

What is predictive maintenance?

Predictive maintenance technologies let you

:
 Predictive maintenance technologies let you
hear what your enterprise assets are trying to
tell you. The machines in your factories, your
!eet of trucks, your industrial equipment –
they’ve been talking to you for years. They’ve
been telling you when they’re about to break
down and what they need to run longer and
more smoothly.
Predictive maintenance allows businesses to anticipate failures and schedule
maintenance when and where it’s immediately needed. It arms businesses with
the information they need to push for peak performance from their valuable
assets yet remain con"dent that they’re not pushing them too far and risking a
costly breakdown.

Predictive maintenance de#nition: Predictive

maintenance seeks to prevent equipment failure and
downtime by connecting IoT-enabled enterprise assets,
applying advanced analytics to the real-time data they
deliver, and using the resultant insights to inform
educated, cost-e$ective, and e%cient maintenance
protocols.

Why is predictive maintenance so

important to today’s businesses?
:
 Modern businesses are in a time of unprecedented change and competition. The
Amazon E#ect has led to a rapid rise in consumer demands for control,
personalization, and speed. A shifting trade and political climate has left many
companies struggling to maintain a#ordable supply and manufacturing
relationships. And as more and more businesses undergo digital transformation,
competition is rising and the margin for error is increasingly slim. As a result,
today’s business leaders are looking to gain a competitive edge through smart
solutions, which predict when asset maintenance is needed, help increase cost
e$ciency, and streamline their often complex enterprise asset management
requirements.

Swiss Federal Railways use case

Hear how predictive maintenance helped achieve operational excellence.

What is the di$erence between

reactive,
Join the webcast preventive, and predictive
maintenance?
The di#erence in these three maintenance models lies not so much in how the
maintenance tasks are undertaken, but when.

■ Reactive maintenance: This is essentially the act of doing nothing until

something breaks. As a maintenance strategy, this is not usually practiced by
large companies for obvious reasons. However, it can be an unintended practice
if certain parts and components are left out of the regular rotation of traditional
maintenance schedules. Reactive maintenance always happens after the
fact.
■ Preventivemaintenance: This is informed by past performance and the
knowledge and experience of engineers and operators. It includes routine,
periodic, planned, or time-based maintenance. Indeed, it often prevents
breakdowns, but unfortunately it can be inexact, which may lead to expensive
:
 maintenance before it’s needed or to unnoticed weaknesses in the maintenance
process. Preventive maintenance happens at times that are pre-set,
often long in advance.
■ Predictive maintenance: This is possible when Internet of Things (IoT)
networks integrate all enterprise assets into a live ecosystem. The ability to
transmit and analyze data in real time, means that live asset condition – rather
than calendars – become the foundation for maintenance protocols. Predictive
maintenance happens in real time, exactly when and where it’s needed
needed.

The following chart (adapted from Deloitte) displays the progression of

technological capabilities throughout industrial revolutions and the resulting
impact on maintenance strategies and equipment e#ectiveness.

Industrial Industry Industry Industry Industry

Revolution 1.0 2.0 3.0 4.0

Technological
Mechanization,
Mass Automatization,
Digital
Innovation steam production, computer solutions,
power electrical power IoT cloud
energy systems

MaintenanceReactive Preventative Preventative Predictive

Policy maintenance maintenance maintenance maintenance

Technology Visual Instrumental Sensor Sensing

inspection inspection monitoring data and
predictive
analytics

Overall <50% 50-70% 70-90% >90%

Equipment
E#ectiveness
:
 How do predictive maintenance and
IoT analytics work?
The "rst step in the predictive maintenance process involves the collection of real-
time data and information from connected IoT network assets across the
business. This data must then be stored and managed in such a way that it can
readily be processed, accessed, and analyzed. The “predictive” component comes
into play when arti"cial intelligence (AI) and machine learning technologies are
applied to the data to make it start telling a useful and actionable story.

