Using PI System Analytics to

See Rotating Equipment

Life Expectancy:
Best Practices for Consuming Wireless IIoT
Smart Vibration Data to Reduce
Downtime

Dated: Jan 2018

Submitted by: Steve Edwards, OSIsoft Field Service Engineer [email protected]
Contributors: Ted Birky, David Gray, Joe Van Roosendaal, Carroll Sparks, Chris Felts, Heath
Howland, Alexander Fiset
OSIsoft, LLC
 Abstract

Using a PI System to analyze overall vibration readings is a breakthrough. Smart EDGE/IoT wireless
vibration sensors are opening new doors. The low costs of connecting these data sources to a PI System
allow you to stream and store life expectancy on machines that have never been monitored. This will
create a tremendous boost in reliability and safety performance for many companies.

In this post we present best practices for consuming overall vibration data with a PI System. We
demonstrate how to forecast machine health, alert users and capture feedback with PI Analysis Service.
We provide sample templates for PI Asset Framework. We demonstrate effective dashboards and event
visualization with web displays in PI Vision™ . We present several best practices that leverage local
operations staff as key data consumers. We show you how to configure a standard PI System to stream
forecasted machine failure into future data. We demonstrate “plug-n-play” flexibility techniques needed
when sensors are often moved between machines in the field. We give you guidelines for selecting smart
vibration sensors. We present how to build vertical asset health rollups. We present best practices for
notification of machine health issues using eMail and SMS messages.

A PI System provides visibility into rotating equipment operation. Boosted tag counts streaming from
production operations transform real-time operational data into fast, actionable insights. Barriers
between Operational Technologies (OT) and IT are removed with PI Asset Framework. PI web-based
visualization tools accelerate discovery through mobility.

Target audience
Plant Engineering designers and managers
Plant Operations managers
Plant Maintenance managers, Engineers, Mechanics
Plant Process engineers and Process technologists
Plant Reliability Engineers

OSIsoft, LLC
Jan 2018
Executive Summary
Unplanned rotating machine failure drives up costs and risks injury in nearly all plants. If you could only
see the failures coming for more rotating machines in your plant…

More online sensors. Fewer inspections.

Fewer spare parts. Improved safety. These are
powerful moves toward world class. The only
questions are… how to do it well and avoid
overspending.

A PI System and EDGE/IIoT smart wireless

vibration sensors can make it happen.

Use a PI System to log overall vibration data,

analyze, forecast and inform your decision-
makers. PI does the work behind the scenes.
What you experience is simple beyond
1 Typical bank of centrifugal pumps
complex.

Witnessing the early stages of rotating equipment failure is essential to finding latent root causes.
Response times are measured in minutes. Online PI analytics near the asset give local operations team’s
immediate focus.

Traditional vibration monitoring uses contracted experts,

hard-wired sensors, dedicated processors, and dense data
storage. This “deep dive” approach is well suited for ultra-
critical assets with high replacement cost. But what about
the rest of the plant?

In power generation there is a class of assets called Balance

of plant (BoP). These are the large quantities of pumps and
fans that support the site. Often these are unmonitored for
health due to cost of traditional hard-wired sensors. These
now are prime targets for improvement.
2 Industrial process air blower
Don’t overspend on complex layers of contracted
service-provider relationships that ultimately are
too distant to sustain. With the PI System, keep a
larger portion of machine health decisions within
the reach of your operations staff.

We will show you how in this post.

3 Rooftop Air Handler Units

OSIsoft, LLC
Jan 2018
Goals
For rotating equipment, vibration data from accelerometers
is the fundamental visibility into machine health. For BoP
assets you often don’t need very much detail to be
effective. Replacement costs are low making the decision-
making thresholds low. Early detection of machine failure
can boost repairs and reduce replacements.

The goals of this post are to help you answer questions like
these for all of your machines…

1) I know the machine is sick but is it getting worse?

