Goforth Et Al 2022 Network Analytics For Infrastructure Asset Management Systemic Risk Assessment

Network Analytics for Infrastructure Asset Management
Systemic Risk Assessment

Eric Goforth, S.M.ASCE 1; Wael El-Dakhakhni, F.ASCE 2; and Lydell Wiebe, M.ASCE 3
Abstract: The ever-increasing investment gap for deteriorating infrastructure has necessitated the development of more effective asset
management (AM) strategies. However, information asymmetry among AM stakeholder silos has been recognized as a key challenge
in implementing effective AM strategies. The connectivity within the AM system introduces systemic risks (possibility of dependence-
Downloaded from ascelibrary.org by 139.255.192.18 on 03/13/24. Copyright ASCE. For personal use only; all rights reserved.
induced cascade failure) to the entire AM system operation when information asymmetry occurs. This study describes a toolbox to enable
asset management stakeholders to assess such systemic risks through a network analytics approach. The network, representing the AM
system, is examined through its centrality measures to identify the most critical subject areas within the AM system. These subject areas
are subsequently paired with assets’ key performance indicators (KPIs). Within the developed toolbox, descriptive analytics provide trans-
ferrable KPI insights between stakeholders to reduce key asset information asymmetry. In parallel, predictive analytics forecast KPIs, ensur-
ing stakeholder awareness of future asset performance to allow for appropriate preparation. Subsequently, prescriptive analytics employ
heuristic-based optimization for optimal configuration of the AM network. The five tools presented are as follows: (1) dependence iden-
tification and network modeling; (2) network centrality analysis; (3) descriptive analytics of critical subject area paired KPI; (4) KPI-based
predictive analytics; and (5) prescriptive analytics for optimal network configuration. The utility of the developed toolbox is demonstrated for
Tools 1–3 using a real AM system network and KPIs associated with power transmission infrastructure outages. Based on the analyses,
managerial insights are drawn to illustrate the usefulness of the developed approach in improving information asymmetry within the AM
system, subsequently mitigating dependence-induced systemic risks. DOI: 10.1061/(ASCE)IS.1943-555X.0000667. This work is made
available under the terms of the Creative Commons Attribution 4.0 International license, https://creativecommons.org/licenses/by/4.0/.
Author keywords: Analytics; Information asymmetry; Infrastructure asset management; Key performance indicators (KPI); Network
analysis; Systemic risk; Transmission outages.
Introduction most infrastructure used daily by Canadians is more than 20 years

old and deteriorating rapidly. This report also noted that effec-
Infrastructure assets in Canada and the United States continue tive AM plan implementation and operationalization within asset-
to deteriorate each year, widening the gap in infrastructure spend- intensive organizations are critical to maximizing the impact of
ing needed to improve the asset conditions to serviceable levels limited resources. In this respect, Uddin et al. (2013) concluded
(Infrastructure Canada 2018; McBride and Moss 2020). The that infrastructure AM encompasses the systematic and coordinated
American Society of Civil Engineers (ASCE 2021) report gave planning and programming of investments, design, construction,
America’s infrastructure an overall grade of C–, up from a D+ maintenance, operation, and in-service evaluation of physical in-
in 2017. This modest improvement was partially attributed to asset frastructure and associated components. Additionally, Ross (2019)
owners adopting asset management (AM) techniques to prioritize concluded that AM can be described as the collective term for the
spending constrained by the limited funding resources; however, structured decision making and execution of plans to optimize the
the lingering low grade is attributed to massive maintenance back- balance among infrastructure performance, efforts, and risk through
logs, deteriorating infrastructure condition, and a lack of compre- the use of available assets as well as procurement of future assets.
hensive asset inventory and consistent condition data (ASCE
2021). The Canadian Infrastructure Report Card (2019) states that
Asset Management System Model
1
Ph.D. Candidate, Dept. of Civil Engineering, McMaster Univ.,
The Institute for Asset Management (IAM) developed the concep-
1280 Main St. West, Hamilton, ON, Canada L8S 4L7 (corresponding
author). ORCID: https://orcid.org/0000-0002-0552-4714. Email: goforte@ tual AM model in 2014 as a guide for AM professionals to imple-
mcmaster.ca ment and operate an AM approach in their organizations (IAM
2
Director of INTERFACE Institute and Professor, Dept. of Civil Engi- 2015). The IAM is the international professional body for AM pro-
neering and School of Computational Science and Engineering, McMaster fessionals, and generates AM knowledge, best practice guidance,
Univ., 1280 Main St. West, Hamilton, ON, Canada L8S 4L7. Email: eldak@ and awareness of the benefits of the AM discipline for individuals,
mcmaster.ca organizations, and wider society (IAM 2021). The AM system
3
Associate Professor, Dept. of Civil Engineering, McMaster Univ., model has also been referred to as the six box model, as there
1280 Main St. West, Hamilton, ON, Canada L8S 4L7. ORCID: https:// are six connected AM divisions (Strategy & Planning, Asset Man-
orcid.org/0000-0001-9754-0609. Email: wiebel@mcmaster.ca
Note. This manuscript was submitted on June 10, 2021; approved on
agement Decision-Making, Life Cycle Delivery, Asset Information,
October 14, 2021; published online on February 10, 2022. Discussion per- Organization & People, and Risk & Review), as shown in Fig. 1
iod open until July 10, 2022; separate discussions must be submitted for (IAM 2015) and summarized as follows. Strategy & Planning
individual papers. This paper is part of the Journal of Infrastructure Sys- aligns the organization’s AM activities to fit within a consistent
tems, © ASCE, ISSN 1076-0342. plan that has been developed and approved by all stakeholders.
© ASCE 04022006-1 J. Infrastruct. Syst.
J. Infrastruct. Syst., 2022, 28(2): 04022006

Fig. 1. IAM conceptual model. (Reprinted with permission from Institute for Asset Management 2015, © 2019 The Institute of Asset Management.)
Asset Management Decision-Making reviews the challenges and Background

makes decisions regarding how each of these stages occurs within
the main areas of an asset’s life: asset acquisition/creation; opera- Key to effective AM is the collective and coordinated effort that
tion; maintenance; and end of life disposal, decommissioning, or involves the collaboration of multiple stakeholders (e.g., engineer-
renewal. Life Cycle Delivery involves the entire lifespan of the asset, ing, operations, AM, finance, project management, and owner).
from acquisition/creation, through operation/maintenance, and These AM stakeholders form silos in the absence of necessary col-
finally, end of life disposal, decommissioning, or renewal. Asset laboration. These stakeholder silos have been shown to be the ma-
Information is typically input into an AM process, created or modi- jor hurdle in the implementation and operation of an effective AM
fied by a process, or the output of a process. Organization & People system within an asset-intensive organization (Pell et al. 2015; de
involves a review of the organizational structure, roles, responsibil- la Pena et al. 2016; Golightly et al. 2018). AM implementation and
ities, and contractual relationships. Risk & Review identifies the operation failures typically occur when stakeholders experience
risks related to an asset’s life cycle delivery, and understands information asymmetry because of inadequate information-sharing
and manages such risks; establishes a feedback mechanism within protocols and/or failure to readily share key information that could
the organization to allow for input on AM objectives, strategy, and mutually benefit their infrastructure’s AM system (Brunetto et al.
plan; and supports the continued improvement and development of 2014; Xerri et al. 2015; Golightly et al. 2018).
AM activities. Information asymmetry occurs when one party in a relationship
Overall, there are 39 AM system subject areas, outlined by the has more or better quality real-time or historical information than
Global Forum’s Asset Management Landscape (2014), across the another (Bergh et al. 2019). Such information asymmetry creates
six AM system divisions, as shown in Fig. 2. The 39 subject areas systemic (i.e., dependence-induced) risks within the AM system. In
were designed to illustrate the breadth of activities within the scope such a scenario, key stakeholders do not receive vital information
of AM, the interrelationships between activities and the need to and cannot respond in a timely manner to the impact of different
integrate them, and the critical role for AM to align with and deliver stakeholders’ decisions within the implementation and operation of
the strategic plan goals of an asset-intensive organization (IAM their AM system. In other words, they are precluded from using
2015). The connectivity among the AM subject areas within the real-time or historical information to support their decision making
AM system introduces systemic risks, such as the possibility of (Bergh et al. 2019). This, in turn, causes isolation of an AM subject
dependence-induced cascade failure. This might occur when func- area node or breaks some of its information flow links to other nodes,
tionality failure(s) through either one or more AM subject areas, potentially inducing a cascade failure throughout the AM system
nodes, or information flow links cascade throughout the remaining network. An example of systemic risk in a different industry oc-
functional AM subject areas, thus hindering relevant decision- curred in international banking in 2008, when the failure of Lehman
making abilities. Brothers caused the collapse of the global banking sector—and,

