Available online at www.sciencedirect.

com

ScienceDirect
IFAC PapersOnLine 55-19 (2022) 121–126
Conceptual
Conceptual framework
framework for
for smart
smart
Conceptual
maintenance
Conceptual framework
based on
framework for smart
distributed
for
maintenance for smart
based on distributed
Conceptual framework smart
maintenance based on
intelligence
maintenance based on distributed
distributed
intelligence
maintenance based on distributed
intelligence
∗
intelligence
intelligence ∗ ∗
Terrin
Terrin Pulikottil
Pulikottil ∗ Luis Luis A. A. Estrada-Jimenez
Estrada-Jimenez ∗ Jose Jose Barata
Barata ∗
Terrin Pulikottil ∗∗∗ Luis A. Estrada-Jimenez ∗∗∗ Jose Barata ∗∗∗
Terrin
Terrin Pulikottil
Pulikottil ∗ ∗ Luis A. Estrada-Jimenez ∗ Jose Barata ∗
Luis A. Estrada-Jimenez
∗ Centre
∗ Centre of
of Technology
Technology and Systems, Jose Barata
and Systems,
UNINOVA
UNINOVA Instituto Instituto∗ Centre
∗ Desenvolvimento
of Technology
Desenvolvimento de
deand Novas
and Tecnologias
Tecnologias Caparica,
Systems,
Novas Caparica,
∗ Centre
Portugal of
(e-mail:Technology Systems,
tpulikottil@uninova.pt)
UNINOVA Instituto
PortugalCentre of Technology
Desenvolvimento de and Systems,
Novas
(e-mail: tpulikottil@uninova.pt) Tecnologias Caparica,
UNINOVA
UNINOVA Instituto Instituto Desenvolvimento
Desenvolvimento de
de Novas Tecnologias Caparica,
Novas Tecnologias Caparica,
Portugal
Portugal (e-mail: tpulikottil@uninova.pt)
Portugal (e-mail:(e-mail: tpulikottil@uninova.pt)
tpulikottil@uninova.pt)
Abstract:
Abstract: An An efficient
efficient maintenance
maintenance strategy strategy willwill provide
provide more more transparency
transparency and and competi-
competi-
tiveness
Abstract:
tiveness for
for An manufacturers
An efficient
manufacturers in smart
maintenance manufacturing
strategy
in smart manufacturing will era. Even
provide
era. more
Evenmore though
though there
transparency have
there haveand been several
beencompeti-
several
Abstract:
maintenance
tiveness
Abstract: for efficient
frameworks
manufacturers
An efficient maintenance
considering
in smart
maintenance the strategy
emerging
manufacturing
strategy will
will provide
technologies,
era. Even
provide more they
thoughtransparency
still need
there
transparency haveto and
address
been
and competi-
chal-
several
competi-
maintenance
tiveness for frameworks considering
manufacturers in smart the emerging technologies,
manufacturing era. Even they still
though need
there haveto address
been chal-
several
lenges
tiveness like
maintenance
lenges for
like big
big data
manufacturers
frameworks
data storage,
storage, system
in smart
considering
system resilience,
manufacturing
the multi-modality
emerging
resilience, era. Even
technologies,
multi-modality sensing,
though
they
sensing, holistic
there
still need
holistic approach,
haveto been
address
approach, future-
several
chal-
future-
maintenance
proof,
maintenancegeneric frameworks
framework
frameworks considering
and so
considering on. the
To
the emerging
address
emerging technologies,
these challenges,
technologies, they
theythisstill need
paper
still need to address
presents
to address a chal-
smart
chal-
lenges
proof,
lenges like
like big
generic
big data
framework
data storage,
storage, andsystem
system resilience,
so on.resilience,
To address multi-modality
these challenges,
multi-modality sensing,
sensing, holistic
thisholistic approach,
paper approach,
presents afuture-future-
smart
maintenance
lenges
proof, like
maintenancegeneric framework
bigframework
data
frameworkstorage, for
and
for shop-floor.
systemso on.
shop-floor. To First,
resilience,
address
First, a
a brief
these
brief insight
multi-modalitychallenges,
insight into
sensing,
into key
this
key properties,
holistic
paper
properties, performance
approach,
presents a future-
performancesmart
proof,
indicators,genericlevelsframework
and and so on.
on. To address these challenges, this paper presents aa smart
maintenance
proof,
indicators,
maintenance
genericlevels and enablers
framework
framework
framework
for
enablers
for
of
of smart
andshop-floor.
so smart
shop-floor.
Tomaintenance
First,
address
maintenance
First, a
theseare
a brief
brief are described.
insight
challenges,into key
described.
insight into
The
this
key
conceptual
Theproperties,
paper
properties,
presents
conceptual framework
performance
framework
performance
smart
is based
maintenance
indicators,
is on distributed
framework
levels
based on levels and
distributed intelligence
for
enablers shop-floor.
intelligence of and
smart
and has has two
First, a
maintenancemajor
brief
two majorare blocks
insight
are -
described.
blocks innate
into key
- innateThe and
The adaptive
properties,
conceptual
andconceptual intelligence.
performance
framework
adaptive intelligence.
indicators,
Tool wear
indicators, monitoring
levels and
and enablers
is presented
enablers of
of smart
as
smart maintenance
aa has
use case.
maintenance are described.
described. The conceptual framework
framework
is
Tool
is based
wear
based on
on distributed
monitoring
distributed isintelligence
presented
intelligence and
as
and use
has two
case.major
two major blocks
blocks -- innate
innate and
and adaptive
adaptive intelligence.
intelligence.