2) How much longer can this machine run and how
confident are we?
3) If we know life expectancy do we need a spare?
4) Did we get a good repair?
5) Has this machine or any others like it ever had these
symptoms before?
6) How can we teach a new employee the history of each
machine?
7) Who saw the machine the last time it failed and what
did they say about it?
4 Wireless Smart vibration sensor transmitting vibration
8) Did we capture the root cause from the last time it readings and surface temperature of a motor.
failed?

Method
Using a PI System to store and analyze overall
vibration readings is the breakthrough method
presented here.

For BoP assets, the operations teams simply

seek “clean” machines. You need less about
why the machine is not healthy because the
corrective action is often to swap for a
refurbished unit. You push the heavier
diagnostics work to the offline repair teams.
Back in the plant the mission is simpler. You
confirm a machine is “clean” on entry, monitor
online, detect “unclean”, track progression and 5 Typical chemical pump pair (A/B) being inspected.
plan for maintenance accordingly.

There are three pieces that must be done well to make this method work.

OSIsoft, LLC
Jan 2018
The human element is still today the most important consideration.
In plants it is the people that make the operating decisions. I read
recently an article in Scientific American by Mark Fischette where
IBM concluded our computing capability equates to the brain power
of a house cat. I always knew cats were smart. IBM estimates that it
will take 880,000 processors to demonstrate the capability of a
human brain. We are best at creative thinking and our computers
are best at repetitive operations. It is important to understand this
and design for it. We will do that here.

The second piece is IIoT hardware. It’s what we techies all want to
talk about. The Industrial Internet of Things (IIoT) is truly moving the
needle. Wireless sensors with on-board processors are scrubbing and
delivering data at ultra-low cost. It is important to choose the right
sensors.

Finally, a PI System must be in place to store data, shape it, slice it

and notify the right people. It must be intelligent so as to not add
labor. Streaming analytics are the repetitive tasks done for us by the
PI System. The PI System can store, aggregate, scrub, forecast and
alert you. Web-based visualization, SMS text messages and email
messages pull users in the loop on preventive maintenance.

Selecting Sensors for Rotating Equipment

Let’s step back for a moment and cover some vibration monitoring basics. The key measurement
technology is an accelerometer. It can capture the smallest of forces before any of our five human
senses. What you can look for today are sensors and EDGE/gateway routing devices that have
processing capability to pre-process the raw data. With BoP assets we need only consume overall
results (typically one data point per hour) for storage and analysis.

Below are some suggestions for selecting overall vibration sensors:

1. First, use wireless sensors. They avoid costs of running conduit/wire while allowing for
flexibility to move sensors readily. Surprisingly, wiring costs in plant settings are very high,
particularly in hazardous facilities.
2. Look for sensors that use mesh-networking to navigate obstructions in your plant.
3. Use hardened industrial wireless protocols and equipment standards such as WirelessHART for
increased cyber security and reliability in a plant setting.
4. Use sensor suppliers with trusted brand names for longevity in reliability and support. Don’t cut
corners on sensor prices without first testing reliability and ruggedness.
5. Don’t forget to buy sensors rated for the hazardous area classifications and ambient conditions
they will see in your plant.
6. ** Most folks in these industries (Chemical, Oil&Gas, Power Gen, Facilities, Food, Pharma, and
others) have numerous pumps, fans, blowers, compressors, and gearboxes. That means lots of

OSIsoft, LLC
Jan 2018
 shaft bearings. Bearing failure is far and away the most common failure mode and should get
the majority of your focus. Make sure your sensor can provide a bearing fault overall signal.
What I mean is a signal designed specifically to detect early stages of bearing defects. This is a
streaming analytic derived from acceleration data. The key here is that the sensor does the high
frequency computations and leaves only a simple overall stream being sent to the PI System.
Sample rates of one event per hour to the PI System is often sufficient.
7. Be sure to have an overall sinusoidal velocity signal stream as well. This is another derived
overall signal from the accelerometer. Velocity is nice counterpart to bearing fault overall to
form a signature for the machine’s health. Each can be leading or lagging.
8. Surface temperature is a helpful complement to accelerometer data. Temperature can be a
lagging indicator of poor health. Look for this on board your accelerometer sensor all in one
convenient unit.

