Deterioration Prediction of Building components

Pedram Keshavarzrad, Sujeeva Setunge, Guomin Zhang

Melbourne RMIT University

School of Civil, Environment & Chemical Engineering
January 2014

Abstract: Local government agencies in Australia manage around 140,000 community buildings
which provide essential services to the community. These are low to medium rise structures which
often have varying functional requirements and usage. Building assets are the second largest asset
group managed by local government. Efficient management of these assets requires understanding of
deterioration of building components, identifying effective condition monitoring methods, forecasting
deterioration and resulting maintenance expenditure, decision making considering risk, cost and
sustainability throughout life cycle of assets. The current approach adopted by most councils is
reactive decision making based on condition data collected at a given point of time.

Due to the large number of components forming a building, deterioration prediction of buildings can be
complex. For example, the IPWEA publication, NAMS uses a building hierarchy with 320 inspectable
components for defining community buildings. Deterioration prediction of a building requires
understanding of the deterioration of components of buildings and the resultant effect on a complete
building. The paper presents a method of deriving building component deterioration curves using the
NAMS useful lives, percentage change in conditions and a five level condition rating scheme. Using
the proposed method, deterioration curves have been derived for 320 building components defined in
NAMS. The curves are then validated using the condition data collected by a local council in
Melbourne.