Tool
is wear
based
Copyright on ©monitoring
distributed
2022 The is presented
intelligence
Authors. This isas
and
an a use
has
open case.
two
access major
article blocks
under -
the innate
CC and
BY-NC-ND adaptive
license intelligence.
Tool
Tool wear
wear monitoring
monitoring is
is presented
presented as a
a use
use case.
asdistributed case.
(https://creativecommons.org
Keywords:
Keywords: Predictive
Predictive maintenance, /licenses/by-nc-nd/4.0/)
maintenance, distributed intelligence,
intelligence, human human immune immune system,system, smartsmart
manufacturing,
Keywords:
manufacturing, tool
Predictive wear
tool wear monitoring
maintenance, distributed intelligence, human
monitoringdistributed intelligence, human immune system, immune system, smartsmart
Keywords:
manufacturing,
Keywords: Predictive
tool wear
Predictive maintenance,
monitoringdistributed intelligence, human immune system, smart
maintenance,
manufacturing,
manufacturing, tool
tool wear
wear monitoring
monitoring
1.
1. INTRODUCTION
INTRODUCTION and
and implementation
implementation modules. modules. Luo Luo et et al.
al. (2018)
(2018) focusfocus onon aa
1. INTRODUCTION more
and specific
implementation
moreimplementation application
specific application modules.by utilizing
Luo
by utilizing et deep
al.
deep learning
(2018)
learning module
focus on
module a
1. INTRODUCTION
1. INTRODUCTION and
for modules. Luo et al. (2018) focus on a
Maintenance in in smart
smart manufacturing: As manufacturing more
As manufacturing andhealth
for health prediction
specific
implementation application
prediction and
and by
modules.considering
utilizing
Luo etdeep
considering the impulse
learning
al. (2018)
the impulse response
module
focus on a
response
Maintenance manufacturing: more
of
for
more the specific
machine.
health application
prediction
specific All
application three
and by utilizing
utilizing deep
research
considering
by works learning
the impulse
deep validate
learning module
their
response
module
industry
Maintenance
industry is moving
is movingin towards
smart
towards the
manufacturing:fourth industrial
the fourth industrialAs revolution
manufacturing
revolution of of the machine. All three research works validate their
Maintenance in smart manufacturing: As manufacturing for health
framework prediction
on All and considering the impulse response
and shifting
industry
Maintenance
and shifting
industry is
to
toinsmart
is moving
moving
smart
smart manufacturing
towards
towards
the fourth there
manufacturing:
manufacturing
the fourth
As has
industrial
there
industrial
also
manufacturing
has also raised
revolution
raised of
revolution
for the
framework
the on machine
machine.
health prediction
machine. machine
All
wear
three
and wear
three
mechanism
research
considering
mechanism
research
the (tool
works
works
impulse
(tool wear
validate
wear
validate
or ball
ortheir
response
ball
their
the
and needs
shifting
industry is for tosmart
moving smart maintenance
manufacturing
towards the systems.
fourth there
industrial These
has alsosystems
raised
revolution screw
framework
of
screw the wear).
wear). on
machine. machine
All wear
three mechanism
research works (tool wear
validate or ball
their
the
and needs
shifting for smart maintenance systems. These systems framework on machine wear mechanism (tool wear or ball
not
the only
needs
and only
not shifting forto
should
should
smart
predict
tosmart
smart manufacturing
future failures
maintenance
predict manufacturing
future failures there
systems. but also
there
but
has
also
These
has also
need
also
need
raised
systems be screw
to be
raised
to
framework
Data-driven
screw
wear). on machine wear mechanism (tool wear or ball
challenges (mentioned below) has encouraged
the
more
not
the needs
resilient
only
needs for
should
for smart
and maintenance
exhibit
predict
smart future
maintenance systems.
anti-fragility.
failures
systems. An
but These
efficient
also
These needsystems
main-
to
systems be screw wear).
Data-drivenwear). challenges (mentioned below) has encouraged
more
not resilient
only should and exhibit
predict anti-fragility.
future failures Analso
but efficient
need main-
to be some researchers
Data-driven
some researchers to
to utilize
challenges utilize Digital
Digital Twin.
(mentioned below)Xu
Twin. Xuhaset al.
al. (2019)
etencouraged
(2019)
tenance
more
not only
tenance system
resilient
should
system would
and
would provide
exhibit
predict provide the
anti-fragility.
future the manufacturers
failures Analso
but
manufacturers efficient
needcompet-
main-
to be some
compet- Data-driven
presents a two
researchers
Data-driven challenges
stage
to
challenges (mentioned
framework
utilize Digital
(mentioned - below)
Virtual
Twin.
below) &Xuhas encouraged
physical
has et al. stage.
(2019)
encouraged
more
itive
tenance
more resilient
advantage
system
resilient and
and exhibit
would
and ability anti-fragility.
provide
exhibit to resist
the
anti-fragility. marketAn
manufacturers
An efficient
disturbances.
efficient main-
compet-
main- presents
some a two stage
researchers to framework
utilize Digital - Virtual
Twin. &Xu physical
et al. stage.
(2019)
itive
tenanceadvantage
system and
would ability
provide to resist
the market disturbances. Virtual
presents
some model
a two
researchers is trained
stage
to utilize using
framework DigitalUnsupervised
- Virtual
Twin. &Xu Deep
physical
et al. Learn-
stage.