One sensor that meets the requirements above is the CMWA-8800 from SKF. SKF a well-known bearing
manufacturer and trusted global brand. This is the only sensor I have direct experience with in this
context. There are many other sensor suppliers in the market that may provide an equivalent device.

Units of Measure
You’ll need some new units of measure (UOM’s) added to your PI System for vibration monitoring.

For the bearing fault indication you will use the units given by your chosen sensor supplier. Each
supplier defines their metric differently. For the CMWA8800 sensor from SKF it is defined as enveloped
acceleration (gE). This can be added to a PI System in a new UOM class of called Acceleration
Enveloping. You will also need a rate of change for this unit to support forecasting analytics. This will
be a new UOM class called Acceleration Enveloping Rate.

For overall Acceleration data you will need a new UOM class that includes several UOMs to express the
overall result of a sinusoidal wave. We called this class Sinusoidal WaveForm Acceleration. In this class
we add the following UOMs: Acceleration RMS, Acceleration Peak, Acceleration Peak to Peak,
Acceleration Average.

For overall Velocity data you will need a new UOM class that includes several UOMs to express the
result of a sinusoidal wave. We called this class Sinusoidal WaveForm Velocity. In this class we add the
following UOMs: Velocity RMS, Velocity Peak, Velocity Peak to Peak, Velocity Average. Conversion
factors can be found here.

OSIsoft, LLC
Jan 2018
Run Counters
Counters keep track of how often a machine has been started and how long it has been running. This
information is necessary for making decisions about when to repair a machine. Run counters should be
a basic piece of your equipment templates such that they appear consistently in the same place across
all of your machine visuals.

Below find some useful run counters and the expression code for them.

Forecasting
It is important to express your results in the simplest form. Successful teams forecast well in any
organization. For a machine we use LifeExpectancy (Months or Days) and the ForecastedFailureDate
(MM-DD-YYYY). To supplement these we store a 30day forecast, not unlike a weather forecast. Simple
to understand for occasional users.

We said earlier that bearing failure is our focus. Let’s expand on that by looking at one typical failure
sequence. Small defects in the balls or rollers of a bearing assembly can occur, like a scratch or a chip.
This can be caused by lack of lubricant, poor quality metals, poor assembly or numerous other
possibilities. This is the beginning of poor health. Initially, there is nothing to hear, smell, touch or see.
Not sure about taste, no data for that one. However, there is data to see in the bearing fault overall
(gE) sensor stream from an accelerometer. Over time the defect damages the other surfaces it touches.
The bearing fault sensor increases as more run hours allow for more physical damage. You may begin to
see increases in overall velocity (in/sec peak). Later, this damage may begin to make a sound and
create heat. The surface temperature may increase. Finally, the bearing gets super-hot and it locks up.

To forecast a time series sensor stream is leverage its history. In doing this you are going beyond the
capability of a traditional alarm system. To forecast utilize the linear regression tool in PI AF to fit lines
to bearing fault signals over time that are trending up. We frame these regressions in time starting from
the last maintenance date to the present. These fits are all done in continuous streaming fashion by the
PI System.

For this post, we focused on this particular failure mode/sequence merely to demonstrate the
technique. You will find many more scenarios to design for and I hope you can use the same techniques.

We demonstrate forecasts in three formats in this post.

1. Bearing Life Expectancy – Time remaining until bearing fault overall reaches its HiHi threshold.
2. Bearing Forecasted Failure Date – Timestamp when bearing fault will reach its HiHi threshold.

OSIsoft, LLC
Jan 2018
 3. Bearing Fault Overall in 30days – Projected bearing fault overall reading in 30days from now.

Techniques in PI Analytics Service

PV While Running – For analyzing the vibration history we must only fit to data taken while running.
The easy way (albeit a tad expensive) is to create a second PV tag in the data archive that only contains
values taken while machine was running. Use a natually scheduled AF analytic to make this happen.
Use a trait to house this second PI tag.