Introduction The second step in developing an asset

management plan for buildings requires an
Community buildings managed by 700 local
accepted condition monitoring plan. The current
government agencies in Australia utilise around
data collection methods for community buildings
$250 billion capital investment yearly. These
use a discrete condition rating based on regular
require an integrated asset management plan to
visual inspection. Most local councils (LCs) use a
optimise service delivery at constrained budgets.
1-5, 1-10 or 0-10.
Moreover, significant aspects of service to
community should be addressed in decision
making by this sector. As the third step in proactive decision making, a
The first step in establishing an integrated asset good deterioration forecasting model is required at
management plan for buildings is establishment of component level. This is a current major gap in
an acceptable building hierarchy which allows knowledge. IPWEA (NAMS, 2009) presents a
division of buildings in to elements. IPWEA (2009) generic deterioration curve combined with useful
has developed a guideline for building asset lives for building components. These can be used
management which includes a lowest level as a basis. However, deterioration prediction
division of a building in to 320 elements. This requires consideration of data collected during
provides a standard method of dividing a building visual inspections as well.
in to elements for effective management.
In the final stage of an integrated AM plan, be the inability to analyse networks with high level
deterioration curves should be integrated with a of condition data variables (Morcous et al. 2002).
decision making process, which will allow
Artificial Neural Networks (ANN), Fuzzy set theory
proactive decision making instead of the current
and Case-Based Reasoning (CBR) are artificial
reactive methods.
intelligence (AI) models, which have been used for
Work presented here presents derivation of deterioration prediction of infrastructure. Al-
deterioration curves for 320 building components Barqawi & Zayed, (2006) used ANN for modelling
used in NAMS using a condition rating scheme of deterioration trend of sewer constructions, bridge
1-5. Preliminary curves have been derived using deterioration (Cattan & Mohammadi, 1997); oil &
the NAMS guidelines and then are converted to gas pipe lines (Sinha & pandy,2002). Most of AI
transition matrices for Markov process which based techniques need a large number of data
allows use of a probabilistic approach complicating the process of prediction for large
incorporating the stochastic nature of number of input data.
deterioration.
Statistical models have been used for
phenomenon affected by random variables leading
to uncertain outputs. As the probability of transition
Methodology
an asset from a condition to another condition has
Deterioration prediction is a significant stage in not been recognized and will be a function of
whole life cycle of building management process variables, it can be appropriate for modelling
(Edirisinghe et al.,2010). In most of the asset deterioration prediction.
management frame works, there is an evidence of
Among probabilistic models Markov process has
forecasting deterioration which is used to evaluate
been used for time-dependent statistical models.
serviceability and risks of failure.
Markov chain as a discrete time Markov process
Deterioration trend of community buildings has has been used in deterioration prediction in
been always affected by useful life cycle of key bridges (Madanat & Ibrahim, 1995), and sewers
components. Determining the useful life of a (Baik et al., 2006). In 2007 Sharabah introduced a
component is based on different variables related weighting model for building assemblies using
to each organisation policy and asset Markov process data collected from Victorian city
management plan. These variables can be altered councils.
by influencing different factors e.g. financial
Gamma process is another statistical model used
constraints, serviceability, functionality, capacity or
in time failure and rate of deterioration. It has been
others even age of an asset.
used on creep of concrete (Cinlar et al.,1977),
The models used for prediction of deterioration fatigue crack growth (Lawless &M.Crowder, 2004).
trend can be classified in three categories;
A previous research project conducted at RMIT
Deterministic models, statistical/stochastic models
University (Mohseni et al, 2012; Kalutara et al,
and Artificial Intelligence models (Tran 2007).
2013) delivered a conceptual framework for asset
Deterministic models are linear or non- linear management of community buildings and
methods and often used when components have a developed a probabilistic deterioration model
certain relationship with each other. Moreover, in based on condition data. The framework has been
linear methods condition is an exact value and can developed in to a working software tool for asset
be estimated by measurable or observable factors. management. However, the deterioration models
There are examples of time linear and power low developed focussed on building component
models for water mines and pavements (Kliener groups at a higher level due to the limited
and Rahani, 2001 and Lou et al.2001). availability of condition data.
Exponential deterministic models for pipe have
This paper aims to address the gaps in knowledge
been introduced by Mishlani & Madanat (2002).
in forecasting deterioration at the building
Deterministic models are simple and
component level to enable decision making for
understandable by practitioners but limitations can
proactive asset management strategies.
The suggested prediction method is based on deterioration which is a function of condition- life is
combination of deterministic methods; factor possible when actual deterioration of material is
model, life cycle analysis model and probabilistic gathered and analysed (NAMS 2009).
method using Markov Chain process. In fact,
Each organization has its own policy for modelling
deterioration prediction is divided in two stages.
and rating the condition of their assets and
The first is determination of the pace of trend
definition of required performances that will be
considering the factors accelerating or
based upon different factors such as serviceability,
decelerating the trend of deterioration of
financial constraints, safety, maintenance and
components; and the second is analysing the
functionality.
sequence of deterioration from a higher level
condition to a lower condition by using Markov Performance requirements can be measured by
chain transition process. In IPWEA publication comparing current level of service to the planned
which has been developed in consultation with a target and can be linked to acceptable conditions
large number of local councils through New meeting standards, organization policies and
Zealand and Australia (Australian Building customer value.
Component Guidelines- NAMS 2009), the most In this paper, a five-level-condition rating which is
common components of a building have been simple to implement has been selected for
classified in to three levels, component group, developing the idea. Since the other most
component type and more than 320 components. common rating scheme is 1-10, mapping of other
For each component, a useful life cycle has been schemes to the selected scheme is considered to
introduced. For the first stage, the useful life cycle be acceptable.
of the components have been selected from
NAMS 2009. Residual Useful Life Cycle & Condition Rating

ISO 15686 Part 2 recommended considering Using life cycle of components, it will be possible
factors impacting the useful life cycle of the to estimate the remaining useful life (residual life
assets, by multiplying coefficients by initial cycle typically used in water and pipe industries).
predicted useful life of an asset. Factors such as weather, attributes of an asset
and geographical considerations and functionality
Markov chain, which is the discrete time are the variables of determination of an asset
Markovian process, presents the stochastic remaining life. As mentioned before, for
process illustrating the sequence of phases developing the idea, NAMS guide lines have been
moving from a condition to another condition considered for predicting the useful remaining life
through discrete time with an uncertain manner. In of a component. Figure 1 depicts the method of
this model future conditions are related to present calculating the useful life cycle.
condition without considering previous conditions.
In Markov chain process, the sequence of moving Remaining Life = Designed life – Long life to date
from a state to other state has been identified
through a matrix, named transition matrix. The
transition matrix presents a set of finite conditions
(1,2,3,4,5,……..) and pij ; identifying the probability
of moving from condition i to condition j. As there
has been a need to at least three consecutive sets
of condition reports for calibration of transition
matrix, in this model, information derived from
analysing component deterioration based on
NAMS 2009 guidelines is used.
Figure 1: Remaining Life
Condition assessment that evolves the concept of
remaining effective life of an asset can be
estimated through manufacturer manuals,
Also condition assessment which is a five rating
questionnaires and reliable references. In fact,
condition assessment is stated in Figure 2.
 State A State B State C State D