(2019)
itive
tenanceadvantage
system and would ability
provide the manufacturers
to resist market disturbances.
manufacturers compet-
compet- Virtual Virtual model
presents
ing and a two is trained
stage using Unsupervised
framework - Virtual & Deep Learn-
physical stage.
15%
itive of
of the
the total
itive advantage
15% advantage totalandcost
cost
and
for
for a
ability
ability ato typical
resist
totypical
organization
market
market disturbances.
resist organization
is
is spend
disturbances. ing
spend Virtual and connected
presents model
a two stage
connected
model is
to
is trainedthe
the physical
toframework
trained physical
using
entity
entity using
using Unsupervised
- Virtual
Unsupervised using Deep
Deep
& physical
Deep
Trans-
Deep Learn-
stage.
Trans-
Learn-
on
15% maintenance
of the total activities
cost for (Boston
a typical Consulting,
organization 2013).
is An
spend fer
ing
VirtualLearning.
and
fer Learning.connected
model Luo
is
Luo to et
to al.
the
trained (2020)
physical
using
et al. (2020) presents
entity
Unsupervised
presents a
usingmore
a more Deep
Deep elaborate
Trans-
Learn-
elaborate
on
15% maintenance
of activities (Boston Consulting, 2013). An fer ing and
and connected et the physical entity
entity using Deep Trans-
on of the
efficient
15%maintenance
efficient the total
maintenance cost
totalactivities
maintenance
for
for aa(Boston
strategy
coststrategy typical
typical
would
organization
wouldConsulting,
reduce 40%
organization
reduce
is
is spend
40%2013).
of
of main-
An fer
spend
main-
multi-domain
ing Learning.
multi-domain connected
Learning.
model
Luo
model
Luo
to
et
considering
al. (2020)
the physical
considering
al. (2020)
physical,
presents
physical,
presents a
simulation
a more
using Deep
simulation
more
and
elaborate
Trans-
and
elaborate
on maintenance
tenance
efficient
on cost, 50%
maintenance
maintenance activities
of downtime
strategy
activities (Boston
(Boston and
would Consulting,
3-5%
reduceof
Consulting, capital
40% 2013).
of
2013). An
invest-
main-
An experimental
multi-domain
fer Learning. model.
model
Luo et This hybrid
considering
al. (2020) approach
physical,
presents a is
more also
simulation imple-
and
elaborate
tenance cost, 50% of downtime and 3-5% of capital invest- experimental
multi-domain model.
model This hybrid approach is also imple-
efficient
ment
tenance
efficient maintenance
(McKinsey
cost, 50% Global
maintenance strategy
of downtime would
Institute,
strategy and 3-5%
would reduce
2015).
reduce 40%
40% of
of Fortune
capital main-
Busi-
ofinvest-
main- mented
experimental
multi-domain
mented in model considering
in predicting
model.
predicting the hybridphysical,
the cutting
This
considering
cutting tool
approach
physical,
tool wear. simulation
is also imple-
wear.simulation
and
and
ment
tenance (McKinsey
cost, 50% Global
of downtime Institute,and 2015).
3-5% of Fortune
capital Busi- experimental
invest- model. This hybrid approach is also imple-
ness
ment
tenance Inside
(McKinsey
cost, (2019)
50% projects
Global
of downtime maintenance
Institute,and 2015).
3-5% of global
Fortune
capital market
Busi-
invest- mented
experimental in predicting
model. This the cutting
hybrid tool approachwear. is also imple-
ness Inside (2019)Global
projects maintenance global market Other
mented researchers
in have developed frameworks considering
ment
size
ness
mentin in(McKinsey
its current
Inside
(McKinsey current stage Institute,
(2019)Global
projects reach 2015).
to maintenance
Institute, $23 billion
2015). Fortune
billion
global
Fortune Busi-
inmarket
2026 Other
Busi- Other in predicting
mentedresearcherspredicting have the cutting
cutting tool
thedeveloped tool wear.
frameworks
wear. considering
size
ness its
Inside (2019) stage
projects to reach
maintenance $23 global in 2026
market solutions
solutions based
researchers
based onhave on Internet
have developed
Internet of Things
frameworks
of Things (Civerchia
(Civerchia et
et al.,
considering al.,
(30%
size
ness in Compounded
its current
Inside Annual
stage to
(2019) Annual
projects Growth
reachRate).
maintenance $23 billion
Rate). These numbers
inmarket
globalnumbers Other
2026 solutions
2017), researchers
Cloud Manufacturing developed (Wangframeworks
et al., considering
(30%
size
clearly
(30%
size
Compounded
in its
show
Compounded
in its current
the high
current stage
potential
Annual
stage
Growth
to reach
for
Growth
to reachan $23
$23
efficient
Rate).
These
billion in
maintenance
These
billion numbers
in 2026
2026
Other
2017), Cloud
solutions
based
researchers
based
on
on
Internet
have
Manufacturing developed
Internet
of
of
Things
(Wangframeworks
Things al., 2017)et and
(Civerchia
et(Civerchia 2017) et
considering al.,
and
al.,
clearlyCompounded
(30% show the high potential for an efficient maintenance Multi-Agent
2017),
solutions Cloud
based Systems (Peres
(Peres et
Manufacturing
on Internet al.,
of(Wang 2018).
