Professional Documents
Culture Documents
Robust Final Project-Alizadeh
Robust Final Project-Alizadeh
Robust Final Project-Alizadeh
Abstract
Optimal maintenance alternatives are those solutions that minimize the life-cycle cost of an infrastructure network
while fulfilling reliability and functionality requirements over a given planning horizon. This research aims at
finding a Robust Optimization approach to determine the optimal set of maintenance alternatives for a network of
infrastructure facilities (namely bridges). The proposed approach can produce successfully-performing maintenance
scenarios under the perturbation of bridge condition grades while maintaining well-balanced maintenance strategy
both in terms of bridge performance and maintenance cost. Thus a well-balanced planning strategy in terms of the
mutually conflicting performance and cost measures can be achieved.
1. Introduction
Infrastructure management systems have been developed to apply the life-cycle costing approach to optimize
maintenance decisions at both network and project levels and achieving network/project performance requirements
under financial constraints.
This is extremely important for most facility managers because of the limitations on the availability of resources
required to fulfill even urgent maintenance needs. Since the early 1980s, many optimization techniques have been
adopted for this purpose. Although these optimization techniques have provided satisfactory results, many agencies
still prefer using traditional methods of maintenance optimization in spite of their arbitrary nature and relatively
low degree of accuracy. These methods are mostly heuristic and based on subjective ranking and priority rules
developed by domain experts. The reason of this preference is due in-part to the mathematical complication of
formulating a maintenance optimization problem using these techniques in addition to the computational
complexity associated with large size networks.
Here is where Robust Optimization can be effective. Robust Optimization has been successfully solving large scale
problems maintaining both feasibility robustness and optimality robustness of the problems.
model1
This problem is a multi-choice knapsack problem (MCKP).
The objective function is to maximize the network wide reward, which is a function of multiple performance
indicators.
-BudgetConstraint, MCKP)
constraint is replaced by a condition constraint (which
means that there is still only one constraint in total). It is possible to transform the Formulation 2 problem to be
very similar to Formulation 1. For example, the decision maker might want to minimize the agency costs to attain
given condition targets or risk levels, while simultaneously optimizing for multiple objectives. The non-budget
constraint could be, for example, that the average health index of the network be at least Hmin, or that the average
vulnerability rating of the network be no more than Vmax. The problem can then be formulated as the following
MCKP:
model2
model3
The problem is an MCMDKP.
-Budge Constraints, MCMDKP)
maker seeks to determine minimum possible costs to
achieve condition and risk targets, while maximizing the network wide rewards in terms of multiple performance
criteria. The formulation is as follows:
model4
In the presence of uncertainty, the gamble method is used to derive the functional form of a multi-criteria utility
function. This method establishes the scaling constants for a given set of performance measures.
One approach is an additive utility function and the other approach is the multiplicative utility function.
Establishing the scaling constants for a given set of performance measures, if the sum of the acquired constants add
up to 1, this implies that an additive utility function is appropriate for the given multi-criteria problem (i.e., the
weighted values or utilities of the performance measures can be added together to
obtain the overall performance). However, if the sum is not equal to 1, then a multiplicative utility function is more
appropriate
Because the measures in the bridge preservation ratings set (NBI ratings, health index, and sufficiency rating)
overlap to a much large extent, it can be argued that a multiplicative functional form would be more appropriate for
intra-set relative utility functions, and an additive utility function is appropriate for relative overall goals.
Thus, the multi-criteria utility function, in the presence of uncertainty, is of multiplicative form for all intra-set
performance measures. This function is given as follows (Keeney
Table 2 presents the scaling constants ki established by the gamble method for individual performance measures.
Table 2: scaling constants
Tables 3 refers to relative weighs assessed through NCHRP research, report 590:
Table 3: recommended relative weights, Overall goals.
Hence the overall utility function concerning individual sets of performance measures weighed against each other
according to relative weights mentioned in table 3 would be calculated by the following equation:
=0.360*U (Bridge preservation) +0.205*U (safety) + 0.150*U (protection from extreme -events) +0.175*U
(agency Cost) +0.110*U(user cost) (1)
Where
U (bridge preservation) = [0.42*U (NBI condition ratings) +1]*[0.53*U (health Index) +1]
*[0.34*U (Sufficiency rating) +1] (2)
U(protection from extreme events)=[0.38*U(Scour vulnerability rating)+1]*[0.42*U(fatigue vulnerability
rating)+1]*[0.37*U(earthquake vulnerability rating)+1] (3)
U (Traffic safety) = [0.40*U (Geometric rating) +1]*[0.49*U (inventory rating) +1] (4)
As for costs, having in mind the time value of money,
= (5)
( )
= (6)
( )
Being the user cost associated with the jth maintenance action applied at time Ti if project j is implemented
for bridge k
Being the Agency cost associated with the jth maintenance action applied at time Ti if project j is
implemented for bridge k
As for assessing single criterion utility functions, according to NCHRP research, report 590, table 4 determines the
relation between each performance measure utility function with its corresponding value function.