Fig. 2. IAM conceptual model subject areas. (Reprinted with permission from Institute for Asset Management 2015, © 2019 The Institute of Asset
Management.)
without government bailouts to other major banking institutions, 2017). Decentralized decision making in infrastructure restoration
the collapse cascade would have been far greater (López-Espinosa was shown to be an effective approach by Crowther (2008) as well
et al. 2015; Miller 2017). In this example, there was information as Talebiyan and Dueñas-Osorio (2020). The AM system is typi-
asymmetry in the mortgage-backed securities where risky mort- cally decentralized with AM stakeholders in charge of and mak-
gages were packaged as high-quality debt, leading to the seller having decisions pertaining to specific AM subject areas (Golightly
ing better information than the buyer (Tarver 2020). In this study et al. 2018). Thus, the current study views and analyzes the AM
through a network analytics lens, the systemic risks involved with stakeholders as a decentralized system, while still proposing a
implementing and operating an AM system pertain to information centralized information database solution that addresses the main
asymmetry due to either (1) node failure(s), which represent a spe- challenge of information asymmetry between AM stakeholders
cific AM subject area losing functionality and thus its ability to con- and ensures that AM stakeholders are not overwhelmed with too
tribute to the overall (AM system) network, or (2) link failure(s), much information.
which represent an interruption in the information flow between A decentralized system can be represented as a network consist-
AM subject area nodes. ing of connected nodes and links, forming a web of connected com-
It is also important to understand that organization structure can ponents (Barabási 2016). The nodes simulate the components of a
be either decentralized or centralized. In a decentralized structure, system, whereas the links represent the dependencies among these
the organization is divided into smaller teams in charge of specific nodes. Networks are often analyzed using specific measures related
aspects of the organization and decision making occurs at various to either system components (i.e., node-based or link-based) or the
levels within the organization (Graybeal et al. 2018). In a central- entire connected system (i.e., network-based). Node-based measures
ized structure, one or a select few individuals make the important focus on centrality analysis as it relates to the node’s importance in
decisions (e.g., resource allocation) and provide strategic direction the network by assessing the connectedness of that node to other
for the organization (Graybeal et al. 2018). A decentralized struc- network nodes. There are different centrality measures applied in a
ture offers many benefits for quick decision and response time and wide variety of applications (Derrible 2012; Lee et al. 2013; Estrada
skilled and specialized management. Johnson & Johnson, for and Knight 2015; Das et al. 2018; Ezzeldin and El-Dakhakhni
example, has successfully adopted this management structure across 2019; Goforth et al. 2020).
its over 200 operating companies (Weldon 2008; Mohamad et al. Analytics facilitate the realization of business objectives through
2017). However, drawbacks of a decentralized structure include co- reporting of data to analyze trends (i.e., descriptive analytics), creat-
ordination issues between teams working toward a company’s stra- ing prediction models for forecasting (i.e., predictive analytics), and
tegic goal, and individual teams prioritizing their own goals over optimizing processes to enhance performance (i.e., prescriptive ana-
the organization’s goals (i.e., teams operating as silos) (Vantrappen lytics) (Tsai et al. 2015; Delen and Ram 2018). Analytics have been
and Wirtz 2017). The benefits of a centralized organization struc- applied in various studies within multiple infrastructure industries to
ture include clarity in decision making, streamlined implementation improve AM processes. Descriptive analytics applications focus on
of policies and initiatives, and control over the strategic direction of deriving insights into performance trends from complex data,
the organization (e.g., Apple) (Graybeal et al. 2018). Drawbacks of mainly through visualizations (Abdelfatah et al. 2013; Barker et al.
the centralized system arise when employees have difficulty pro- 2017; Black et al. 2018; Mukherjee et al. 2018). Predictive analytics
viding feedback on operations, and lower management levels have applications use historical data within machine learning models to
limited flexibility to influence changes (Vantrappen and Wirtz predict an output (e.g., health index, condition, outage severity, or

asset remaining life) (Zhou et al. 2016; Dehghanian et al. 2019; descriptive, predictive, and prescriptive analytics using infrastructure
Yang et al. 2019; Piryonesi and El-Diraby 2020). Prescriptive ana- KPIs (e.g., average outage duration, bridge condition index, and the
lytics applications attempt to optimize intervention and maintenance number of water main failures per 1,000 km) and the AM network
planning and scheduling (Qiu et al. 2013; Chen et al. 2015; Heng structure specific to the organization owning, managing, and oper-
et al. 2016; Abu-Samra et al. 2020). Such applications have ating the infrastructure assets to reduce the information asymmetry,
demonstrated the benefits of using analytics to improve the specific with the aim of ensuring risk-informed and effective decision
subsets of AM, but a disconnect remains between the use of ana- making. This concept, applied in a manufacturing operational
lytics and the bigger picture that considers systemic risks within the performance study by Prajogo et al. (2018), illustrated that good
infrastructure AM system. information management practices within an organization can have
In addition, although studies have shown that analytics provide a a significant impact on overall business performance. The latter was
competitive edge when integrated into business processes (Delen accomplished through sharing information, using information tech-
et al. 2018; Scheibe et al. 2019; O’Neill and Brabazon 2019; Hassan nology tools within an organization, and sharing information with
2019), it is critical to first identify the key hubs within an organiza- supply chain partners. As such, Prajogo et al. (2018) emphasized
tional structure through which information flows, for the organiza- that organization management must look for ways to facilitate the
tion to operate effectively (McDowell et al. 2016). This concept has sharing and centralized management of information across internal
been studied in organizational networks, identifying key stakeholders and external organizational boundaries.
(e.g., companies, people, or departments) that are critical to the func-
tionality and effective operation of the organization (Barão et al.
2017; Ujwary-Gil 2019; Eisenberg et al. 2020). Nonetheless, to the Network Measures
best of the authors’ knowledge, the identification of AM systemic
risks with a method to reduce the information asymmetry within the Complex network theory allows for the modeling of complex sys-
AM system subject areas (i.e., hubs) is yet to be developed. As such, tem connections through a network of nodes and links (Boccaletti
in this study, a toolbox is created to integrate network analysis and et al. 2006; Barabási 2016; Salama et al. 2020). This section pro-
data analytics for an AM system model that incorporates the decen- vides a background of some relevant node- and network-based
tralized nature of the AM stakeholders and subject areas, and presents measures. Within the context of this study, nodes represent the main
a centralized database whereby AM subject area-specific information AM system subject areas and links represent the connections
between the subject areas. The links within the network are directed,
can be accessed by the stakeholders responsible for those subject
indicating information, documentation, knowledge, and/or policy
areas. This approach will ensure consistency across all AM subject
being transferred from a source node to a target node, and un-
areas while also preventing AM stakeholders from being overloaded
weighted, as each connection is viewed as equally important (unless
with information, allowing them to focus on only the information
otherwise specified) to the overall operation of the AM system. The
necessary to make decisions within the AM subject areas they are
level of connectedness within an AM system necessitates a network-
responsible for (Herrera et al. 2011; Prajogo et al. 2018).
based model to understand each node’s importance to the AM
This paper first outlines the study goals and objectives, followed
system. An adjacency matrix (A) can be formed that describes
by a description of the considered network measures. Subsequently,
the connectivity and disconnection between the AM system network
the developed toolbox is presented to describe five distinct tools that
nodes. Each element of the adjacency matrix, Aij , is either 1, illus-
identify critical AM system subject areas using network analysis
trating a direct connection between nodes i and j (i ≠ j), or 0
and that uses analytics with infrastructure asset key performance
otherwise (Barabási 2016). Specific node-based centrality measures
indicators (KPIs) to reduce the systemic risks caused by information
that relate to the AM network model are summarized below.
asymmetry between dependent AM subject areas. In the current
Betweenness centrality identifies nodes that play a central role
study, due to data restrictions, the utility of the toolbox is demon-
in connecting to other nodes in the network (Freeman 1977). The
strated using only the first three of five tools considering an AM
betweenness centrality of node i (BCi ) measures the total number
conceptual model developed by IAM and power transmission in-
of shortest paths passing through node i as expressed in Eq. (1)
frastructure outage KPIs. Finally, managerial insights are drawn to
illustrate how asset managers can reduce the systemic risks within X ρjk ðiÞ
an AM system. BCi ¼ ð1Þ
j≠i≠k
ρjk
where ρjk = number of shortest paths connecting node j to node k;

Study Goal and Objectives and ρjk ðiÞ = number of shortest paths connecting node j to node k
that traverse node i in the network.
The study goal is to mitigate systemic risks within an AM system
Closeness centrality represents how close a node is to all other
that are created by information asymmetry among dependent AM
network nodes (Estrada and Knight 2015). The closeness centrality
subject areas. This study attempts to break down the silos within
of node i (CCi ) is determined by finding the shortest path using
which infrastructure stakeholders operate, so all stakeholders use
either weighted or unweighted links for node i as
the same information to make decisions, ensuring cohesiveness
among stakeholders working toward their AM goals and objectives N−1
with a set of described tools. The described tools will enable AM CCi ¼ P ð2Þ
j dði; jÞ
stakeholders to identify the network structure of an AM system by
modeling the complex connections within such a system. In addi- where dði; jÞ = shortest path distances between nodes i and j; and
tion, the described tools will allow AM stakeholders to analyze the N = total number of network nodes.
resulting network to identify dependence-induced systemic risks Degree centrality assesses the relative influence of nodes as
to implementing and operating an effective AM system, achieved the number of degrees (links) that a node directly shares with
by identifying the critical subject areas within the network using other nodes (Estrada and Knight 2015). As such, the degree
network measures (e.g., node- and link-based centralities) specific centrality of a node i (DCi ) is defined using the adjacency matrix
to the AM system network structure. Finally, the tools will employ A ¼ ðai;j Þ as