Things et(Civerchia
al., 2017)et and al.,
system.
clearly show the highAnnual
(30% Compounded
system. potential
Annual Growth
for anRate).
Growth efficient
Rate). These numbers
maintenance
These Multi-Agent
2017),
numbers Multi-Agent
2017), Cloud
Cloud
Systems
Manufacturing
Systems (Peres
Manufacturing
et
et
al.,
(Wang
al.,
(Wang
2018).
2018). et
et al.,
al., 2017)
2017) and
and
clearly
clearly show
show the the high
high potential
potential for for an
an efficient
efficient maintenance The framework presented by these researchers
system.
Related Works:
system. Works: Most Most of of the
the research
research works
maintenance Multi-Agent
works presented
presented in in TheThe framework
Multi-Agent Systems
Systems (Peres
presented
(Peresby et
et al.,
these
al., researchers are
2018).
2018). are de-
de-
Related
system. veloped
veloped for
framework
for aa specific
presented
specific technology
by these
technology or
or application
researchers
application but
are as
but de-
as
the
the last
Related
last decade
Works:
decade in
Most
in the
the field
of the
field of
of maintenance
research
maintenance works in
in shop-floor
presented
shop-floor in The
veloped
The framework
manufacturing for
framework a presented
becomes
specific
presented by
complex
technology
by these
the
these or researchers
developed
application
researchers are
framework
but
are de-
as
de-
Related
focused Works:
on Most
integration of the
ofthethe research
emerging works presented
technologies in
like manufacturing
veloped for becomes
aa specific complex
technology the developed framework
the last Works:
Related
focused
the last
decade
on
decade
in
Most
integration
in
theoffield
the of
field theofresearch
of
maintenance
emerging
maintenance
works in
in
shop-floor
presented
technologies shop-floor
in manufacturing
like will
veloped
will not
not beforable
be able to
to provide
becomes
specific
provide the
complex
technology theor
the necessary
application
ordeveloped
necessary support.
application
support.
but
but as
framework as
Big-Data,
focused
the last
Big-Data, on Internet
decadeintegration
Internetin of
the Things
of
field
of Things theof (IoT),
emerging
maintenance Machine
(IoT), Machine technologies
in Learning,
shop-floor
Learning, like manufacturing
will not be ablebecomes
manufacturing becomes
to provide complex the
the necessary
complex developed
the developed support. framework
framework
focused on integration of the emerging technologies like Current Challenges: We
Cloud
Big-Data, computing,
on Internet
focusedcomputing,
Cloud
Big-Data,
integration
Internet
Multi-Agent
of Things
Multi-Agent
of
SystemMachine
(IoT),
of the emerging
System and
and Digital
technologies
Digital Twin.
Learning, like will
Twin. will not
Current not be able
able to
Challenges:
be We have
to provide
provide the
have identified
necessary
theidentified
a list
list of
necessarya support. of challenges
support. challenges
Cloud
Big-Data, computing, of Things
InternetMulti-AgentThings (IoT), SystemMachine
(IoT), and Digital
Machine Learning,
Learning, from
Twin. Current
from recently
Challenges:
recently published
published We literature.
have
literature. These
identified
These a challenges
list of
challenges needs
challenges
needs
Data
Cloud Analytic
computing, plays a pivotal
Multi-Agent role
System in pushing
and the tradi- Current Challenges: We have identified a list of challenges
Data
CloudAnalytic
computing, plays a pivotal role
Multi-Agent System and Digital
in pushing Digitalthe Twin.
tradi- from
Twin. to
Currentbe considered
recently
to berecently Challenges:
considered while
published We developing
literature.
have
while literature. identified
developingThese a
These smart
a
a smart list maintenance
challenges
of needs
challenges
maintenance
tional
Data maintenance
Analytic plays techniques
a pivotal to
role smarter
in pushing maintenance.
the tradi- from
framework and published
its implementation in challenges needs
tional
Data
Yan
tional
Dataet
maintenance
etAnalytic
al. (2017)plays
maintenance
Analytic
techniques
aa pivotal
presents
plays a Big-Data
techniques Big-Data
pivotal
to smarter
role in pushing
framework
to smarter
role in pushing
maintenance.
the
the tradi-
with
maintenance.
to
from
focus framework
tradi- to
berecently
framework
be
considered
considered itswhile
andpublished while
developingThese
literature.
implementation
developing inaa the
smart
the
smart
shop-floor,
maintenance
challenges
shop-floor,
maintenance
needs
Yan
tional al. (2017)
maintenance presents a
techniques to framework with focus to be and
considered its implementation
while developing in a the
smart shop-floor,
maintenance
on
Yan the
tional
on spatiotemporal
et al.
the (2017) presents
maintenance
spatiotemporal properties
a Big-Data
techniques
properties of smarter
toof the data maintenance.
framework
smarter
the data and
and invisible
with
maintenance. focus framework
invisible
•• Huge
frameworkHuge and volume
volume
and its of
of sensor
sensor data
its implementation
implementation data in for maintenance
inforthe shop-floor,
themaintenance
shop-floor, has
has
Yan et al.
information
on the (2017)
Yan et al. (2017) mining
spatiotemporal presents
from
presents a Big-Data
these
properties
athese
Big-Data data.
of framework
the Asdata
frameworka with
similar
and focus
work,
invisible
with work,
focus • posed
Huge challenge
volume ofon data
sensor management
data for and
maintenancestorage. has
information mining from data. As a similar posed challenge
•• posed
Huge volume onsensor
data management and storage.has
on
Li the
et
on et spatiotemporal
al.