Emergency Inhibit – Ideally we feed machine health diagnoses from the PI System back to your
SCADA/Control system as an output point to inhibit operation of the machine. This is especially useful
to guarantee that machine health policies are enforced to prevent damage. This inhibit signal can be an
alarm point in the SCADA/Control system or a interlock logic input.

Linear Regression – Shown below is the stack of commands for executing a look back at vibration history
since last maintenance. We are focused on bearing fault overall. We want to characterize from the
data whether it is trending up and how fast. We require a minimum amount of good data points and
require that the machine be running before enabling the linear regression fit. In the analysis we only
look back at events while the machine is running. The linear fit yields a slope (m), intercept (b) and
goodness of fit ratio (R-squared). As you can see the code is quite short and simple. Once we see a
positive slope (m) and an goodness of fit ratio > 90% we release the forecast to update archived tags for
the user to see. You’ll notice we are archiving just about every item since this is a demonstration. In a
real build you could archive fewer items

Tracking Unhealthy Runtimes with PI Event Frames

Event frames (EF’s) in the PI System are great tool for machine health work. A PI System captures the
start and end times for periods of unhealthy machine operation. You might say a PI System is like an
assistant that organizes information you may need later while you are playing golf or reading a book.

PI Vision can overlay multiple time series events on top of each other regardless of when they occurred.
For example, you can quickly compare a failure that happened last year against a current failure in
progress. Another useful example is to overlay failures across sister machines. Time-series overlay tools

OSIsoft, LLC
Jan 2018
are an absolute must for finding answers quickly in large data sets. They are a huge part of allowing
your PI System to work for you. Visit our PI System YouTube channel for Event Frames for more details.

In the example display below we compare nine previous failures of the same type machine. Very quickly
we can conclude that all nine failures were similar in shape and likely share the same root cause.

EF: Symptoms

We present in this post a very simple EF implementation you can start with and expand. The PI analysis
service triggers the event frames automatically from real-time sensor values. When Bearing Fault
overall exceeds its high limit for a proving time period the trigger fires. The PI System classifies the
symptoms (See the Table below) which is used in visual displays and emails. This list of symptoms can
grow as your teams learn more about their machines. The list is centralized in your PI library and easily
propagated to all deployments. This list can become key intellectual property for your company.

OSIsoft, LLC
Jan 2018
EF: Reason Codes

Reason codes allow users to capture what they found to be the root cause for each failure. Drop down
lists are needed for selecting the reason code. These lists serve to teach users what to look for. Like
symptoms, reason lists are always expanding as you learn. This list can grow and become more
intellectual property. The list is centralized in your PI library and easily propagated to all deployments of
health sensors.

EF: Acknowledgement and Comments

OSIsoft, LLC
Jan 2018
Manual data entry in rotating equipment work is essential. We all contribute our best when we know
that our efforts are seen and used. Diverse teams have good tools for traction to communicate well.
They organize manual data alongside sensor data and engineering data. Comments about what people
see, hear, feel, and smell at the machine are very important. Photos and videos are extremely powerful.
Other files such as shop teardown reports and independent performance reports can all be valuable
evidence when solving difficult problems. Save all manual data in your failure records. Accurate
recollection of the details leads to discovery.

As a manager, engage your users by reading their comments and responding when possible. Operations
managers, staff engineers, operators, vendors and mechanics are all vital in finding root causes of
machine failures. Manual data entry bridges gaps across work crews and across plant sites. The PI
System stores comments along
with timestamp and username to
help you make this happen.

A web-based platform allows users

to contribute wherever they may
be. (Mobile or desktop). At right,
see the standard event frame
panels built into PI Vision. These
require no programming or setup.
Features include time series
overlays in the middle, with
commenting and
acknowledgement controls to the
right.

As you can see in the image,

the comments pane includes
information about emails
sent, who acknowldeged the
event and when. Comments
are saved with timestamp
and user login too. There are
simple controls for attaching
files and viewing them.
These are essential tools for
collaborating effectively.

OSIsoft, LLC
Jan 2018
Template Design
Here are some ideas around PI analytics and template design we found helpful when building out this
post.