Table 4: Utility functions related to value functions for each performance measure
For those non-linear utility functions figures 1 through 7 (from NCHRP research, report 590depict their corresponding
value functions:
As mentioned in Table 4, and depicted in figures 1-7, for the following ratings,
FSVR, EVR, OR, IR, GR, SR.HI,
U(x) =V(x)
Obtaining each single criterion utility function and incorporating equations 1 through 6 into either one of models 1
through 4, will result in a deterministic network–level bridge maintenance model. As aforementioned, there exists
significant uncertainty in describing the deteriorating process of the bridge components. Thus, a reliable
maintenance planning should be able to guarantee that the scheduled maintenance scenarios can keep the structure
safe even under the worst conditions. This phenomenon is termed as robust maintenance planning in this study.
, ∈ ( , )
.
Max =∑ ∑ ∈ 102 ∗ 1 − ∗
s.t.
∑ ∑ ∈ ≤
∑ ∑ ∈ ≥
∑ ∑ ∈ ≥
∑ ∈ = 1, = 1,2, … ,
∈ {0,1}, = 1,2 … . , , ∈
Where
is the utility function to be maximized, since HI is an index available in most bridge inventories , a utility
function sensitive to HI is sought after
For this simple network of 3 bridges, as can be observed the total utility benefit of the network has reached, 235.889,
for bridge #1, alternative 2, for bridge #2, alternative 1 and for bridge #3, alternative maintenance3 has been proved
to be optimal , amongst the 3*3=9 alternatives available. The total cost of the network maintenance and repair has
added up to 8469 monetary units.
References
Ok SY, Lee SY, Park W. Robust Multi-objective maintenance planning of deterioration bridges against
uncertainty -42.
NCHRP Report 590. Multi-Objective Optimization for Bridge Management Systems. Transportation
Research Board, 2007.
ﻫﺎ ﺗﺎﯾﯿﺪ آﯾﺎ ﻣﺪل اراﺋﻪ ﺷﺪه و ادﻟﻪ اﻧﺘﺨﺎب آن ﻗﺎﻧﻊ ﮐﻨﻨﺪه ﻣﯽ ﺑﺎﺷﺪ؟ و
ﻣﯽ ﻓﺮﻣﺎﯾﯿﺪ؟
ﭼﺮا در اﯾﻦ ﻣﺪل ،ﺗﺎﺳﻒ ﻫﻤﯿﺸﻪ ﺻﻔﺮ اﺳﺖ؟ در ﺣﻘﯿﻘﺖ ﻣﺪل ﺑﺎ ﺗﮑﯿﻪ ﺑﺮ ﺗﻨﻬﺎ ﻣﺘﻮﺳﻂ ﻋﻤﻠﮑﺮد در ﺣﺎل ﺑﻬﯿﻨﻪ ﺳﺎزی اﺳﺖ ﯾﻌﻨﯽ از ﺑﯿﻦ ﻣﻮ و
ﻻﻣﺒﺪا ،ﻣﻮ ﻋﻤﻼ ﺑﻼ اﺳﺘﻔﺎده ﻣﺎﻧﺪه اﺳﺖ.
ﭼﮕﻮﻧﻪ ﻣﯽ ﺗﻮان از ﺷﯿﺖ ﻫ ﺎی اﮐﺴﻞ آراﯾﻪ ﻫﺎی ﺳﻪ ﺑﻌﺪی ﺧﻮاﻧﺪ؟ ﻣﻦ ﭼﻮن راه آن را ﻧﯿﺎﻓﺘﻢ ﻣﺠﺒﻮر ﺷﺪم از ﻣﺜﺎل ﻋﺪدی ﺳﺎده ای اﺳﺘﻔﺎده
ﮐﻨﻢ ﮐﻪ ﺑﻪ ﺗﻌﺪاد ﮐﻤﯽ داده داﺷﺘﻪ ﺑﺎﺷﺪ.
در ﺿﻤﻦ اﮔﺮ ﻣﺮاﺣﻞ ﻣﺪل ﺗﺎ اﯾﻨﺠﺎ ﻣﻮرد ﺗﺎﯾﯿﺪ ﺟﻨﺎﺑﻌﺎﻟﯽ اﺳﺖ ،ﻣﺮﺣﻠﻪ اﻋﺘﺒﺎر ﺳﻨﺠﯽ ﻣﺪل را در ﮔﺰارش ﺗﮑﻤﯿﻠﯽ ﺧﺪﻣﺘﺘﺎن اراﺋﻪ ﺧﻮاﻫﻢ داد
ﺑﺎ ﺗﺸﮑﺮ
زﻫﺮه ﻋﻠﯿﺰاده
۹۳۸۷۱۸۲۸