information
theal. (2017)mining
spatiotemporal focusfromproperties
on data
properties of
of the
mining
these mining
data. theAs data
by
data and
and 33invisible
using
ausing
similar groups
work,
invisible • volume of
Multi-modality
Huge challenge
Multi-modality sensing
ofon sensor
sensing
data for
for maintenance
considering
data management
data
considering heterogeneous
and
maintenancestorage.has
heterogeneous
Li
information (2017)miningfocusfromon data
these data. by
As aausing
similar groups
work, posed challenge on data management and storage.
of
Li models
et al.
information -
(2017) Statistics,
miningfocusfromon Machine
data
these Learning
mining
data. by
As and
similarHeuristic
3 groups
work, data from
• Multi-modality
posed challengemulti-source
data from multi-source sensing
on data is less explored.
considering
management
is considering heterogeneous
and
less explored.heterogeneousstorage.
of
Li models
et al. - Statistics,
(2017) focus on Machine
data Learning
mining by and Heuristic
using 3 groups • Multi-modality sensing
Algorithm.
of
Li models
et al. -The
(2017) framework
Statistics,
focus on also
Machine
data includes
Learning
mining by decision
and
using support
Heuristic
3 groups • data from multi-source
Multi-modality sensingis considering
less explored.heterogeneous
Algorithm. The framework also includes decision support data
of
of models
models --The
Algorithm. Statistics,
framework
Statistics, Machine Learning
also includes
Machine Learning and
and Heuristic
decision support
Heuristic data from
from multi-source
multi-source is is less
less explored.
explored.
Algorithm. The framework also includes decision support
2405-8963 Copyright © 2022 The Authors. This is an open access article under the CC BY-NC-ND license.
Algorithm. The framework also includes decision support
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2022.09.194
122 Terrin Pulikottil et al. / IFAC PapersOnLine 55-19 (2022) 121–126
• Frameworks focuses mainly on just a specific part or
operation of a machine/system. Also, the methods
only focus on constant operating conditions.
• Variation in the consistency of the machine in its
life cycle makes it hard for the development of a
proper statistical or physical model based method.
data driven methods lacks extensive fault data and
understanding of equipment physical characteristics.
• Maintenance frameworks are specific to only one
maintenance strategies (e.g. condition-based or pre-
dictive).
• Maintenance strategies focus mainly on improving the
overall equipment effectiveness and ignores other key
aspects like energy or resource reduction.
• Current maintenance frameworks don’t address new
property requirement of the maintenance system like
resilience and anti-fragility.
Present Work: In order to fulfill these challenges, we
proposes a smart maintenance conceptual framework. Ini-
tially, we provide the base for the development of the
framework by giving insights to the properties, perfor-
mance indicators, levels and enablers of a smart main-
tenance system (Section 2). The proposed framework is
based on two major blocks of intelligence - innate and
adaptive (Section 3). The developed framework is further
explained by considering tool wear monitoring as a use
case (Section 4).
2. SMART MAINTENANCE - REQUIREMENTS AND
ENABLERS
In this section, we propose the key properties, levels , per-
formance indicators and enablers of a smart maintenance
system. These act as the foundation on which we develop
our conceptual framework.
2.1 Key Properties
We propose 4 key properties which need to be taken
into consideration while designing a smart maintenance
system. Some properties (Robustness and Pro-activity)
are innate properties of the system which are present in
the system before a failure occurs. These properties help
the system by preventing/prolonging the occurrence of
the failure. Some properties (Resiliency and Anti-fragility)
are adaptive properties of the system which are activated
after the failure has occurred. As an adaptive property,
the system can choose to be more resilient or anti-fragile.
Anti-fragility is preferable than resiliency.
Table 1. Key properties for a smart mainte-
nance system
Property Property Type Definition
Robustness Innate Ability to resist failure
Ability to predict failure and act
Pro-activity Innate
ahead of time
Ability to recover from failure to
Resiliency Adaptive
the same state
Anti-fragility Adaptive
Ability to recover and be more
robust after failure
2.2 Maintenance levels
Considering the stages of implementation of the mainte-
nance strategies in the shop-floor, We propose 6 levels of
Fig. 1. Six levels of maintenance strategies
maintenance (Fig. 1) in the shop floor (Inspired from Lee
et al., 2020 and Metso et al., 2018). These levels could
help manufacturers to understand how much evolved their
maintenance policies are in present. With the understand-
ing of the current stage, the manufacturers could aim to
raise above the ladder.
• Level 1 - Corrective Maintenance : Maintenance is
carried out after a fault/breakdown occurred
• Level 2 - Interval-based Maintenance: Maintenance
is carried out during a specific time period (e.g. every
week, every month) irrespective of the condition of
the system
• Level 3 - Model-based or Condition-based Mainte-
nance : Maintenance is carried out based on a math-
ematical or simulation or sensor data-driven model of
the system
• Level 4 - Predictive Maintenance : Maintenance is
carried out based on predictive model (developed us-
ing historical data) and the current machine condition
(sensor data)
• Level 5 - Prescriptive Maintenance : Maintenance
is carried out considering the current state of the
machine and the work flow schedule
• Level 6 - Self Maintenance : Maintenance is carried
out by the system itself considering its current state
and work flow
In practice, shop-floor doesn’t exist in one specific level.
Critical equipment in a machine, expensive machines and
bottle-neck system mostly will be in a higher maintenance
level than other less important machines.