There are four basic stages to the architecture of predictive maintenance and an
Industrial IoT (IIoT) network:

The predictive maintenance process

:
 1 . Sensing and gathering data using predictive maintenance technologies (for
example, thermal imaging or vibrations)

2 . Transmitting that data – in real time – across the network to a central business
system

3 . Applying intelligent technologies like AI and machine learning analytics to that

data, to get it to deliver the most useful and relevant insights

4 . Taking rapid action on those data-driven insights to establish the maintenance

and response protocols (both human and automated) required

Monitoring asset conditions and

leveraging predictive maintenance
technologies
Predictive maintenance is made possible through cyber-physical systems that
help to integrate machines and enterprise assets into an intelligent IoT network. It
starts by identifying the asset conditions that need to be monitored, then "tting
sensors and establishing an IoT network, and, "nally, gathering and analyzing data
from that network to deliver actionable "ndings and insights. Identifying these
conditions to be monitored is the "rst crucial step toward transforming a
business’s enterprise asset management through predictive maintenance.

Monitoring conditions
Initially, managers must establish the conditions that need to be monitored for
each machine. That analysis may be visual, auditory, thermal, or – most typically –
a combination of those criteria and more. The technological step at this point
consists of determining the correct sensors and monitoring tools to be "tted:

■ Vibrationanalysis: Small changes in vibration patterns may indicate

imbalance or misalignment, while high vibration levels may indicate impending
bearing or other problems. Vibration analysis can give early warnings of failure
:
 and is particularly useful in detecting imbalance, misalignment, mechanical
looseness, or worn or damaged parts.

■ Sound and ultrasonic analysis: Under normal operation, most systems

create steady sound patterns. Changes in the reference sound pattern can
indicate wear or other types of deterioration. Ultrasonic analyses can also give
information about the overall health of the system by translating high-frequency
sounds (such as those produced by steam or air leaks) into the audible range.

■ Infraredanalysis: As with ultrasonic analysis, thermography also uncovers the

hidden by using infrared analysis to translate temperature changes into a visible
spectrum. Even very subtle changes to normal operational temperatures can
warn of impending problems.

■ Fluidanalysis: Beyond simply monitoring the levels and temperature, the

physical and chemical analysis of %uids can give valuable information about the
condition of mechanical components. By seeing the rate of degradation in
coolants and lubricants, preventive steps can be taken as soon as these insights
warrant.

■ Others: Other predictive maintenance technologies are specialized for various

unique industrial needs. They include: laser alignment, electrical circuit
monitoring, crack detection, corrosion monitoring, electrical resistance changes,
and other industry-speci"c means of measuring corrosion or deterioration.
:
 Public Cloud Private Cloud

Environment Publicly-shared computing Private computing

resources resources

Auto-Scaling High Can be limited

Security Good – but depends on Most secure – all data

security of the vendor stored in private data

Reliability Medium – depends on High – all equipment

Internet connectivity and premise or hosted by
service provider availability dedicated private clou
provider

Cost Low – pay-for-what-you- Moderate to high – ca

need model and no need for require on-premise
on-premise storage and resources such as a d
infrastructure center, electricity, and
sta#

Who Is It for? Companies that want to Government agencies

take advantage of the latest healthcare providers,
SaaS apps and elastic IaaS and any business tha
while keeping costs low handles a lot of sensit
data

Predictive maintenance technologies

Once the above criteria are established, the appropriate sensors and monitors
must be "tted and connected to a central business system, most commonly an
enterprise resource planning (ERP) system, via a cloud-connected IoT network.
:
 Finally, the necessary AI-driven software solutions must be in place to support the
various algorithms and analytics processes necessary to deliver actionable
insights and recommendations from the data gathered.

■ IoT network: When enterprise assets are augmented with sensors, processing
ability, and other technologies, they are able to send and receive data – usually
via cloud connectivity – to and from a central business system. This comprises
an IoT network and underpins the predictive maintenance strategy.

■ IoTgateways: Many older assets still work perfectly well, yet their analog
technology predates digital integration. These machines can be "tted with IoT
gateway devices, which may include cameras, microphones, and thermometers,
to gather and transmit real-time data on their operational states.

■ Cloud
connectivity: Cloud connectivity delivers the on-demand availability of
computer system resources. In an IoT network comprised of multiple industrial
assets, it’s critical that multi-location data centers be integrated into a single
database and system.

■ Modern database and ERP: Legacy disk-based databases are not well
equipped to manage the voluminous and non-linear data that comprises Big
Data and complex data sets. Furthermore, predictive maintenance uses AI and
machine learning to perform advanced analytics on such data. This whole
process is best served by a modern AI-powered ERP with an in-memory
database that is fast, responsive, and almost in"nitely scalable.