Number one, I strongly recommend that IIoT health sensors like the ones referred to in this work be
given their own element template. Position health sensors as optional “care providers” that can be
attached to “patients”. This gives you tremendous freedom to evolve the diagnostic algorithms and
then push those improvements with one click to all deployments of that sensor. You should expect to
make many advancements to your machine health analytics as you see more and more results. The
machine health analytics captured in these templates become important intellectual property honed
to your business. This template design will also provide flexibility to move sensors readily to and from
machines with minimal work to do in the PI System.

Let’s look at the actual buildout in more detail. Note the images below illustrate a typical BoP
centrifugal pump. The only existing piece of sensor data we have is from the motor starting equipment
telling us a discrete Running/Stopped status.
We give the motor a Drive End (DE) health
sensor. Large motors will require additional
sensors. We must also give the pump a drive
end health sensor to detect its local bearing
faults. This layout is what we will use in this
post and shown at right.

Now, what does this look like in the PI System.

As you see illustrated, the health sensors are
child elements of the pump and motor. The
pump and motor are the “patients”. Data is
exchanged between patient and care provider.

The list below contains the information provided

by a typical BoP asset to its health sensor care
provider element. Status and LastMaintenance
date are key items involved in the analytics of
the health sensor.

Each health sensor is responsible in turn for providing results back to its patient. These results come
from the forecast calculations and event frame health triggers. The list below contains the health
results that are communicated back up to the patient. To facilitate this exchange of health info back to

OSIsoft, LLC
Jan 2018
the patient we use a PI Analysis Rollup at the patient element because in we have multiple care
providers on one patient, they will be combined.

For example, let’s say we add three more health sensors to our pump. The rollup for pump
LifeExpectancy would simply be the minimum of the LifeExpectancy reported by all care provider
children. This means as you move sensors around in the field, you need only move the sensor in the PI
System. The templates know how to attach themselves for data flow to/from the patient. This is super
easy to configure in the PI System. See below the configuration.

The wireless vibration health sensor template attributes are shown below. Categories help organize into
groups and traits are used on several attributes to provide depth and organization. The best way to

OSIsoft, LLC
Jan 2018
review this template is to download it into your PI System and explore it.

OSIsoft, LLC
Jan 2018
Enumeration Sets
Enumeration sets are code lists you can build in PI System to describe behavior or status in the form of a
discrete list. Two enumeration sets were built for this demonstration. These sets were mentioned
earlier in the event frame section. These sets are included in the PI AF kit that comes with this post.

Necessary Visualization
Good visualization design is critical so that users can read these new data streams without effort. You
want your staff on their feet talking to people and spending time with the machines in your process.
You don’t want them at their desk building queries. Web-based based user interfaces are important to
allow folks to use smart phones and tablets in addition to desktops.

Dashboards - It is important to shape data and make it fall in front of the user at the right time. A
“collection” in PI Vision is effective at making smart dashboards. We can draw a dashboard panel for
one asset, specify an asset query criteria and then PI Vision will continually populate the screen with
duplicate panels for all assets that meet our criteria.

One critical dashboard we will need here is a collection of all unhealthy machines. See the image below.
In this dashboard we show the name, life expectancy [in days], run status, and health status of the
machine. We highlight sensor data using yellow and red colors if hi limits are reached. The trend
provides a quick glance but can be clicked to expand. A detailed table is provided with the remaining
useful metadata.

OSIsoft, LLC
Jan 2018
OSIsoft, LLC
Jan 2018
You might find it useful to have high level dashboards with very brief summaries for all machines with
health sensors. We have only four machines in our build here, but imagine this display dynamically
finding and displaying hundreds of machines.

Another technique to query and report by exception in PI Vision is the asset comparison table. Here we
see a more dense view of results across several machines. An example is shown below. Again the use of
yellow and red coloring is used to highlight unhealthy machines throughout the grid. In this example
three machines are running, two are unhealthy and both of those machines are showing early stage
bearing failure. Imagine this table looking at hundreds of motors in a plant sorted by health status.
Machines that need attention rise to the top. That tremendous boost in visibility we talked about
getting from wireless smart devices comes out here.