2.3 Key Performance Indicator (KPI)
The evolution of KPIs for measuring a maintenance system
is presented below, along our proposal of a new KPI.
Machine Availability : Historically, KPI for maintenance
is measured as the percentage of actual production time
to potential production time. This gives us as an idea of
the time lost due to maintenance. Some other variants for
understanding machine availability are Down time, Mean
Time Between Failures (MTBF), Mean Time To Repair
(MTTR) and so on. Planned Maintenance Percentage and
Schedule Compliance gives an idea of the efficiency of the
maintenance strategy implemented.
 Terrin Pulikottil et al. / IFAC PapersOnLine 55-19 (2022) 121–126
Fig. 2. Smart Maintenance Radar Chart
Overall Equipment Effectiveness (OEE): This is a widely
used KPI for maintenance (Kechaou et al.,2022). OEE not
just takes into consideration the machine availability but
also the impact of the maintenance on the product quality
and the production speed (1).
OEE = Availability ∗ Quality ∗ Speed
Other variance of OEE are Overall Operations Effective-
ness (OOE) and Total Effective Equipment Performance
(TEEP) which reduces the machine availability consider-
ing operator performance and production planning time.
Smart Maintenance Radar Chart: Even though OEE cover
the impact of maintenance on some aspects of the shop-
floor performance beyond machine availability, It still
misses the impact of maintenance in some areas which
affect the production cost and safety. We propose a Smart
Maintenance Radar Chart which considers aspects like
energy consumption, Resource reduction, data privacy,
human safety, and satisfaction. The radar chart has three
axes (Fig. 2) considering the possible difference of impact
on the axes due to maintenance. For example, a main-
tenance strategy considering the improvement of OEE
only will have a negative impact on the resource utiliza-
tion or human satisfaction. An ideal maintenance strategy
should consider the improvement of all three axes. The
manufacturers could provide different weights to each axis
depending on their priorities. OEE and Energy & Resource
Reduction are a quantitative data and the Third axis on
Security, Safety and Satisfaction is a qualitative data.
2.4 Maintenance Enablers
Enablers are the technologies that aid in the improvement
of maintenance activities. Recent technological enablers
are linked with the advanced of digitization of indus-
tries and the convergence of physical assets with its vir-
tual counterpart; i.e., Cyber-Physical Production Systems
(CPPS). Conceptually CPPS represent future industrial
automation and its desired functionalities fall in the so
called ”self-x” properties (Lee et al., 2015).
123
A 5C-CPPS architecture is proposed by Lee et al. (2015) to
illustrate the different phases required to implement CPPS
from the connection to the configuration itself (self-x
properties). We use this work to map various technological
enablers related to maintenance operations as shown in
Fig. 3 and listed below:
(1) Smart connection: Includes physical devices (e.g.,
Sensor networks, gateways, physical assets) and stan-
dardized technology and protocols (e.g., OPC-UA,
MQTT) to acquire relevant information from the
shop-floor.
(2) Data to information: Considers methods to store
(distributively) encrypt (cyber-security), process (in
the edge) and analyze data (data analytics). This
data becomes relevant information to be used by the
maintenance system.
(3) Cyber: The analysis of the information from the shop
floor is supported by distributed computation i.e. ar-
tificial intelligence (Multi-Agent Systems) and Cloud
computing, providing: decentralization, modularity
and scalability to the system. The use of real time
simulations (digital twin) allows forecasting of future
scenarios e.g. to artificially create data that can be
used by the maintenance system. The use of aug-
mented and virtual reality technologies has been also
popularized as a digital tool to increase engagement
and interaction between the operator and the shop
floor.
(1) (4) Cognition: In this phase enough information is pro-
vided to diagnose a maintenance problem (if some)
and to support the decision making of the operator.
(5) Configuration: In this phase the system is expected
to at least partially execute a maintenance operation,
providing healing mechanisms to the physical assets,
controlling and setting up necessary parameters and
adapting current maintenance operations.
Fig. 3. Novel enablers of maintenance activities and its
relation to the 5 cognition phases of CPPS
3. SMART MAINTENANCE FRAMEWORK
We propose a novel framework for smart maintenance
system. The framework is designed considering the key
124 Terrin Pulikottil et al. / IFAC PapersOnLine 55-19 (2022) 121–126
Fig. 4. General overview of the smart maintenance frame-
work
properties required for smart maintenance (Section 2.1)
and has the ability to reach the highest maintenance level
(Section 2.2). The framework has drawn inspiration from
the human immune system for developing a future proof
system. Human immune system has a distributed intelli-
gence with innate & adaptive immunity and a extensive
collection of memory cells for all pathogens what could
attack the human body.
3.1 General Overview
The smart maintenance framework consists of 4 blocks
(Fig. 4),
(1) Physical Asset: This block takes care of all the activ-
ities carried out by the physical asset namely system
robustness, sensing and actuation.
(2) Innate intelligence: This is the first stage of the
maintenance system and activities are performed near
the physical asset. In an efficient system, most of the
maintenance activities are carried out by just the first
two blocks. If required, Innate system asks the help
of the adaptive system.
(3) Adaptive intelligence: The adaptive intelligence pro-
vides the required support to the innate component
in term of developing the properties of resiliency and
anti-fragility.
(4) Knowledge Base: It acts as the library for the adap-
tive components. All developed models by the adap-
tive components are stored in this block.