State A 0.4 0.3 0.2 0.1

State B 0 0.2 0.4 0.4

State C 0 0 0.2 0.8

Figure 2: Condition rating
State D 0 0 0 1

Discrete Time Markov Chain

Discrete Time Markov Chain is a finite-state Figure 3: A typical Transition Matrix
stochastic process in which the defining random
variables are observed at discrete points in time
(Sharabah et al., 2006). If an element is in state Figure 3 shows a typical transition matrix. The
“i”, there is a fixed probability, Pij of it going in to probability of an element being in a given state at
state j after the next time step. Pij is called a a given point of time can then be depicted by the
“transition probability”. The matrix P whose ij
th set of curves shown in Figure 10.
entry is Pij is called the transition matrix. It also An initial distribution ‘v’ is a single row matrix
consists of a set of set of finite states S representing the number of elements in each
(1,2,3,………n). Markov Chain should satisfy two state. In Markov chain after one time step the new
conditions: distribution will be the result of multiplying initial
Pij ≥ 0, and ∑ Pij≤ 1 distribution v by the transition matrix P

This mean if an element is in state i, there is a (Pii) Distribution after 1 step: vP

probability that this element will stay in state i, and The distribution one step later, obtained by again
(1- Pii ) will move to next state j. multiplying by P, is given by
Present state at time t is i: Xt = i (vP)P = vP
2.`

Next state at time t + 1 is j: Xt+1 = j Therefore distribution after 2 steps = vP

2

Conditional Probability Statement of Markovian Similarly, the distribution after n steps can be
Property: obtained by vP
n

Pr{Xt+1 = j | X0 = k0, X1 = k1,…,Xt = i} = Pr{Xt+1 P2 is the two-step transition matrix for the system.
= j | Xt = i} ---------(1) Similarly, P3 is the three-step transition matrix,
Discrete time means t  T = {0, 1, 2, . . . } and Pn is the n-step transition matrix. This means
that the ijth entry in Pn is the probability that the
system will pass from state i to state j in n steps.
0.4 For calibration of transition matrix there is a need
for introducing the probability of transition
0.1 0.3 elements. One of the major challenges for
A calibrating the transition matrix has been sourcing
0.2 condition data, to establish the probability of
transition. Currently asset owners have used
1 D 0.4 B 0.2
inspection data for calibration of transition matrix.
0.8 0.4