X
N measure refers to network dependence and highlights the systemic
DCi ¼ aij ð3Þ risk from the cascading effects of failed AM system subject areas.
j¼1
This centrality measures the direct influence of a node on its Network Analytics Toolbox
connected nodes.
Eigenvector centrality quantifies the extent of node connected- To address the study goal and objectives, the following five tools
ness to other important (i.e., high degree centrality) nodes (Thai and were developed, as shown in Fig. 3: (1) dependence identification
Pardalos 2012). The relative centrality score of node i (xi ) for and network modeling, where the AM network structure is identi-
adjacency matrix A ¼ ðai;j Þ is fied and modeled; (2) network centrality analysis, to identify the
critical AM subject areas causing systemic risk; (3) descriptive
1X
xi ¼ a x ð4Þ analytics of critical subject area paired KPI, to develop targeted
λ j i;j j visualizations that focus on only the necessary information for de-
cision making relevant to specific AM subject areas; (4) KPI-based
where λ = a constant eigenvalue, and the equation could be predictive analytics, to forecast KPI metrics to enable more proac-
rearranged in vector notation as the eigenvector equation Ax ¼ λx. tive decision making; and (5) prescriptive analytics for optimal net-
This centrality indicates node importance as its connection to other work configuration, to minimize AM network systemic risks. Each of
important nodes and nonconnection to unimportant nodes in the these tools will be described in detail later in the paper to explain how
network. each tackles the study goal and objectives.
In addition to node failures, links might also fail, representing
information asymmetry between nodes. Therefore, it is important
to understand the importance of each link to the functionality of Tool 1: Dependence Identification and Network
the network. A centrality metric related to the importance of infor- Modeling
mation flow in a network is the link betweenness centrality (LBC) Tool 1 describes the process for identifying the network structure of
(Teixeira et al. 2016), defined as the number of shortest paths that an AM system. The first step in implementing Tool 1 involves mod-
traverse the link (Freeman 1977). In practical terms, the LBC is a eling an organization’s AM system in terms of its specific subject
measure of how central a link is to the network, and in the case areas that describe the implementation and operation of its AM sys-
of the AM system network, it measures the criticality of a specific tem as a network. In this respect, the connections between subject
link to information asymmetry. High-ranking links contribute sys- areas are identified based on expert AM knowledge. The links
temic risks to the network, as their failure would lead to a cascading between nodes represent a decision, information or data transfer,
failure throughout the network. Although the links within this study strategy, or policy that is passed from one node (source node) to
are specified as unweighted, future extensions of this toolbox might another (target node). It is these links that define the dependence
incorporate link types and weights that relate to the type or criticality between subject areas to form the AM system network. These links
of information that is passed between nodes (e.g., raw, preprocessed, are indicated in the adjacency matrix, as illustrated in Fig. 4, where
figures, decisions, or connections to AM standards). a link presence is indicated by a 1 and its absence is indicated by a
In addition to the aforementioned node- and link-based central- 0. As mentioned earlier, the links do not have an associated weight
ity measures, there are network-based measures that quantify the value, as the link represents the presence of a connection in the
connectedness of the overall network structure (Estrada and Knight form of a decision, information or data transfer, strategy, or policy
2015; Opdyke et al. 2017; Valentin et al. 2018). The most relevant that is passed from one source node to another (target one). The
to the AM network are described below. adjacency matrix can then be used to visualize the network, as illus-
Network density (ND) represents the ratio of actual links within trated in Fig. 4. For example, Fig. 4 shows 10 AM subject areas in
a network to the potential links that could be formed within the net- the adjacency matrix and illustrates a potential AM network model
work if the network were fully connected (Barabási 2016). It can be with nodes representing subject areas and links representing the
calculated using the following equation for a directed network: connections between subject areas.
l
ND ¼ ð5Þ Tool 2: Network Centrality Analysis
nðn − 1Þ
Tool 2 describes the process for analyzing the resulting network to
where l = number of links in the network; and n = number of nodes identify dependence-induced systemic risks to implementing and op-
in the network. The network density is a measure of the network’s erating an effective AM system. The AM network layout, as gener-
health and effectiveness. The ratio has values ranging from 0 (a ated using Tool 1, is employed by Tool 2 to calculate the centrality
completely unconnected network) to 1 (a fully connected network). measures—identifying the importance of each AM subject area
In AM applications, it can assess the level of connectedness within an AM system. Such centralities, illustrated in Fig. 5, highlight
between all subject areas in sharing information and indicate the potential node/link systemic risk in the AM system where the central-
susceptibility to network failure (for low values). ities are converted to a ranked list of node/link importance. In this
Average degree centrality (ADC) is the ratio of the summation respect, betweenness centrality is a measure of the importance of
of the degree centrality values for all nodes to the total number of the subject area to the overall implementation of the AM system
nodes in the network (Barabási 2016). It can be calculated using the within an organization. Closeness centrality is a measure of indirect
following equation for all i nodes: AM information flow between the not-directly connected nodes. De-
P gree centrality is a measure of the criticality of AM subject areas to
DCi
ADC ¼ ð6Þ the dependent subject areas. Eigenvector centrality is a measure of
n
node connectedness and importance to other highly connected nodes,
where DCi = degree centrality for node i; and n = total number of identifying subject areas that have a strong influence on other impor-
nodes in the network. This measure indicates how quickly disruptions tant AM subject areas. An organization would need to determine the
can diffuse throughout the network. Within the context of AM, this centrality measure that is most relevant to its implementation strategy.

Tool 1: Dependence Identification and
Network Modelling Tool 2: Network Centrality Analysis
AM Subject AM Betweenness Closeness Eigenvector
AM10 Degree
Area Division Centrality Centrality Centrality
AM1 AMD1 0 0.21 5 0.39
AM7 AM8 AM2 AMD3 374.09 0.26 19 1.00
AM1 AM2 AM3 AM4 AM3 AMD2 0 0.23 7 0.75
AM2 AM9 AM4 AMD4 275.09 0.3 13 0.88
AM5 AMD2 242.91 0.33 9 0.49
AM7 AM6 AM5 AM8 AM6 AMD3 378.03 0.31 9 0.43
AM6 AM5
AM7 AMD1 267.48 0.35 5 0.14
AM9 AM10 AM8 AMD4 83.15 0.3 5 0.13
AM1 AM9 AMD1 99.52 0.28 13 0.31
AM10 AMD2 34.48 0.28 10 0.26
AM4 AM3
Tool 4: KPI-Based Predictive Analytics

Tool 3: Descriptive Analytics of
Predict
Critical Subject Area Paired KPI Inputs: Y1
Feature X1
Output(s)
Feature X2
Feature X3 Y2
| Machine
Feature Xn Learning Y3
Model |
Output(s): Yn
Feature Y1
|
Feature Yn
Tool 5: Prescriptive Analytics for Optimal Network

Configuration
Prescriptive
Analytics
Fig. 3. Overview of the asset management network analytics toolbox.
Asset Management Conceptual Model
AM1 AM2 AM3 AM4

Network Model
AM7 AM6 AM5 AM8
AM9 AM10
Adjacency Matrix
AM1 AM2 AM3 AM4 AM5 AM6 AM7 AM8 AM9 AM10
AM1 0 1 0 1 1 0 0 0 0 0
AM2 1 0 0 1 0 1 1 1 1 1
AM3 0 1 0 0 1 1 0 1 1 0
AM4 0 0 0 0 0 1 0 0 0 0
AM5 1 0 1 0 0 1 1 1 1 0
AM6 1 0 0 1 0 0 1 0 0 1
AM7 1 1 0 0 0 1 0 1 1 0
AM8 1 0 1 0 0 1 0 0 1 1
AM9 1 1 0 1 0 0 1 0 0 1
AM10 1 0 0 0 0 1 1 0 1 0
Fig. 4. Tool 1: Dependence identification and network modeling of AM subject areas.
For example, an organization conducting a preliminary screening of Therefore, the importance ranking identifies the most critical subject
its AM structure would utilize eigenvector centrality to determine areas and links exposed to the dependence-induced systemic risk
subject areas that influence other highly influential subject areas, involved with the implementation and operation of an AM system.
allowing the organization to focus their attention on a few subject
areas to maximize the impact of improvement in their AM system. Tool 3: Descriptive Analytics of Critical Subject
When the calculated node- or link-based centrality measures are Area-Paired KPI
high, there is a greater likelihood of the AM network failing if such Tool 3 describes the process for implementing descriptive analytics
an important subject area or link were to become dysfunctional. for infrastructure asset KPIs. The subject areas most exposed to

Asset Management Network Network Measures Table
AM Subject AM Betweenness Closeness Eigenvector
Degree
Area Division Centrality Centrality Centrality
AM1 AMD1 0 0.21 5 0.39
AM2 AMD3 374.09 0.26 19 1.00
AM3 AMD2 0 0.23 7 0.75
AM4 AMD4 275.09 0.3 13 0.88
AM5 AMD2 242.91 0.33 9 0.49
AM6 AMD3 378.03 0.31 9 0.43
AM7 AMD1 267.48 0.35 5 0.14
AM8 AMD4 83.15 0.3 5 0.13
AM9 AMD1 99.52 0.28 13 0.31
AM10 AMD2 34.48 0.28 10 0.26
Ranked Subject Area Importance