The interaction within/between these blocks are highly
distributed (Fig. 5). Some of the emerging technologies
that support these interactions are multi-agent Systems,
swarm learning and federated learning.
3.2 Physical Asset
The Physical Asset block consist of 3 key modules,
1. System Robustness: This module helps in improving the
robustness of the system. Predominant activities are car-
ried out during the design stage of the physical asset. Some
examples of improving the system robustness includes,
• Tougher and more resistant material properties
• Smart programming to deal with possible errors
• Fail safe mechanisms to prevent failures
Fig. 5. Smart maintenance framework
This is an iterative process and not just restricted to design
stage of the system. The iteration is carried out considering
the information from the innate component.
2. Sensor Network: This module takes care of sensing
and data acquisition process. A wide variety of sensors
could be utilized for maintenance activities for measuring
vibration, energy (power & current), acoustic, tempera-
ture, pressure, load, speed, ultrasonic, and torque. With
the implementation of Internet of Things, the monitoring
devices could abstract the measured value and passed it
through a gateway (messages into data packets).
3. Response System: This performs the instructions pro-
vided by the Innate intelligence on the physical asset.
This could be as simple as providing alarm signals to
as complicated as actuators performing self-maintenance.
Some common response systems includes PLCs, Control
Systems, User Recommendation Systems, Virtual & Aug-
mented Reality (VR/AR) and so on.
3.3 Innate intelligence
Innate intelligence (local intelligence) is developed and
maintained by the Adaptive intelligence (global intelli-
gence), but it’s the first layer of protection for the mainte-
nance system. It is physically near to the physical asset. An
enabler used for this implementation is Edge Computing.
It consists of two modules,
1. Data processing: This module converts the sensory
data into structured data. The processing algorithm is
developed by the Adaptive intelligence block. The sensory
data could include semi-structured data (e.g., XML) and
unstructured data (e.g., Images, Voices, Videos). Useful
features are extracted from these data.
2. Data Analytics: This module converts the extracted
features into useful information. This information is also
converted into actionable instructions or sent to Adaptive
intelligence for further analysis. A generic algorithm fol-
lowed is presented below,
data is compared with existing model,
 Terrin Pulikottil et al. / IFAC PapersOnLine 55-19 (2022) 121–126
if data within threshold value then
data is deleted
else if abnormal data identified in past then
Sent command to the response system
abnormal data is deleted
else
Send the abnormal data to Adaptive Intelligence
end if
This module helps the system to be proactive and respond
before failure occurs.
3.4 Adaptive intelligence
Adaptive intelligence is the main engine of the mainte-
nance system. It develops models, initially considering
historical data or simulation model and later adapts the
developed model considering the new abnormal data. An
enabler used for this implementation is Cloud Computing.
1. Model Generation: A model is generated from historical
data &/or multi-physics simulation (with boundary condi-
tion mapping). The model is tested under different param-
eters for potential failures. The model is then validated
for failure prediction. Another approach for generating
virtual model include development of physical degradation
model. This developed model can be further modified by
experiments on the physical asset.
2. Pre-processing algorithm generation: This module con-
siders historical sensory data from different sensors to
create algorithms for structuring the data. The algorithm
includes data cleaning, reduction, and transformation. The
data could be characterized using its spatiotemporal prop-
erties. Later, various techniques could be used for structur-
ing the data, including semantic web, ontologies, feature
extraction, expert knowledge, clustering.
3. Invisible pattern Analysis: This module tries find out
the relationship between the extracted features and in-
visible factors. The invisible factor might include sensory
data like environmental temperature, humidity, or energy
consumption etc., production data like product quality,
production speed, etc. or worker data like staff working
efficiency. After the relationship mining is performed a
causality analysis is carried out.
4. Drift Analysis: The abnormal data sent by the innate
intelligence helps in identifying data drift or concept drift.
Once identified, the existing model is updated, and sent to
innate intelligence for adaption. A copy of the developed
model is also stored in the knowledge base.
5. Decision Making: This module decides which type of
analysis should be carried out with the historical, the-
oretical, experimental, or simulation dataset. The mod-
ule chooses the algorithm based on the dataset from the
knowledge base. The Decision Making takes into con-
sideration the Smart Maintenance Radar Chart and the
maintenance planning & scheduling.
3.5 Knowledge Base
All the existing models and algorithms are stored in the
knowledge base. The library is also constantly updated
with new models developed by the adaptive intelligence
125
and by expert knowledge. An enabler used for this im-
plementation is Cloud Storage. A shorter list of existing
models and algorithms available in the knowledge base is
listed below,
• Statistics [regression analysis, cluster analysis, dis-
criminate analysis]
• Machine Learning [supervised, unsupervised, semi-
supervised, reinforcement]
• Heuristic [genetic, artificial neural network, fuzzy
logic systems, case-based reasoning]
• Diagnosis [decision tree, support vector machine, as-
sociation rules]
• Prognosis [fuzzy logic prediction, deep neural net-
work, match matrix prediction]
• Hybrid approach [Kalman filter, particle filter, ensem-
bled learning]
• Planning & Scheduling [particle swarm optimization,
ant colony optimization, bee colony algorithm]
4. USE CASE - TOOL WEAR MONITORING
Wear monitoring is an important activity for shop-floor
consisting of many CNC machines. Researchers have done
extensive research in this aspect with particular focus on
tool and ball screw wear (see related works in section 1).
Tool wear monitoring is considered here as a use case to
better understand the different blocks of the framework.