OSIsoft, LLC
Jan 2018
Detailed Display - A detailed display is always useful to give the data full context. Here we try to put
everything we know. This screen is useful for training purposes. The content we show here is typical for
a BoP pump asset in a chemical plant. Typically the only process signal for this pump is Status [Running
or Stopped]. So, we add health sensors and metadata. We use the banner for overall summary and we
use several panels below to show historical and realtime data. These displays promote “emotional
discovery” of trends and relationships that will lead to finding root causes of failure.

Visit our YouTube online course content for PI Vision for more details about visualization.

XvY Plotting - Provide an XY plot of bearing fault vs overall velocity as part of your standard visual
template for each sensor. Plotting XY is a simple way to help users recognize relationships. In the
example prodvided, you see gE plotted against Velocity (In/Sec Pk) over the time period of a poor health
event. In this case a cluster of gE readings at low velocity are meaningful. Later, increasing velocity
becomes well corellated with gE. This provides a signature that is unique and useful in understanding
the physical root cause inside the machine.

OSIsoft, LLC
Jan 2018
Recommended
Metadata
Much like run counters, there are
more key pieces of metadata that
should be kept near. The mfr and
model of the asset along with rated
size and performance from the
Engineering design. This includes
direct links to engineering drawings.
The same is true for maintenance
and reliability data in the Centralized
Maint Info Sys (CMMS). Provide links
so that information is easily
discovered. Another idea is to keep
the replacement cost for each asset
in front of the user to help in
decision making. The physical
location of the asset is useful to
avoid mistakes. The criticality is
helpful to harmonize everybody on
what is important and help make
decisions faster and cleaner.

These are only a start for one basic

class of pump. You will have
templates for many types of rotating
equipment, each with unique
metadata. Metadata fields are easily
expanded in the PI System. PI AF and
PI Vision are excellent aggregators of
metadata as shown here. You can
expect to always be improving the
metadata for rotating equipment
assets.

Be sure to learn and utilize layered

templates (called derived templates
in the PI System) in your buildout of AF. In the templates that come with this post you will see we use
an asset top-level parent template. This template contains attributes that apply all assets regardless of
class. The Manufacturer attribute for example appears in the asset top-level. Rated Head on the other
hand is specific to the centrifugal pump class.

For more detail visit our YouTube Learning channel here…

https://www.youtube.com/watch?v=dCx5_Aw5x24

OSIsoft, LLC
Jan 2018
Notifications
Well-designed emails and text messages are super important for rotating equipment work. Good
messaging reduces labor and empowers people . Bad messaging can sour the whole experience .

1. Messages should be triggered automatically based on analytics.

2. Messages should be sent quickly to help first responders capture key evidence. It is important
that everybody know about the issue ASAP.
3. Allowing many people to be near the details via messages helps solve hard problems.
4. Both real-time and historical data surrounding the issue should be provided in messages.
5. Content should be accurate and targeted for what a person on their feet will need. Don’t
expect users to do any searching.
6. If it’s not in your PI System now, it probably can be.
7. Several users must have authority to configure messages so that edits are done swiftly.
8. Messages should include static text such as work instructions, troubleshooting tips.
9. Messages should include a path for the user to easily respond to the event with feedback
(comments) and see the latest real time status on smartphone, tablet, or desktop.
10. Centralize message formats as part of the health sensor template.
11. Don’t be shy about including information in the message.
12. Thoroughly test your messages before scaling up to more machines. Start with one machine
and a small group of users. Trigger and test as many scenarios as possible.
13. Work to get candid feedback on what is not missing or unnecessary.
14. When you see that the design is well accepted then scale up to remaining machines.

In the end, message content represents key intellectual property that you can hone to your business and
share across a large enterprise. Expect message content to continually evolve as you learn more.

Visit our YouTube channel on Notifications for more details on how to extend event frames into email
messages and web services.