C
0.2
Calibration of Transition Matrix The condition rating based on remaining life
(NAMS, 2009) is suggested as:
As discussed before, a Transition Matrix
represents the probability of moving from a state Condition 1-2 = %100- %55 of Remaining Useful Life
to other state or even remaining in current state. Condition 2-3 = %55-%37 of Remaining Useful Life
Figure 1 depicts sequence of the deterioration
Condition 3 -4 = %37-%25 of Remaining Useful Life
recommended by NAMS 2009 for core approach
planning and in rating condition assessment, Condition 4-5 = %11-%25 of Remaining Useful Life
which gives a trend for an element altering from a Figure.4 shows the remaining life and condition
higher condition to lower conditions during its long considering the base useful life equal to 43 years
life. The mentioned curve can be broken down in regarding NAMS, 2009 guidelines
to 4 lines with different equations and trends that
can represent the trend of transition moving from
one condition to another condition.
The co-efficient of X on each line is the slope of
the line and on the other hand reflects the trend of
the line or transition from condition to other Condition 5 4 3 2 1
condition. Using these co-efficient, each line of an
Life (years) 38.27 32.25 27.09 19.35 0
n x n matrix can be shaped which is introduced as
transition matrix and represents the trend of
Remaining 4.73 10.75 15.91 23.65 43
transition from condition to other condition. Whilst life (Years)
this is an approximate method, it addresses a
major gap in decision making at the component Figure4: Remaining Life – Condition Table
level.
Deterioration Curves
Using above table for remaining life a typical
Once the transition matrix is developed, future deterioration trend curve will be derived (Fig. 5).
conditions can be evaluated by multiplying a
vector “v”, which presents the initial condition (eg.
1 0 0 0) by the transition matrix. A table will
be formed representing the sequence of transition
from one state to other states during the life cycle
of an asset/component. VP shows the transition
for the first year. By powering the transition matrix
(P) by 2,3,4,………… it will be possible to provide
the probability of condition transition for
n
2,3,4,…………. years. In fact, VP provides the
condition for year n.
Using calculated data it will be possible to derive Figure5: Deterioration Curve
preliminary expected deterioration curves for
building components. These preliminary curves
can be improved with regular collection of By breaking down the curve to the lines
inspection data. representing the transition trend, transition matrix
will be formed (Fig. 6).
Demonstration of the method
More than 320 curves have been derived for
community buildings presenting the deterioration
trend for elements.
From gathered information from Kingston city
council, Fire service Category has been selected Figure6: Transition Matrix
to demonstrate the method
After multiplying transition matrix by initial Fig.9 depicts the sequence of condition transition
condition vector, deterioration table will be formed of the asset/component based on its age without
16250 8937.5 6012.5 4062.5 1787.5
considering the maintenance or rehabilitation
0
1
1
2
0
3
0
4
0
5
0 1
measures.
1 0.9483 0.0517 0 0 0 1.0517
2 0.899273 0.094058 0.006669 0 0 1.107396
3 0.85278 0.128417 0.017509 0.001294 0 1.167316
4 0.808692 0.15594 0.030678 0.004476 0.000215 1.231582
5 0.766882 0.177633 0.044843 0.009684 0.000958 1.300202
6 0.727235 0.194366 0.059058 0.016776 0.002565 1.373072
7 0.689637 0.206891 0.072674 0.025449 0.00535 1.449985
8 0.653982 0.215856 0.085264 0.035323 0.009575 1.530651
9 0.620171 0.221822 0.096568 0.046 0.015438 1.614713
10 0.588109 0.225269 0.106449 0.057099 0.023074 1.701761
11 0.557703 0.226615 0.114858 0.068271 0.032553 1.791355
12 0.52887 0.226215 0.121809 0.079221 0.043886 1.883037
13 0.501528 0.224376 0.127359 0.089701 0.057036 1.976343
14 0.475599 0.22136 0.131596 0.099518 0.071927 2.070815
15 0.45101 0.217393 0.134622 0.108528 0.088447 2.166008
16 0.427693 0.212667 0.136549 0.116629 0.106462 2.261501
17 0.405581 0.207344 0.137493 0.123759 0.125823 2.356898
18 0.384613 0.201566 0.137566 0.129889 0.146367 2.451832
19 0.364728 0.195448 0.13688 0.135015 0.167928 2.545967
20 0.345872 0.189092 0.135538 0.139157 0.190341 2.639004
21 0.32799 0.18258 0.133637 0.142352 0.213441 2.730673
22 0.311033 0.175985 0.131264 0.144647 0.237071 2.820739
23 0.294953 0.169363 0.128501 0.146101 0.261083 2.908998
24 0.279704 0.162764 0.12542 0.146777 0.285335 2.995277
25 0.265243 0.156228 0.122085 0.146744 0.3097 3.07943
26 0.25153 0.149788 0.118554 0.146069 0.33406 3.161341
27
28
0.238526
0.226194
0.143469
0.137294
0.114877
0.111098
0.144821
0.143067
0.358307
0.382347
3.240914
3.31808
Fig.9: condition transition
29 0.2145 0.131277 0.107256 0.140871 0.406097 3.392787
30 0.20341 0.125432 0.103383 0.138294 0.429481 3.465004
31 0.192894 0.119767 0.099508 0.135393 0.452438 3.534714
32 0.182921 0.11429 0.095653 0.132223 0.474913 3.601916
33 0.173464 0.109004 0.09184 0.12883 0.496862 3.666622
34
35
0.164496
0.155992
0.10391
0.09901
0.088084
0.0844
0.125261
0.121556
0.518248
0.539041
3.728855
3.788645 Typical Condition – Age curves
36 0.147927 0.094303 0.080799 0.117752 0.55922 3.846034
37 0.140279 0.089786 0.077289 0.11388 0.578766 3.901069
38
39
0.133027
0.126149
0.085456
0.081309
0.073877
0.070569
0.10997
0.106047
0.59767
0.615925
3.953802
4.004291
A few of condition – Age curve has been depicted
40
41
0.119627
0.113442
0.077342
0.07355
0.067367
0.064275
0.102134
0.098249
0.633529
0.650483
4.052596
4.098781 in the following graphs in Figure 10 that are
42 0.107578 0.069927 0.061294 0.094409 0.666793 4.142913
43
44
0.102016
0.096742
0.066468
0.063168
0.058423
0.055664
0.090628
0.086918
0.682465
0.697509
4.185058
4.225285
derived using above mentioned method.
45 0.09174 0.060021 0.053014 0.083288 0.711937 4.263662
46 0.086997 0.057021 0.050472 0.079747 0.725763 4.300258
47 0.082499 0.054163 0.048036 0.076301 0.739001 4.335141
48 0.078234 0.051441 0.045704 0.072954 0.751667 4.368378
49 0.074189 0.04885 0.043473 0.06971 0.763777 4.400036
50 0.070354 0.046384 0.041341 0.066572 0.775349 4.430179
51 0.066717 0.044038 0.039304 0.063541 0.7864 4.458871