Physical Asset : The system robustness could be improved
by careful choosing of the tool material, tool shape, tool
design, chip-breaker design etc. The IoT Sensors net-
work includes vibration, temperature, acoustics, and en-
ergy/current sensor. Considering the abnormal data from
the IoT sensors (e.g. high temperature on work piece
surface) the innate intelligence provides command to the
response system. These commands include, automatic slow
down of the spindle speed, start coolant supply or increase
the amount of coolant, power off, alarm signal to the
operator.
Innate Intelligence : Statistical feature of the signals like
mean value, max-min point, standard deviation, kurtosis
RMS, Skewness etc. are extracted. These features are
then compared with the existing model to determine if
the values are within the threshold. New abnormal data
are sent to adaptive intelligence for further analysis. This
module also predicts the tool life and sends the information
to physical asset.
Adaptive Intelligence : Initially, the tool wear model
is developed considering tool wear mechanism, Finite
Element Method and experiments (varying machining
parameters like spindle speed, feed rate etc.). This model
is then validated with historical sensor data. The sensor
data especially, vibration and acoustic are pre-processed to
extract only the important features. The pre-processing is
performed both in time and frequency domain. Once the
model is validated, it is checked for data or concept drift
when an abnormal data is received (e.g., due to change in
tool supplier). Once drift is identified, the decision-making
module decided on the perfect algorithm to develop a more
adaptable model. The developed model is then sent to
Innate Intelligence.
126 Terrin Pulikottil et al. / IFAC PapersOnLine 55-19 (2022) 121–126

Knowledge Base : It provides the required library for
data pre-processing and development of a more adaptive
model. The newly created model is also stored for future
reference.
5. CONCLUSION
The key contribution of the paper and take away points
are listed below,
• The paper provides new insights to key properties,
different levels, Key Performance Indicators and en-
ablers for a smart maintenance system
• The paper proposes a novel framework considering
the above mentioned insights based on human im-
mune system
• The framework consist of 4 blocks, with two major
blocks of intelligence - innate and adaptive
• Tool wear monitoring is considered as the use-case for
better understanding of the framework
future work will focus on implementation of the developed
framework in tool wear monitoring for wear prediction and
smart maintenance.
ACKNOWLEDGEMENTS
tool driven by digital twin. Robotics and Computer-
Integrated Manufacturing, 65, 101974.
McKinsey Global Institute, McKinsey Company,
N.Y.N.U. (2015). The internet of things : Mapping the
value beyond the hype.
Metso, L., Baglee, D., and Marttonen-Arola, S. (2018).
Maintenance as a combination of intelligent it systems
and strategies: a literature review. Management and
Production Engineering Review, 9.
Peres, R.S., Rocha, A.D., Leitao, P., and Barata, J. (2018).
Idarts–towards intelligent data analysis and real-time
supervision for industry 4.0. Computers in industry,
101, 138–146.
Wang, J., Zhang, L., Duan, L., and Gao, R.X. (2017). A
new paradigm of cloud-based predictive maintenance
for intelligent manufacturing. Journal of Intelligent
Manufacturing, 28(5), 1125–1137.
Xu, Y., Sun, Y., Liu, X., and Zheng, Y. (2019). A
digital-twin-assisted fault diagnosis using deep transfer
learning. IEEE Access, 7, 19990–19999.
Yan, J., Meng, Y., Lu, L., and Li, L. (2017). Industrial
big data in an industry 4.0 environment: Challenges,
schemes, and applications for predictive maintenance.
IEEE Access, 5, 23484–23491.

This work has received funding from the European Union’s

Horizon 2020 research and innovation programme under
the Marie Sklodowska-Curie grant No. 814078.

REFERENCES
Boston Consulting, Boston, U. (2013). The mainte-
nance advantage in manufacturing achieving excellence
in three dimensions.
Civerchia, F., Bocchino, S., Salvadori, C., Rossi, E., Mag-
giani, L., and Petracca, M. (2017). Industrial internet
of things monitoring solution for advanced predictive
maintenance applications. Journal of Industrial Infor-
mation Integration, 7, 4–12.
Fortune Business Inside, New York, N.U. (2019). Predic-
tive maintenance market size, share industrial analysis.
Kechaou, F., Addouche, S.A., and Zolghadri, M. (2022).
A comparative study of overall equipment effectiveness
measurement systems. Production Planning & Control,
1–20.
Lee, J., Bagheri, B., and Kao, H.A. (2015). A cyber-
physical systems architecture for industry 4.0-based
manufacturing systems. Manufacturing letters, 3, 18–
23.
Lee, J., Ni, J., Singh, J., Jiang, B., Azamfar, M., and
Feng, J. (2020). Intelligent maintenance systems and
predictive manufacturing. Journal of Manufacturing
Science and Engineering, 142(11).
Li, Z., Wang, Y., and Wang, K.S. (2017). Intelligent
predictive maintenance for fault diagnosis and prognosis
in machine centers: Industry 4.0 scenario. Advances in
Manufacturing, 5(4), 377–387.
Luo, B., Wang, H., Liu, H., Li, B., and Peng, F. (2018).
Early fault detection of machine tools based on deep
learning and dynamic identification. IEEE Transactions
on Industrial Electronics, 66(1), 509–518.
Luo, W., Hu, T., Ye, Y., Zhang, C., and Wei, Y. (2020). A
hybrid predictive maintenance approach for cnc machine