Fig. 5. Tool 2: AM network centrality analysis and subject area importance ranking.
systemic risk, as identified from the centrality measures from Tool KPIs, as shown in Fig. 6. Descriptive analytics often includes build-
2, can be further analyzed using Tool 3 when paired with infrastruc- ing a KPI-tailored dashboard that allows user interaction to gain
ture-industry-specific KPIs. Tool 3 visualizations are designed to useful KPI insights (Wexler et al. 2017). Tool 3 facilitates clear
draw their information from a centralized database and only dashboards to be circulated among stakeholders to ensure that
display KPI information directly related to a AM subject area. Tool every stakeholder is informed on the KPIs related to the important
3 visualizations are designed to ensure that stakeholders focus only AM subject areas. Huang et al. (2019) showed that, when stake-
on the information necessary to make decisions within that AM holders can see the impact of their work, they are more likely
subject area rather than becoming overwhelmed by all AM infor- to develop trust in the management processes and therefore share
mation from the centralized database. For example, the subject information internally more readily. Therefore, Tool 3 facilitates
area, outage management, is paired with the KPI avg. outage du- clear visualizations to be circulated among stakeholders that man-
ration year-over-year and the trend in the avg. outage duration, as age the key AM subject areas, ensuring that these stakeholders see
shown in Fig. 6. Fig. 6 presents an example of two KPIs and their only the necessary information related to the decisions they need to
evolving values with time. It should be noted that only the KPI make within the AM subject area they are responsible for. Tool 3
values are expected to continue to change with time (i.e., dynamic) applications, in turn, allow for AM stakeholders to monitor their
as new information becomes available, whereas their pairing to the impact on the AM system for their specific AM subject areas,
systemic risk-critical AM subject areas is expected to remain and ensure that all decisions specific to each AM subject area
largely the same (i.e., static) given that the AM system network are made using consistent information.
is not expected to change with time.
Infrastructure KPIs are metrics of a specified asset or overall
infrastructure network performance. KPIs can be continuous or dis- Tool 4: KPI-Based Predictive Analytics
crete, and can also be qualitative (e.g., low, medium, high) or quan- Tool 4 describes the process for implementing predictive analytics
titative (e.g., 50%–70%) in nature. Although the KPIs are paired to for infrastructure KPIs. Building on the descriptive analytics of
the AM subject areas, making them static in terms of their evalu- Tool 3, Tool 4 is focused on developing a predictive analytics model
ation approach, their values are nonetheless expected to be dynamic for the aforementioned KPIs. Fig. 7 outlines the process for this tool
as they continuously change with time under different conditions by including historical KPI performance within a machine learning
(e.g., climate). For example, the pavement industry uses the pave- model to output forecasted KPI metrics. Machine learning models
ment condition index and international roughness index; the bridge are typically classified as either supervised (i.e., developing a math-
industry uses the bridge condition index; the power industry uses ematical function that maps the relationship between specific input–
the system average interruption frequency index, system average output pairs) or unsupervised (i.e., categorizing the dataset based
interruption duration index, and mean outage duration; and the on similarity, without pre-specifying outputs) (Zumel and Mount
water and wastewater industry uses the number of breaks per year, 2020). Examples of machine learning models include decision trees,
the number of failures per 1,000 km, and leakage of water per year artificial neural networks, and support vector machines (Aggarwal
(Alzoor et al. 2021; Uddin et al. 2013). Because the KPIs differ 2015).
between infrastructure industries, Tool 3 pairs the critically depen- Input data are employed to train the machine learning model to
dent subject areas of the AM system with the relevant KPIs. predict a numerical or categorical output based on the provided
Descriptive analytics can then be used to illustrate these paired contributing features (Hastie et al. 2009). This allows a decision

Top Ranked Subject Areas KPI
Key Performance
Indicator (KPI)
KPI1
KPI2
KPI3
KPI4
KPI5
KPI6
Descriptive Analytics
Fig. 6. Tool 3: Descriptive analytics to inform critical AM subject areas using paired KPIs.
maker to predict a KPI output value based on contributing features The KPI predictive analytics model would forecast future perfor-
(e.g., climatic conditions, economic conditions, geographic loca- mance, allowing stakeholders to be informed and proactively plan
tion, asset characteristics, time, and maintenance history) and past for more effective AM. Tool 4 applications improve the information
historical KPI values (Haggag et al. 2021). Any additional input symmetry in that all stakeholders are aware of and strive toward a
features would be infrastructure-industry-specific, therefore the clear performance goal for their critical AM KPIs, and ensure con-
organization would need to establish which features would be sistency among stakeholders that manage the AM subject areas be-
accessible before building a KPI predictive analytics model. Other cause they can make decisions based on consistent predictive
research studies have successfully predicted specific infrastructure models built specifically to each AM subject area-paired KPI.
KPIs within different industries in isolation (Zhou et al. 2016;
Dehghanian et al. 2019; Yang et al. 2019; Piryonesi and El-Diraby
2020), therefore a summary of some key techniques deployed in Tool 5: Prescriptive Analytics for Optimal Network
those studies to meet the goals of Tool 4 is provided below. Configuration
For example, Fig. 7 is presented as an illustration of this tool to Tool 5 describes the process for implementing prescriptive analyt-
show a forecasted 5-year period for the KPI, average outage dura- ics by optimizing the AM subject area network through adding
tion, used by the power industry. The input features include contri- links to the original network to minimize the average centrality
butions to the outage (e.g., outage cause, failure mode, and climatic measure of choice (e.g., betweenness, closeness, degree, or eigen-
information) and component or system characteristics (e.g., voltage, vector) or multiple centralities, depending on the objective function
affected component/system, time of the outage, and geographic (Thai and Pardalos 2012). The optimization problem would include
location). As would be expected, the outputs from Tool 4 applica- a constraint on the number of allowable links to add to the network
tions are only good for AM decision making if the information used before it becomes too centralized or nonfunctional and there would
for inputs is of high quality and pertains to meaningful data. Koziel be a cost per link addition in terms of new information, policy, or
et al. (2021) investigated the impact of using faulty data in AM decision that would be transferred. The optimization would mini-
decision making and found that there were significant implications mize the systemic risks in the AM network by reducing the impact
on optimal replacement schedules. Therefore, high-quality data of failure related to high centrality nodes or links through the ad-
must be gathered related to each AM subject area-paired KPI to dition or subtraction of links in the network, while still maintaining
facilitate reaching the most effective AM decisions. The two lines the functionality of the AM system network. Such optimization
in Fig. 7 indicate different organizations and their forecasted KPIs. would employ, for example, genetic algorithms or other heuristics,

Historical KPI Performance
Forecasted Future KPI
Machine Learning Model
Predict
Inputs: Output(s)
| Machine
Learning
Model
Output(s): |
Fig. 7. Tool 4: Predictive analytics to forecast future KPI performance using historical input features.
where a population of solutions is generated and evolves until a outage data gathered by the Canadian Electricity Association
(near) optimal solution is obtained (Goldberg 1989). Each solution (CEA). The IAM is the international professional body for AM pro-
within the population represents a single realization of the input fessionals, develops AM knowledge and best practice guidelines,
features (i.e., individual). New individuals are reproduced through and generates awareness of the benefits of the AM discipline for
special evolutionary operators including (1) elitism, where individ- individuals, organizations, and the wider society (IAM 2021). CEA
uals with greater fitness are replicated; (2) crossover, where sets membership includes generation, transmission, and distribution
of two individuals (i.e., parents) are selected based on predefined power utilities and industrial partners from across Canada (CEA
criteria (e.g., random selection or a selection based on the fitness 2020). The toolbox was applied in this setting to display its applica-
value) and subsequently mixed to produce new individuals; and tion within the asset-intensive power transmission industry.
(3) mutation, where single parents are altered randomly to produce
new individuals (Nearchou 2004; Scrucca 2013; Yosri et al. 2021). Project Description
Within the AM system network optimization, each link would
be represented as a feature within the individual and the values The IAM conceptual model was used for building the infrastructure
would be either 1, indicating link presence, or 0, indicating link AM network. To show the connection between important AM sub-
absence. The application of Tool 5 would present an optimized con- ject areas and industry-specific KPIs, a transmission infrastructure
figuration of the AM system such that the systemic risks would be asset outage dataset was obtained from the CEA, covering the
minimized, as illustrated in the new network configuration shown period from 1978 to 2018. The transmission infrastructure network
in Fig. 8. is critical to the reliable delivery of power from generators to sub-
stations and on to customers. Therefore, the effective and efficient
management of transmission infrastructure assets is critical for safe
Toolbox Application to Power Transmission and reliable power delivery. The transmission equipment outage
Infrastructure data are for equipment operating at high voltages of 60 kV and
above (CEA 2018). The outages are recorded for transmission in-
The developed toolbox was demonstrated using the IAM’s concep- frastructure components including transmission lines, cables, trans-
tual model for an AM network and transmission infrastructure asset former banks, circuit breakers, synchronous compensators, static