OSIsoft, LLC
Jan 2018
Sample Verbose Email - Shown below is a sample of a verbose email message that can be generated
using PI Notifications for machine health. The idea here is to include all data surrounding the machine
that may be useful. This can be instructions, realtime data, historical data, metadata, or just about
anything. With this mail it allows the user to stay on their feet and be fully informed. The subject line
includes the name of the asset and the key triggering data value. The body of the email contains sensor
values, forecast, maintenance data, engr links and a description of the triggering mechanism behind the
email.

OSIsoft, LLC
Jan 2018
Sample SMS Text Message – At right find a
sample SMS text message generated by PI
Notifications. The wording is critical since
we have a limited number of characters.
We need to be sure it includes the asset
name, the symptom, and the data value(s)
triggering the event.

Final Thoughts
I hope this post is helpful in your work. The
depth we cover in this post is only
scratching the surface of what is possible for
machine learning in this setting. What I
found in doing this post is that machine
learning algorithms can be broken down
into small, very manageable equations.
Then when these analyses are gathered up
and presented to a user they create insight.
You might even say the result is greater than
the sum of the parts. I was pleased to see
how well these algorithms went into place
in a PI System. I hope your experience will
be the same.

Please feel free to download the templates

used here and push forward in your own
unique setting. We look forward to hearing
what you find.

Should you need support please don’t

hesitate to contact OSIsoft.

OSIsoft, LLC
Jan 2018
Appendix
Data Sources
The raw sensor data used in this white paper was generated by the PI Ramp Soak Interface. This
interface generates patterns of data movement using prescribed time periods, amplitude and signal
noise. The amplitudes of sensor data are based on actual machine failure data. The time periods and
reoccurrence however in these data were accelerated greatly to save time in proving out the analytics.
Therefore the elapsed time results and forecasted timestamps are much more frequent than real
machine data.

References
Impacting the bottom line: 10 real-world examples of oil and gas innovators using data for economic
effect, OSIsoft, LLC. 2017

https://osisoft1980.sharepoint.com/sites/Sales/Marketing%20Assets/wp-Impacting-the-Bottomline-for-
Oil-and-
Gas.pdf?mkt_tok=eyJpIjoiWW1ZME5URmlZVGhrWXpneiIsInQiOiJSVWdjK1pcL2llcUZhYWxwTkoxMyt2V2
gyZXlhZWREdUFOR2JKanJxcCtwWHZJSEVJK2JMYllGQnJJOEI4b2Z1QWFadEFUc2dESEJaaitSWW4yTkhoNjJ
wUU9TSStHVW5tbkpSTURtK0JKK0taSWRwV2g0WW91bExMWkFydllHYXcifQ%3D%3D

IoT Collaboration Bridging Edge to Cloud: Inside Petasense & OSIsoft’s Effort to Improve Manufacturing,
By: Bill Lydon, Editor Automation.com

https://www.automation.com/iot-collaboration-bridging-edge-to-cloud-inside-petasense-osisofts-
effort-to-improve-
manufacturing?mkt_tok=eyJpIjoiWW1ZME5URmlZVGhrWXpneiIsInQiOiJSVWdjK1pcL2llcUZhYWxwTkox
Myt2V2gyZXlhZWREdUFOR2JKanJxcCtwWHZJSEVJK2JMYllGQnJJOEI4b2Z1QWFadEFUc2dESEJaaitSWW4y
TkhoNjJwUU9TSStHVW5tbkpSTURtK0JKK0taSWRwV2g0WW91bExMWkFydllHYXcifQ%3D%3D

Causes of unbalance in rotating machinery, Association of Asset Management Professionals

http://www.maintenance.org/topic/causes-of-unbalance-in-rotating-machinery

SKF Wireless Machine Condition Sensor - CMWA 8800, www.skf.com

http://www.skf.com/us/products/condition-monitoring/surveillance-systems/on-line-systems/wireless-
systems/wireless-machine-condition-sensor/index.html

Rethink Overall Vibration Monitoring, By Trent Phillips, CMRP, CRL, Novelis, 2016

OSIsoft, LLC
Jan 2018
https://www.efficientplantmag.com/2016/08/rethink-overall-vibration-monitoring/

OSIsoft, LLC
Jan 2018