Fig. 7: Transition Table

The last column on the right in the table (Fig. 7)

shows the sequence of deterioration trend which
can be used for providing expected deterioration
curve (Fig. 8)
Foundation

Figure 8: Expected Deterioration Curve

Ceiling
 Discussion
Asset management has always been a challenge
for local councils. Buildings are complex assets
where an integrated plan requires consideration of
deterioration trends of a large number of
constituent elements. Lack of a standardised
method for condition rating, deterioration
prediction and risk assessment is a major gap in
achieving optimised decision making. Most of
assessments are based on experience of
inspectors, and asset owners.
Deterministic methods for condition assessment
have limitations. For improving the reliability of
Electrical
deterministic methods, it is necessary to specify a
relationship between variable and stochastic
events that cannot be addressed through these
methods. Probabilistic methods, Markov Chain,
need transition matrices that rely on at least three
sets of condition data to be formed, which
however in most cases are not available. The
proposed method allows derivation of deterioration
trends based on NAMS guidelines, which then can
be validated using published data.
Another important debate is regarding
aggregation of components conditions for
estimating the condition of a building. Currently, a
Facilities Condition Index (FCI) defined as below
Ext. Wall provides a method of aggregating component
Fig 10- Typical Condition – Age Curves conditions to derive a facility condition.
Facility Condition Index =