Prescriptive
Analytics
Fig. 8. Tool 5: Prescriptive analytics for optimal AM system network configuration.
Table 1. KPIs calculated and published in the CEA’s annual report network model shown in Fig. 9. A transmission utility AM system
would typically include the AM subject areas from Figs. 2 and 9.
KPI Definition
The node color refers to the AM division of the subject area. The
Frequency (per 100 km-year ) The number of outages divided by km network is directed, as typically subject areas pass knowledge, in-
years divided by 100 formation, and policy in only one direction (i.e., from source to
Frequency (per year) The number of outages divided by
target nodes). The link color is the same as the source node. Using
component years
Number of outages The number of major component-related color as a distinguishing feature allows for the identification of
forced outages clusters of AM division-based subject areas. This is shown in Fig. 9,
Total outage duration (h) Total forced unavailable time (i.e., the where the Asset Information and Organization & People division
time required to completely restore a subject area nodes are highly interconnected within their clusters.
component to service) of the component- Conversely, the Strategy & Planning division subject areas are not
related outages only clustered among themselves, but instead are highly connected
Average outage duration (h) Total outage duration divided by the with other subject areas.
number of outages Network modeling is useful for viewing node connections,
Median outage duration (h) 50% of the forced unavailability times are
whereas node-, link-, and network-based centrality analyses are
greater than this value
Unavailability (%) The product of frequency and average needed to identify the highly dependent nodes/links that induce
outage duration in years, expressed as a systemic risk to the AM network. Fig. 10 shows the top 10 subject
percentage of the component’s population areas for each of the previously described centrality measures. Of
note are the Strategy & Planning division subject areas of Asset
Management Strategy and Planning, Asset Management Planning,
and Strategic Planning. These areas all rank high for betweenness
compensators, shunt reactor banks, shunt capacitor banks, and series centrality, degree centrality, and eigenvector centrality. This indi-
capacitor banks. The KPIs recorded and published by the CEA in its cates that, for an organization to implement an effective AM system,
annual report are shown in Table 1 along with a definition of each it must have a strong AM plan and objective targets. In addition,
KPI metric. This demonstration of the toolbox will involve the the Operation and Maintenance Decision-Making and Resourcing
application of Tools 1 to 3 only, as the data needed for implemen- Strategy of the Asset Management Decision-Making division were
tation of Tools 4 and 5 are restricted by transmission infrastructure high-ranking subject areas among the centrality measures.
owner/operators for their internal use. As such, the demonstration The links shown in Table 2 illustrate the critical links contrib-
will focus on describing the utility of the toolbox for the identifi- uting to systemic risks within the AM system. The betweenness
cation of AM subject areas that are most critical to induce systemic centrality of each link was found per the network measure previ-
risk within an AM system. ously described. The links are ordered based on their criticality,
which indicates their importance to the network functionality. Of
note within Table 2, the Resourcing Strategy and Asset Management
Network Analysis Planning nodes have multiple important links, suggesting that it is
Based on available details from the IAM’s conceptual model sub- particularly important for these AM subject areas to have excellent
ject areas and the connections between subject areas as outlined by communication with the connected AM subject areas.
the Global Forum’s Asset Management Landscape, the adjacency In addition to the node-based centrality measures, the network-
matrix (Table S1 in the Supplemental Materials) was developed based measures are important to evaluate the overall resilience
(Global Forum on Maintenance and Asset Management 2014). of the network to potential failures (Barabási 2016). The average
The connections were specified within the report for each subject degree centrality of the AM subject area network is 3.15 and the
area as related subjects and artefacts. The colors of the subject areas network density is 0.08, meaning that only 8% of the potential links
in the adjacency matrix (Table S1) correspond to the AM divisions of a fully connected network connect the AM subject areas. This
from Fig. 2. Tool 1 uses the adjacency matrix to develop the implies that the network is vulnerable to systemic risk because, if

Fig. 9. AM system network.
one or more of the previously identified critical nodes/links were to slider menu along with the median KPI value among all organiza-
be disrupted, the AM system would be greatly impacted. tions over the selected period (e.g., 2014–2018). (2) The line graph
shows the changing KPI for each organization (color) over the
period indicated by the side slider menu. (3) The bottom point graph
Descriptive Analytics is based on the organization selected by the menu (titled Organiza-
Three of the critically dependent subject areas, as determined from tion) and is broken down by voltage class to highlight the long-
the node-based centrality analysis in Fig. 10, are used to illustrate duration outage events with an option to select a specific point to
the use of descriptive analytics for subject area-paired KPI analysis. display the outage features. Descriptive analytics in this application
Asset Management Strategy and Objectives, Asset Management will allow the key stakeholders to track the objective KPIs while
Planning, and Operations and Maintenance Decision-Making were also investigating the long-duration outages to understand and rem-
chosen for illustration because these subject areas ranked high in edy them.
the centrality importance measures previously analyzed. Three of Fig. 12 illustrates the KPI related to two important subject areas
the 39 subject areas were chosen to illustrate the use of Tool 3 for the in Asset Management Planning and Operations and Maintenance
sake of brevity, but organizations should employ descriptive analyt- Decision-Making. The Asset Management Planning subject area
ics to pair each subject area to at least one infrastructure AM KPI. focuses on achieving the AM objectives and identifying related
It should be noted that insights into multiple subject areas can be risks arising from previous asset failures, whereas the Operations
taken from the same figure, as illustrated below. and Maintenance Decision-Making subject area focuses on ensur-
Fig. 11 illustrates the Asset Management Strategy and Objec- ing a predictable and acceptable level of service throughout the
tives subject area focused on developing a long-term plan for man- asset’s life (IAM 2015). Fig. 12 illustrates both subject areas and
aging an organization’s infrastructure assets (IAM 2015). Fig. 11 contains two sections: (1) bar graph showing current year perfor-
shows how an organization can see its AM KPIs (e.g., median mance compared to previous year performance, which can be
outage duration in hours and number of outages) compared with changed using the arrow selector on the right side of the dashboard;
other organizations. In this context, the term organization, shown and (2) sparkline showing the specified subcomponent (selected
in Fig. 11, indicates anonymized transmission utilities that contrib- from the dropdown menu, Subcomponent Group) KPI over the years
uted outage data for all shown years. Fig. 11 depicts the three main specified in the side menu (titled Year). This descriptive analytics
sections from which a stakeholder can obtain information after application allows an organization to monitor the year-over-year
selecting the asset type in the view from the menu (i.e., titled Com- changes in average outage duration to analyze previous asset failures
ponent). The three sections are described as follows. (1) The top bar and develop mitigation plans to ensure that the issues do not arise
graph indicates the KPIs from the period as selected on the side again. In addition, the service level and trending performance of the

(a)
(b)
(c)
(d)
Fig. 10. Top 10 AM subject areas ranked by centrality measures: (a) betweenness centrality; (b) closeness centrality; (c) degree centrality; and
(d) eigenvector centrality.
specified asset subcomponent can be monitored to ensure that an transformer banks shows a decrease in average outage duration—
acceptable level of service is being provided by that subcomponent this would indicate that the operations and maintenance decisions
and to view the impact of operations and maintenance decisions. For being made are positively affecting the performance. This de-
example, the trending performance, indicated by the sparkline, for scriptive analytics application would then be distributed to all AM

Table 2. Top 10 AM network links by betweenness centrality systemic risks within the AM system network. The following sub-
Betweenness sections coincide with the main AM divisions of Strategy & Plan-
Source node Target node centrality ning, Asset Management Decision-Making, Lifecycle Delivery, Asset
Information, Organization & People, and Risk & Review. The in-
Asset management Resourcing strategy 283
planning
sights will be presented from the viewpoint of an asset manager
Resourcing strategy Resource management 282 within an infrastructure asset-intensive organization and are transfer-
Strategic planning Asset management planning 248 able across infrastructure AM industries. The application of Tools 1
Asset management Stakeholder engagement 241 and 2 would be organization-specific, whereas Tools 3 to 5 would
strategy and objectives have industry-specific KPIs and additional features, as outlined in
Maintenance delivery Reliability engineering 138 the respective tool descriptions.
Operations and maintenance Maintenance delivery 133
decision-making
Resource management Competence management 133 Strategy & Planning
Stakeholder engagement Operations and maintenance 114 Based on the centrality measures analyzed in the demonstration of
decision-making the AM network, it was shown that some Strategy & Planning sub-
Asset management strategy Strategic planning 112 ject areas were very important to the successful implementation and
and objectives
operation of an AM system. This provides evidence that the most
Stakeholder engagement Asset management planning 110
critical aspect of a successful AM system is the development of a
clear and precise strategy and plan to improve information sym-
metry among stakeholders—e.g., all stakeholders will be guided
stakeholders to see the positive impact of their coordinated effort in by the same clearly defined strategy and plan and not overwhelmed
improving the operations and maintenance decisions. by too much information. This also ensures that all other dependent
lifecycle stages and decisions that follow will be guided by a clear
strategy.
Managerial Insights
The described network analytics toolbox and the subsequent dem- Asset Management Decision-Making
onstration yield managerial insights that can reduce the information Key subject areas, based on the centrality measures, that are important
asymmetry between AM subject areas by targeting corresponding to effective decision making within an AM system are Resourcing
Primary Cause Group ID: Adverse Environment

Primary Cause Name: Fire
Subcomponent Group ID: Integral Set
Subcomponent: Structure
Outage Duration (hr): 797.7
Fig. 11. Descriptive analytics for Asset Management Strategy and Objectives subject area, including median outage duration and number of outages
KPIs for transmission line assets.