■ AI and machine learning: Pioneering computer scientist John McCarthy

de"nes AI as “the science and engineering of making intelligent
machines.” Machine learning is a subset of AI that uses algorithms to analyze
and understand data. Predictive maintenance solutions are dependent upon AI
and machine learning to not only sort, understand, and learn from enterprise
assets’ operational data – but to extrapolate upon that knowledge with
actionable recommendations and insights.

■ Advanced analytics: AI and machine learning power advanced analytics.

Managers must determine the attributes and conditions to be assessed and the
:
 analytical outcomes that are desired. In this way, the algorithms that inform
advanced analytics can be programmed to be as insightful and actionable as
possible, and to best learn from data and new experiences over time.

■ Digital twins: A digital twin is just that: a virtual recreation of an actual physical
asset. By creating digital twins, managers can visit any possible operational
scenario upon the twin – without any risk of actual real-life damage to a costly
machine or device. This helps to augment predictive maintenance by allowing
machine learning and AI tools to incorporate and learn from experiences that
have never even happened.

Examples of predictive maintenance

use cases
:
 ■ Oil and gas sector: Oil drilling puts enormous wear on assets and can lead to
great risk and danger in the event of a failure. Through real-time monitoring of
changes in oil temperature and the speed of gearboxes in drilling equipment,
predictive maintenance has greatly improved safety and reduced maintenance
costs by up to 38%.
■ Automotive industry: On assembly lines, spot-welding guns perform about
15,000 spot welds each per day. By connecting welding guns around the world
and collecting their operational data, auto manufacturers can gather millions of
data points, leading to unprecedented predictive accuracy on the condition and
state of these assets.
■ Domestic appliance manufacturing: Vibration measurements of the drum
rotation in the dryer production process have helped predict malfunctioning or
breakdown. This predictive maintenance application has eliminated
manufacturing defects by 33% and reduced consumer maintenance costs by
27%.
■ Railroad asset management: “Voids” occur when an empty space develops
under a track leading to potential delay or even derailment. Recent innovation
has led to cab-based monitoring systems that can detect a number of variables
as they roll over the rails. This has led to improved void detection and an overall
rise in customer safety.
■ Steel industry: Anomaly detection is being used to gather real-time readings
of the vibration, rotational speed, and electrical current (amperes) in the cold-
rolling equipment used in steel processing. This application has led to a 60%
improvement in equipment lifetime and greatly reduced losses due to downtime
and delays.

Bene#ts of predictive maintenance

:
 The implementation of predictive maintenance systems has led to impressive
results across multiple industries.

Interestingly enough, it’s not that the idea of predictive maintenance is anything
new. For decades, businesses have strived to achieve greater predictability in their
asset maintenance – but it has taken the rise of technologies like AI and modern
ERP systems to deliver the capacity and functionality needed to achieve predictive
maintenance solutions that really work. The bene"ts of which include:

■ Bettervisibility across your entire operation: Increased visibility into "eld

and other o#-site assets. This allows OEMs and third-party service providers to
o#er better value and more informed services.
■ Lowermaintenance costs and improved asset performance: Predictive
maintenance consistently leads to the better use of existing resources, a
reduction in downtime, and to the life extension of valuable assets.
■ More empowered teams: When asset operators, service providers, and
supply chain managers are armed with data science and real-time analytics,
they can develop maintenance schedules that work – they become planners and
strategists rather than "re"ghters.

Next steps to transforming your supply

chain with predictive maintenance
solutions
:
 Many businesses have not changed their asset maintenance strategies in decades
– despite having modernized other areas of their business. Changing longstanding
processes is challenging and it can be di$cult to get buy-in from your teams. The
most successful business transformation plans begin with a good communication
and change management strategy – to help engage your teams and break down
silos. Speak to your software vendor to learn more about which tools and solutions
will work best for your unique needs and to get you rolling with your road map and
digital transformation journey.

Explore SAP Predictive Asset Insights

Maximize your asset performance with modern predictive maintenance tools.

SAP Insights Newsletter

Learn more
:
 Further reading

 Search...

How is your experience with

this page?



: