets: 35-44.
[6] Sánchez-Silva M, Frangopol DM, Padgett J, Soliman M. Maintenance and operation [32] Elms DG. Achieving structural safety: theoretical considerations. Struct Saf
of infrastructure systems: review. J Struct Eng 2016;142(9):F4016004. 1999;21:311–33.
[7] Pozzi, M., Der Kiureghian, A. (2011) Assessing the value of information for long- [33] Benjamin JR, Cornell CA. Probability, Statistics, and Decisions for Civil Engineers.
term structural health monitoring. In: Kundu, T. (ed.) Proc. SPIE 7984, Health New York: McGraw-Hill Book Company; 1970.
Monitoring of Structural and Biological Systems, Vol. 7984, pp. 79842W–1–14. [34] von Neumann J, Morgenstern O. Theory of games and economic behavior. 2nd
[8] Frangopol DM, Dong Y, Sabatino S. Bridge life-cycle performance and cost: analysis, edition. Princeton University Press; 1947.
prediction, optimisation and decision-making. Struct Infrastruct Eng [35] Faber M. On the treatment of uncertainties and probabilities in engineering decision
2017;13(10):1239–57. analysis. J Offshore Mech Arct Eng 2005;127:243–8.
[9] Thoft-Christensen, P. (2008) Structure of bridge management systems. Optimal [36] Wong SM, Onof CJ, Hobbs RE. Models for evaluating the costs of bridge failure.
bridge maintenance based on risk and reliability, Toluca, Mexico, August, 2008. Bridge Engineering 2005;158(BE3):117–28.
[10] Moubray J. Reliability-Centered Maintenance. 2nd Edition New York: Industrial [37] Björnsson I. Holistic approach for treatment of accidental hazards during con-
Press; 1997. ceptual design of bridges – a case study in Sweden. Saf Sci 2017;91:168–80.
[11] Sørensen JD. Framework for risk-based planning of operation and maintenance for [38] Howard RA, Matheson JE. Influence diagrams. Decision Analysis
offshore wind turbines. Wind Energy 2009;12:493–506. 2005;2(3):127–43.
[12] Goyet J, Boutillier V, Rouhan A. Risk based inspection for offshore structures. Ships [39] Kjærulff UB, Madsen AL. Bayesian networks and influence diagrams – a guide to
Offshore Struct 2013;8(3–4):303–18. construction and analysis. New York: Springer; 2008.
[13] Honfi, D., Lechner, T. & Köhler, J. (2017c) Rational maintenance of timber bridges. [40] Shachter RD, Peot MA. Decision making using probabilistic inference methods.
In: 3rd International Conference on Timber Bridges (ICTB2017), Skellefteå, Proceedings of the Eighth Conference on Uncertainty in Artificial Intelligence. San
Sweden, 26-29 June, 2017. Mateo, CA: Morgan Kaufmann Publishers; 1992. p. 276–83.
[14] Barone G, Frangopol DM. Life-cycle maintenance of deteriorating structures by [41] Koller D, Milch B. Multi-agent influence diagrams for representing and solving
multi-objective optimization involving reliability, risk, availability, hazard and games. Games Econ Behav 2003;45:181–221.
cost. Struct Saf 2014;48:40–50. [42] Honfi, D., Leander, J. & Björnsson, I. (2017a) Decision support for bridge condition
[15] Nielsen JS, Sørensen JD. Computational framework for risk-based planning of in- assessment. In: 4th International Conference on Smart Monitoring, Assessment and
spections, maintenance and condition monitoring using discrete Bayesian net- Rehabilitation of Civil Structures (SMAR 2017), Zürich, Switzerland, 13-15
works’. Struct Infrastructure Eng Taylor & Francis 2017:1–13. September, 2017.
[16] Weber P, Medina-Oliva G, Simon C, Iung B. Overview on Bayesian networks ap- [43] Schlune H. Safety Evaluation of Concrete Structures with Nonlinear Analysis PhD
plications for dependability, risk analysis and maintenance areas. Engineering Thesis Division of Structural Engineering, Sweden: Chalmers University of
Applications of Artificial Intelligence, Elsevier 2012;25(4):671–82. Technology; 2011.
[17] Gubbi J, Buyya R, Marusic S, Palaniswami M. Internet of Things (IoT): A vision, [44] Trafikverket (2016) Samhällsekonomiska principer och kalkylvärden för trans-
architectural elements, and future directions. Future Generation Comput Syst portsektorn: ASEK 6 (In Swedish). Arbetsgruppen för Samhällsekonomiska Kalkyl-
2013;29(7):1645–60. och Analysmetoder inom Transportområdet. https://www.trafikverket.se/for-dig-i-
[18] Trafikverket (2015) BaTMans handbok (In Swedish). https://batmanhandbok. branschen/Planera-och-utreda/Planerings–och-analysmetoder/
trafikverket.se. Samhallsekonomisk-analys-och-trafikanalys/gallande-forutsattningar-och-indata/.
[19] Sustainable Bridges (2007) Guideline for Load and Resistance Assessment of [45] Leander J, Honfi D, Larsson Ivanov O, Björnsson I. A decision support framework
Existing European Railway Bridges, SB-LRA. for fatigue assessment of steel bridges. Eng Fail Anal 2018;91:306–14.
[20] Schneider, J. (1997) Introduction to Safety and Reliability of Structures. Structural

12

Potrebbero piacerti anche