Strategy, Capital Investment Decision-Making, Operations and Asset Information
Maintenance Decision-Making, and Lifecycle Value Realization,
Digital data related to the KPIs for infrastructure assets are critical
as shown in Fig. 10. These critical AM subject areas highlight
to implement the network analytics toolbox. Asset information is
the necessity for all AM stakeholders to be making decisions based
most valuable in a digital format so that it can be used to track KPIs
on the same information. Descriptive analytics applications showed
and implement the tools to improve the value provided by assets
how stakeholders can stay informed of current information and see
throughout their lifecycles. The Asset Information subject areas
the effects of their decisions on the AM KPIs. Fig. 12 illustrates this
are highly connected, as shown in Fig. 9 and Table S1 (Supplemen-
concept by showing the KPI variation related to the specified AM
subject areas, allowing stakeholders to view the impact of their AM tal Materials), indicating that if one were to be disrupted, the others
decisions on the KPI of the assets. would also be disrupted. An asset manager should note that the
design and development of a robust asset information collection,
reporting, and storage system are critical to the development of
Lifecycle Delivery informative results on asset performance, which in turn is necessary
The lifecycle of an asset includes acquisition, operation, mainte- to evaluate the effectiveness of the AM plan being implemented.
nance, and disposal. This cycle is continuously operating within an Throughout the lifecycle of an asset, it is critical to consistently
infrastructure asset-intensive organization, as infrastructure assets collect and store the KPIs associated with an asset’s performance
are at different stages of their lifecycles. As is shown in Fig. 10, so that the KPIs can be monitored on the timeline tailored to
the Lifecycle Delivery subject areas are important for the closeness a specific infrastructure asset (e.g., monthly, yearly, or every
centrality measure, indicating that these subject areas are important 5 years). This will enable Tools 3–5 to be deployed effectively
to the indirect information flow within the AM system. This means to improve the AM system. In addition, having a consistent asset
that information is not passed through a direct connection to a node, information reporting method enables stakeholders to access to the
but through one or multiple other nodes. The nodes that pass in- same information, therefore reducing information asymmetry.
formation to other nodes typically process such information so that
it can be readily used by the following node. For example, Fig. 10
Organization & People
shows that the subject area, Fault and Incident Response, processes
the information from the Contingency Planning and Resilience The Organization & People subject areas were also clustered in the
Analysis subject area before passing it onto the Risk Assessment AM network, suggesting that most subject areas within this divi-
and Management subject area. sion are connected, and any disturbance in one area will affect all
Fig. 12. Descriptive analytics for Asset Management Planning and Operations and Maintenance Decision-Making subject areas and average outage
duration KPI for each transmission asset component.

the others. The asset manager should use this insight to ensure that using descriptive analytics applications. Key managerial insights for
all stakeholders are clear about the AM strategy, goals, and imple- systemic risk identification associated with the AM system and reduc-
mentation plan, to reduce the potential for information asymmetry tion in information asymmetry between AM stakeholders were high-
between stakeholder silos. Stakeholder buy-in to implementing an lighted as they related to each of the major AM divisions.
AM system is critical; strong organizational management is vital so Understandably, for the tools to be effective, implementation of
that stakeholders can see the positive effects of information sym- the tools requires active participation among all AM stakeholders.
metry on infrastructure asset KPIs following the implementation of As in all data-driven models, high-quality input data are necessary
an AM system. to achieve a useful output. Specific, and expected, limitations of
Tool 1 relate to its dependency on an organization’s record-keeping
Risk & Review of its AM system and/or an organization’s level of understanding of
how the AM subject areas are linked together. For Tool 2, if any
The risk and review process is critical in evaluating the effectiveness component (i.e., node or link) changes due to organizational restruc-
of the AM plan within an organization. The use of descriptive ana- turing, then all centrality values would need to be revised. Tool 3
lytics allows asset managers to efficiently evaluate the KPIs for the requires a infrastructure industry-specific expert to pair AM subject
organization’s assets. The targeted descriptive analytics applications areas with relevant KPIs, and there must be relevant data to generate
to specified critical AM subject areas allow asset managers to con- KPIs within the existing database. Tool 4 is influenced by the fea-
centrate on detailed information quickly to minimize the time needed tures available for inputs to the machine learning model (e.g., if all
to search for the result they are looking for. Descriptive analytics also available features are categorical then there is a limited number of
allow for rapid consultation among stakeholders, therefore improv- machine learning models that can be used, and the output will also
ing the review process and necessary collaboration. Deploying Tool be categorical). Finally, the ability of Tool 5 to provide an exact (or
3 also allows for automatic updates to occur in the figures so that a near exact) solution might be affected by the complexity of the ob-
stakeholder does not need to continuously update figures for use in jective function and constraints, indicating that users might resort to
reports. This enables all stakeholders to have access to the same in- heuristics, for example, to reach a solution. In addition, it is expected
formation, allowing them to make decisions with the most accurate that organizations would implement the described tools in sequen-
and up-to-date information. tial order as indicated, and become comfortable with using each tool
before implementing the next tool. By adopting the toolbox pre-
sented in this study, it is expected that stakeholders can reduce the
Conclusion systemic risks within an AM system using AM subject area-specific
Global infrastructure assets are continuously deteriorating, and the Tool 3 outputs that display information from a centralized database,
current condition of infrastructure is poor in both Canada and thus ensuring that an AM subject area’s information is not siloed
the United States. To maximize the value of each dollar spent on from the overall system. This also ensures that AM stakeholders
infrastructure for repair, rehabilitation, replacement, and mainte- make decisions using a consistent information source, reducing the
nance, effective and efficient asset management (AM) practices are likelihood of stakeholders acting in silos and causing information
needed. One of the main challenges in implementing and operating asymmetry.
an effective AM system within an organization is dealing with the
systemic risks caused by information asymmetry among dependent
AM system subject areas. This paper described a network analytics Data Availability Statement
toolbox to identify the systemic risks in AM systems and reduce
information asymmetry by using key performance indicator (KPI) Some or all data, models, or code used during the study were pro-
analytics paired with the critical AM subject areas. Five tools were vided by a third party. (i.e., transmission line outage data). Direct
presented, as follows: (1) Dependence Identification and Network request for these materials may be made to the provider as indicated
Modeling; (2) Network Centrality Analysis; (3) Descriptive Analyt- in the Acknowledgments. The data used to build the AM network
ics of Critical Subject Area-Paired KPI; (4) KPI-Based Predictive were retrieved from IAM (2015) and Global Forum on Mainte-
Analytics; and (5) Prescriptive Analytics for Optimal Network Con- nance and Asset Management (2014).
figuration. Tool 1 describes how to build an AM network from an
organization’s AM system. The connections between the AM sys-
tem subject areas are used to develop an adjacency matrix, which is
Acknowledgments
used then to build an AM system network. Tool 2 employs node- and The data for this study were provided by the Canadian Electricity
network-based centrality measures to determine the most critical no- Association (CEA), and CEA support in the development of this
des to the operation of the AM system network. Tool 3 takes the study is greatly appreciated. The financial support for the study
critical AM subject areas, identified from the node-based centrality was provided through the Canadian Nuclear Energy Infrastructure
measures, and uses descriptive analytics to track KPIs that directly Resilience under Systemic Risk (CaNRisk)–Collaborative Re-
relate to the important AM subject areas. Tool 4 uses historical KPI search and Training Experience (CREATE) program of the Natural
values and additional influencing features within a machine learning Science and Engineering Research Council (NSERC) of Canada.
model to predict future KPIs. Tool 5 uses the existing AM network The support of the INTERFACE Institute and the INViSiONLab
structure and applies optimization to generate the optimal AM sys- is also acknowledged in the development of this study. In addition,
tem network configuration to minimize the systemic risks. the authors thank the anonymous reviewers for their detailed com-
The toolbox was subsequently deployed to demonstrate three of ments in helping to clarify the systemic risk definition.
the five tools using the Institute for Asset Management’s conceptual
model and transmission infrastructure asset outage KPIs. Critical AM
subject areas were identified through the node- and link-based cen- Supplemental Materials
trality measures, and descriptive analytics were deployed so that trans-
mission utilities would be able to track their KPIs as they directly relate Table S1 is available online in the ASCE Library (www.ascelibrary
to the important subject areas. The AM subject areas were described .org).