Maintenance, Repair & Replacement Deficiencies

Cost
Significance of the outcome Current Replacement value of the Facility

By using IPWEA guidelines and Markov chain

process, more than 320 deterioration curves have
been derived using the method demonstrated FCI Condition of building Portfolio
above. 0−2% Excellent
2−5% Good
A building deteriorates due to degradation of 5−10% Fair
materials, elements, assemblies and services. By 10−15% Poor
using deterioration trend curves, as a forecasting >15% Very Poor
tool, it will be possible to track condition of assets Figure 11: Condition relation with FCI
in component level without the need of initial
physical inspections and when no reliable By improving deterioration forecasting and
condition data is available This will assist in condition aggregation, it will be possible to track
optimising asset data collection in near future. and assess the condition of assets without
Further, it can be used for identifying critical physical inspections leading to reductions in
elements to aid in decision making (maintenance additional budget.
or replacement) before failure.
Conclusion Case-Based Reasoning. Journal of
Infrastructure Systems, ASCE, 2002. 8(3): p.
Work presented in the paper leads to following
86-95.
conclusions:
1. A five stage condition rating scheme has 7) Mishalani, R.G. and S.M. Madanat,
been proposed for building assets based Computation of Infrastructure Transition
on IPWEA guidelines; Probabilities Using Stochastic Duration
2. Using the proposed rating scheme, Models. Journal of Infrastructure
deterministic and probabilistic Systems, ASCE, 2002. 8(4): p. 139-148.
deterioration curves have been derived for
320 building components; 8) Al-Barqawi, H. and T. Zayed, Condition
3. Derived curves have been validated using Rating Model for Underground
data collected by a local council; and Infrastructure Sustainable Water Mains.
4. A Facilities Condition Index has been Journal of Performance of Constructed
proposed as a method of aggregating Facilities, 2006. 20(2): p. 126-135.
condition to derive building condition as a
function of condition of components. 9) Cattan, J. and J. Mohammadi, Analysis of
Bridge Condition Rating Data Using
Neural Networks. Computer-Aided Civil
References Engineering, ASCE, 1997. 12(6): p.
419-429.

1) Edirisinghe, R., et al. Council Building 10) Sinha, S.K. and M.D.Pandey,
Management practices, Case studies and Probabilistic Neural Network for Reliability
Road Ahead. in The 5th World Congress on Assessment of Oil and Gas Pipelines.
Engineering Asset Management Computer-Aided Civil and Infrastructure
(WCEAM). 2010. Brisbane, Australia. Engineering, 2002. 17(5): p. 320-329.

2) Tran, H.D., Investigation of Deterioration 11) MOHSENI, H., SETUNGE, S., ZHANG,
G., EDIRISINGHE, R. & WAKEFIELD, R.
Models for Storm Water Pipe Systems.2007,
2011. Deterioration Prediction for
Victoria University. Community Buildings in Australia.
International Journal of the Constructed
3) Kleiner, Y. and B. Rajani, Comprehensive Environment, 1.
Review of Structural Deterioration of
Water Mains: Statistical Models. Urban Water 12) Madanat, S., R. Mishanlani, and W.H.W.
Journal, 2001. 3: p. 131-150. Ibrahim, Estimation of infrastructure
transition probabilities from condition rating
4) Lou, Z., et al., Application of Neural data. Journal of Infrastructure
Network Model to Forecast Short-Term Systems 1995. 19(2): p. 120-125.
Pavement Crack Condition: Florida Case
Study. Journal of Infrastructure 13) Madanat, S. and W.H.W. Ibrahim,
Systems, ASCE, 2001. 7(4): p. 166-171. Poisson Regression Models of Infrastructure
Transition Probabilities. Journal of
5) Mishalani, R.G. and S.M. Madanat, Transportation Engineering, ASCE, 1995.
Computation of Infrastructure Transition 121(3): p. 267-272.
Probabilities Using Stochastic Duration
Models. Journal of Infrastructure 14)Baik, H.S., H.S. Jeong, and D.M.
Systems, ASCE, 2002. 8(4): p. 139-148. Abraham, Estimating Transition
Probabilities
6) Morcous, G., H. Rivard, and A.M. Hanna,
in Markov Chain-Based Deterioration
Modeling Bridge Deterioration Using
Models for Management of Wastewater
Systems. Journal of Water Resources
Planning and Management, ASCE, 2006. 17) Lawless, J. and M.Crowder, Covariates
132(1): p. 15-24.) and random effects in a gamma process
model with application to degradation and
15) Sharabah, A., S. Setunge, and P. failure. Lifetime Data Analysis 2004.
Zeephongsekul, A Reliability Based 10(3): p. 213-27.
Approach
for Service Life Modeling of Council Owned 18) National Asset Management Strategies
Infrastructure Assets., in the NAMS 2009
ICOMS Asset Management Conference.
2007: Melbourne. 19) ISO 15686

16) Cinlar, E., Z.P. Bazant, and E. Osman, 20) International Asset Management Manual
Stochastic process for extrapolating 2006, 2011 (IIMM)
concrete creep. Journal of the Engineering
Mechanics Division, 1977. 103: p.
1069–88.