References Delen, D., G. Moscato, and I. L. Toma. 2018. “The impact of real-time
business intelligence and advanced analytics on the behaviour of
Abdelfatah, M., M. El-Shimy, and H. M. Ismail. 2013. “Outage data analy- business decision makers.” In Proc., 2018 Int. Conf. on Information
sis of utility power transformers based on outage reports during 2002– Management and Processing, 49–53. New York: IEEE.
2009.” Int. J. Electr. Power Energy Syst. 47 (1): 41–51. https://doi.org Delen, D., and S. Ram. 2018. “Research challenges and opportunities in
/10.1016/j.ijepes.2012.10.060. business analytics.” J. Bus. Anal. 1 (1): 2–12. https://doi.org/10.1080
Abu-Samra, S., M. Ahmed, and L. Amador. 2020. “Asset manage- /2573234X.2018.1507324.
ment framework for integrated municipal infrastructure.” J. Infrastruct. Derrible, S. 2012. “Network centrality of metro systems.” PLoS One 7 (7):
Syst. 26 (4): 04020039. https://doi.org/10.1061/(ASCE)IS.1943-555X e40575. https://doi.org/10.1371/journal.pone.0040575.
.0000580. Eisenberg, D. A., J. Park, and T. P. Seager. 2020. “Linking cascading failure
Aggarwal, C. C. 2015. Data mining: The textbook. New York: Springer. models and organizational networks to manage large-scale blackouts in
Alzoor, F. S., M. Ezzeldin, M. Mohamed, and W. El-Dakhakhni. 2021. South Korea.” J. Manage. Eng. 36 (5): 04020067. https://doi.org/10
“Prioritizing bridge rehabilitation plans through systemic risk-guided .1061/(ASCE)ME.1943-5479.0000820.
classifications.” J. Bridge Eng. 26 (7): 04021038. https://doi.org/10 Estrada, E., and P. Knight. 2015. A first course in network theory. Oxford,
.1061/(ASCE)BE.1943-5592.0001733. UK: Oxford University Press.
ASCE. 2021. 2021 report card for America’s infrastructure. Reston, VA: Ezzeldin, M., and W. E. El-Dakhakhni. 2019. “Robustness of Ontario
ASCE. power network under systemic risks.” In Sustainable and resilient infra-
Barabási, A.-L. 2016. Network science. Cambridge, UK: Cambridge structure, 1–20. New York: Taylor.
University Press. Freeman, L. C. 1977. “A set of measures of centrality based on between-
Barão, A., J. B. de Vasconcelos, Á. Rocha, and R. Pereira. 2017. “A knowl- ness.” Sociometry 40 (1): 35–41. https://doi.org/10.2307/3033543.
edge management approach to capture organizational learning net- Global Forum on Maintenance and Asset Management. 2014. “The asset
works.” Int. J. Inf. Manage. 37 (6): 735–740. https://doi.org/10.1016/j management landscape.” In Proc., Global Forum on Maintenance and
.ijinfomgt.2017.07.013. Asset Management. Oakleigh, Australia: Asset Management Council.
Barker, K., J. H. Lambert, C. W. Zobel, A. H. Tapia, J. E. Ramirez- Goforth, E., M. Ezzeldin, W. El-Dakhakhni, L. Wiebe, and M. Mohamed.
Marquez, L. Albert, C. D. Nicholson, and C. Caragea. 2017. “Defining 2020. “Network-of-networks framework for multimodal hazmat trans-
resilience analytics for interdependent cyber-physical-social networks.” portation risk mitigation: Application to used nuclear fuel in Canada.”
Sustainable Resilient Infrastruct. 2 (2): 59–67. https://doi.org/10.1080 J. Hazard. Toxic Radioact. Waste 24 (3): 04020016. https://doi.org/10
/23789689.2017.1294859. .1061/(ASCE)HZ.2153-5515.0000493.
Bergh, D. D., D. J. Ketchen, I. Orlandi, P. P. M. A. R. Heugens, and B. K. Goldberg, D. E. 1989. “Genetic algorithms in search, optimization and ma-
Boyd. 2019. “Information asymmetry in management research: Past ac- chine learning.” In Optimization, and machine learning. Boston, MA:
complishments and future opportunities.” J. Manage. 45 (1): 122–158. Addison-Wesley Longman Publishing Co.
https://doi.org/10.1177/0149206318798026. Golightly, D., G. Kefalidou, and S. Sharples. 2018. “A cross-sector analysis
Black, J., A. Hoffman, T. Hong, J. Roberts, and P. Wang. 2018. “Weather of human and organisational factors in the deployment of data-driven
data for energy analytics: From modeling outages and reliability indices predictive maintenance.” Inf. Syst. e-Bus. Manage. 16 (3): 627–648.
to simulating distributed photovoltaic fleets.” IEEE Power Energy https://doi.org/10.1007/s10257-017-0343-1.
Manage. 16 (3): 43–53. https://doi.org/10.1109/MPE.2018.2801442. Graybeal, P., M. Franklin, and D. Cooper. 2018. “Principles of accounting.”
Boccaletti, S., V. Latora, Y. Moreno, M. Chavez, and D. U. Hwang. 2006. In Managerial accounting. Houston, TX: OpenStax.
“Complex networks: Structure and dynamics.” Phys. Rep. 424 (4–5): Haggag, M., A. Yorsi, W. El-Dakhakhni, and E. Hassini. 2021. “Infrastruc-
175–308. https://doi.org/10.1016/j.physrep.2005.10.009. ture performance prediction under climate-induced disasters using data
Brunetto, Y., M. Xerri, and S. Nelson. 2014. “Building a proactive, engage- analytics.” Int. J. Disaster Risk Reduct. 56 (Nov): 102121. https://doi
ment culture in asset management organizations.” J. Manage. Eng. 30 (4): .org/10.1016/j.ijdrr.2021.102121.
04014014. https://doi.org/10.1061/(ASCE)ME.1943-5479.0000251. Hassan, N. R. 2019. “The origins of business analytics and implications for
Canadian Infrastructure Report Card. 2019. Monitoring the state of the information systems field.” J. Bus. Anal. 2 (2): 118–133. https://doi
Canada’s core public infrastructure: The Canadian infrastructure .org/10.1080/2573234X.2019.1693912.
report card 2019. Ottawa: Canadian Construction Association. Hastie, T., R. Tibshirani, and J. Friedman. 2009. “The elements of statistical
CEA (Canadian Electricity Association). 2018. Instruction manual for learning.” In Springer series in statistics. New York: Springer.
reporting component forced outages for transmission equipment. Ottawa, Heng, F.-L., K. Zhang, A. Goyal, H. Chaudhary, S. Hirsch, Y. Kim, M. A.
ON: CEA. Lavin, and A. Raman. 2016. “Integrated analytics system for electric
CEA (Canadian Electricity Association). 2020. “Overview—Canadian industry asset management.” IBM J. Res. Dev. 60 (1–2): 1–12. https://
Electricity Association.” Accessed April 16, 2020. https://electricity doi.org/10.1147/JRD.2015.2475955.
.ca/about-cea/overview/#about. Herrera, F., G. Chan, M. Legault, R. M. Kassim, and V. Sharma. 2011. The
Chen, L., T. F. P. Henning, A. Raith, and A. Y. Shamseldin. 2015. “Multi- digital workplace: Think, share, do. London: Deloitte & Touche LLP.
objective optimization for maintenance decision making in infrastruc- Huang, K., M. Li, and S. Markov. 2019. “The information asymmetry
ture asset management.” J. Manage. Eng. 31 (6): 04015015. https://doi between management and rank-and-file employees: Determinants and
.org/10.1061/(ASCE)ME.1943-5479.0000371. consequences.” In Proc., AAA 2019 Management Accounting Section
Crowther, K. G. 2008. “Decentralized risk management for strategic (MAS) Meeting. 1–53. Miami: Social Science Research Network.
preparedness of critical infrastructure through decomposition of the IAM (Institute for Asset Management). 2015. Asset management—An
inoperability input-output model.” Int. J. Crit. Infrastruct. Prot. 1 (3): anatomy Ver3. Bristol, UK: Institute for Asset Management.
53–67. https://doi.org/10.1016/j.ijcip.2008.08.009. IAM (Institute for Asset Management). 2021. “About the IAM.” Accessed
Das, K., S. Samanta, and M. Pal. 2018. “Study on centrality measures in May 25, 2021. https://theiam.org/about-us/.
social networks: A survey.” Social Network Anal. Min. 8 (1): 1–11. Infrastructure Canada. 2018. Investing in Canada: Canada’s long-term
https://doi.org/10.1007/s13278-018-0493-2. infrastructure plan. Ottawa, CA: Infrastructure Canada.
Dehghanian, P., B. Zhang, T. Dokic, and M. Kezunovic. 2019. “Predictive Koziel, S., P. Hilber, P. Westerlund, and E. Shayesteh. 2021. “Investments
risk analytics for weather-resilient operation of electric power systems.” in data quality: Evaluating impacts of faulty data on asset management
IEEE Trans. Sustainable Energy 10 (1): 3–15. https://doi.org/10.1109 in power systems.” Appl. Energy 281 (Nov): 116057. https://doi.org/10
/TSTE.2018.2825780. .1016/j.apenergy.2020.116057.
de la Pena, C., D. Gonzalez Fernandez, and J. Rodriguez Gonzalez. 2016. Lee, S. H., J. Y. Choi, S. H. Yoo, and Y. G. Oh. 2013. “Evaluating spatial
“How analytics can improve asset management in electric power net- centrality for integrated tourism management in rural areas using GIS
works.” Accessed June 14, 2018. https://www.mckinsey.com/industries and network analysis.” Tourism Manage. 34 (Feb): 14–24. https://doi
/electric-power-and-natural-gas/our-insights/. .org/10.1016/j.tourman.2012.03.005.

López-Espinosa, G., A. Moreno, A. Rubia, and L. Valderrama. 2015. “Sys- Scheibe, K. P., S. Nilakanta, C. T. Ragsdale, and B. Younie. 2019. “An
temic risk and asymmetric responses in the financial industry.” J. Bank. evidence-based management framework for business analytics.” J. Bus.
Financ. 58 (Sep): 471–485. https://doi.org/10.1016/j.jbankfin.2015.05 Anal. 2 (1): 47–62. https://doi.org/10.1080/2573234X.2019.1609341.
.004. Scrucca, L. 2013. “GA: A package for genetic algorithms in R.” J. Stat.
McBride, J., and J. Moss. 2020. “The State of U.S. infrastructure.” Software 53 (4): 1–37. https://doi.org/10.18637/jss.v053.i04.
Accessed April 6, 2021. https://www.cfr.org/backgrounder/state-us Talebiyan, H., and L. Duenas-Osorio. 2020. “Decentralized decision mak-
-infrastructure. ing for the restoration of interdependent networks.” J. Risk Uncertainty
McDowell, T., H. Horn, and D. Witkowski. 2016. Organizational network Eng. Syst. 6 (2): 04020012. https://doi.org/10.1061/AJRUA6.0001035.
analysis. Seattle: Deloitte. Tarver, E. 2020. “How financial markets exhibit asymmetric information.”
Miller, B. B. 2017. “Information asymmetries conceal fraud and systemic Accessed August 30, 2021. https://www.investopedia.com/ask/answers
risks in the U.S. banking industry.” Accessed July 29, 2021. https:// /042915/how-do-financial-market-exhibit-asymmetric-information.asp.
corpgov.law.harvard.edu/2017/08/19/information-asymmetries-conceal Teixeira, A. S., F. C. Santos, and A. P. Francisco. 2016. “Spanning edge
betweenness in practice.” Stud. Comput. Intell. 644 (3): 3–10. https://
-fraud-and-systemic-risks-in-the-u-s-banking-industry/.
doi.org/10.1007%2F978-3-319-30569-1_1.
Mohamad, A., Y. Zainuddin, N. Alam, and G. Kendall. 2017. “Does decen-
Thai, M. T., and P. M. Pardalos. 2012. Handbook of optimization in
tralized decision making increase company performance through its in-
complex networks: Communication and social networks. New York:
formation technology infrastructure investment?” Int. J. Account. Inf.
Springer.
Syst. 27 (Oct): 1–15. https://doi.org/10.1016/j.accinf.2017.09.001. Tsai, C. W., C. F. Lai, H. C. Chao, and A. V. Vasilakos. 2015. “Big data
Mukherjee, S., R. Nateghi, and M. Hastak. 2018. “A multi-hazard approach analytics: A survey.” J. Big Data 2 (1): 1–32. https://doi.org/10.1186
to assess severe weather-induced major power outage risks in the U.S.” /s40537-015-0030-3.
In Reliability engineering and system safety, 283–305. New York: Uddin, W., W. R. Hudson, and R. Haas. 2013. Public infrastructure asset
Elsevier. management. New York: McGraw-Hill Education.
Nearchou, A. C. 2004. “The effect of various operators on the genetic Ujwary-Gil, A. 2019. “Organizational network analysis: A study of a uni-
search for large scheduling problems.” Int. J. Prod. Econ. 88 (3): versity library from a network efficiency perspective.” Library Inf. Sci.
191–203. https://doi.org/10.1016/S0925-5273(03)00184-1. Res. 41 (1): 48–57. https://doi.org/10.1016/j.lisr.2019.02.007.
O’Neill, M., and A. Brabazon. 2019. “Business analytics capability, organ- Valentin, V., N. Naderpajouh, and D. M. Abraham. 2018. “Integrating
isational value and competitive advantage.” J. Bus. Anal. 2 (2): 160–173. the input of stakeholders in infrastructure risk assessment.” J. Manage.
https://doi.org/10.1080/2573234X.2019.1649991. Eng. 34 (6): 04018042. https://doi.org/10.1061/(ASCE)ME.1943-5479
Opdyke, A., F. Lepropre, A. Javernick-Will, and M. Koschmann. 2017. .0000638.
“Inter-organizational resource coordination in post-disaster infrastruc- Vantrappen, H., and F. Wirtz. 2017. “When to decentralize decision mak-
ture recovery.” Construct. Manage. Econ. 35 (8–9): 514–530. https:// ing, and when not to.” Accessed July 29, 2021. https://hbr.org/2017/12
doi.org/10.1080/01446193.2016.1247973. /when-to-decentralize-decision-making-and-when-not-to.
Pell, R., R. Svoboda, R. Eagar, P. Ondko, and F. Kirschnick. 2015. Effective Weldon, W. 2008. “Johnson & Johnson CEO William Weldon: Leadership
infrastructure asset management—A holistic approach to transformation. in a decentralized company.” In Proc., Wharton Leadership Conf.
Hong Kong: Arthur D. Little. Philadelphia: Wharton Univ. of Pennsylvania.
Piryonesi, S. M., and T. E. El-Diraby. 2020. “Data analytics in asset man- Wexler, S., J. Shaffer, and A. Cotgreave. 2017. The big book of dashboards.
agement: Cost-effective prediction of the pavement condition index.” Hoboken, NJ: Wiley.
J. Infrastruct. Syst. 26 (1): 04019036. https://doi.org/10.1061/(ASCE) Xerri, M. J., S. Nelson, and Y. Brunetto. 2015. “Importance of workplace
relationships and attitudes toward organizational change in engineering
IS.1943-555X.0000512.
asset-management organizations.” J. Manage. Eng. 31 (5): 04014074.
Prajogo, D., J. Toy, A. Bhattacharya, A. Oke, and T. C. E. Cheng. 2018.
https://doi.org/10.1061/(ASCE)ME.1943-5479.0000306.
“The relationships between information management, process manage-
Yang, S., W. Zhou, S. Zhu, L. Wang, L. Ye, X. Xia, and H. Li. 2019. “Fail-
ment and operational performance: Internal and external contexts.” Int.
ure probability estimation of overhead transmission lines considering
J. Prod. Econ. 199 (May): 95–103. https://doi.org/10.1016/j.ijpe.2018
the spatial and temporal variation in severe weather.” J. Mod. Power
.02.019. Syst. Clean Energy 7 (1): 131–138. https://doi.org/10.1007/s40565
Qiu, Q., J. A. Fleeman, D. R. Ball, G. Rackliffe, J. Hou, and L. Cheim. -017-0370-4.
2013. “Managing critical transmission infrastructure with advanced Yosri, A., Y. Elleathy, S. Hassini, and W. El-Dakhakhni. 2021. “Genetic
analytics and smart sensors.” In Proc., IEEE Power and Energy Society algorithm-markovian model for predictive bridge asset management.”
General Meeting, 1–6. New York: IEEE. J. Bridge Eng. 26 (8): 04021052. https://doi.org/10.1061/(ASCE)BE
Ross, R. 2019. Reliability analysis for asset management of electric power .1943-5592.0001752.
grids. New York: Wiley. Zhou, K., C. Fu, and S. Yang. 2016. “Big data driven smart energy man-
Salama, M., M. Ezzeldin, W. El-Dakhakhni, and M. Tait. 2020. “Temporal agement: From big data to big insights.” In Renewable and sustainable
networks: A review and opportunities for infrastructure simulation..” energy reviews, 215–225. New York: Elsevier.
In Sustainable and resilient infrastructure, 1–16. New York: Taylor & Zumel, N., and J. Mount. 2020. Practical data science with R. Shelter
Francis. Island: Manning Publications.

Goforth Et Al 2022 Network Analytics For Infrastructure Asset Management Systemic Risk Assessment

Uploaded by

Copyright:

Available Formats

You might also like

Goforth Et Al 2022 Network Analytics For Infrastructure Asset Management Systemic Risk Assessment

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Goforth Et Al 2022 Network Analytics For Infrastructure Asset Management Systemic Risk Assessment

Uploaded by

Copyright:

Available Formats

Network Analytics for Infrastructure Asset Management

Systemic Risk Assessment

Introduction most infrastructure used daily by Canadians is more than 20 years

© ASCE 04022006-1 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

Asset Management Decision-Making reviews the challenges and Background

© ASCE 04022006-2 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-3 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

where ρjk = number of shortest paths connecting node j to node k;

© ASCE 04022006-4 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-5 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

Tool 4: KPI-Based Predictive Analytics

Tool 5: Prescriptive Analytics for Optimal Network

Fig. 3. Overview of the asset management network analytics toolbox.

Asset Management Conceptual Model

AM1 AM2 AM3 AM4

AM7 AM6 AM5 AM8

Fig. 4. Tool 1: Dependence identification and network modeling of AM subject areas.

© ASCE 04022006-6 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

Ranked Subject Area Importance

© ASCE 04022006-7 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-8 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

Forecasted Future KPI

Machine Learning Model

© ASCE 04022006-9 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

Fig. 8. Tool 5: Prescriptive analytics for optimal AM system network configuration.

© ASCE 04022006-10 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

Fig. 9. AM system network.

© ASCE 04022006-11 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-12 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

Primary Cause Group ID: Adverse Environment

© ASCE 04022006-13 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-14 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-15 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-16 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

© ASCE 04022006-17 J. Infrastruct. Syst.

J. Infrastruct. Syst., 2022, 28(2): 04022006

You might also like