Technology REPORT

The Next Step for Data Analytics

— Driving Business Strategy
At this stage of the game, many of us are confidently turning the data from our machines into actionable insights.
We’re enjoying actual successes from our smart approaches to manufacturing. But let’s widen the focus a bit—
let’s take a look at what, in a broader sense, we are learning from this data learning. Are we properly using insights
to alter business strategy? Are we enabling data to dictate enterprise-wide changes? We should be, and the
features in this report provide guidance on getting there.

Sponsored by
Spreadsheets weren’t
designed for time series
data analytics. Seeq is.

Time series data analysis poses unique challenges.

Difficult and time-consuming work in spreadsheets is a thing of the past. Seeq’s multiple applications enable
you to rapidly investigate and share insights from data stored in multiple enterprise data historians, such as
OSIsoft PI, Honeywell PHD, and GE Proficy, as well as contextual data sources like SQL Server, Oracle, and MySQL.

Seeq’s support for time series data and its challenges – connecting, displaying, interpolating, cleansing,
and contextualization – relieves you of hours and days of fruitlessly searching for insights in your process
manufacturing data. Seeq helps you get more value from the data that you’ve already been collecting, and
gives organizations data transparency and the ability to execute on those insights.

Learn more at [Link]

Asset Situational Investigation & Operational

Optimization Awareness Troubleshooting Excellence

© 2019 Seeq Corporation. All Rights Reserved.

Technology REPORT

CONTENTS

The edge, the cloud & the hybrid approach 4

Insights from time-series data: Context required 6

Organizational responses to data overload 11

Edge is the new cloud...but only if it’s intelligent 13

www.S m a r t I nd us t r y.c om
- 3-
 Smart Industry: Tec h n ol ogy REPORT

The edge, the cloud & the hybrid approach

By Oded Sagee, senior director of embedded and integrated solutions for Western Digital

p With the increased integra- connected machines, critical sys- from) and context (why it’s import-
tion of AI and machine-learning tems and other sensor-laden devices ant and how it can be used) of all
technologies into industrial in the factory. The most common the data coming in from all the
environments (such as smart use case for machine learning connected devices at the edge?
factories), IT managers seek the in IoT is for predictive analyt-
best strategies for enabling all ics, according to an IDC survey, THE ANSWER IS TRICKY
the benefits—tangible as well as where 83% of the respondents Take the factory environment as an
potential—that will affect factory were developing predictive-analyt- example: a hybrid approach collects
efficiency, now and into the future. ics capabilities. data from the edge and moves it up
New capabilities like predictive Optimizing use of this data to the cloud. IT managers, equipped
maintenance are made possible requires IT managers to store, with the context they have for all
by recent developments driving move, manage and properly uti- the edge devices in the factory, can
real-time insights generated from lize the data. But, what’s the best filter, dissect, organize and com-
the data that’s being captured, pro- strategy to capture the necessary mand this huge amount of data to
cessed and analyzed from all the content (what it is and where it’s determine the best ways to integrate

A hybrid approach collects

data from the edge and
moves it up to the cloud.

www.S m a r t I nd us t r y.c om
- 4-
 Smart Industry: Tec h n ol ogy REPORT

and optimize AI technologies for Data that’s kept on the gateway strategic business perspective, fac-
the factory. In a smart factory, data is sent to the cloud during off-peak tories can enhance their operations
is not funneled directly to the cloud hours when timeliness is no longer and output through data analysis
because valuable factory-floor con- a concern. More machine-learning both at the edge and in the cloud.
text can be lost in this transition. algorithms are performed in the Data acted upon at the edge can
Additionally, latency, security and cloud over time to produce patterns prevent mishaps in the production;
data integrity are all considerations and insights that can be used for line from a machine error, data sent
for accuracy and safety. predictions and future operations. to the cloud, or the “core,” enables
Instead, a new hybrid model The variety of sensors has mul- long-term analysis of variance fac-
is emerging where data is col- tiplied in recent years, enabling tors to render savings or heighten
lected from all devices at the edge granular monitoring of the entire efficiency within the factory.
and stored on a local gateway factory. IDC predicts the number Machine-performance deviations
equipped with necessary AI and of deployed sensors to exceed 80 can be detected (if not predicted),
machine-learning capabilities to billion worldwide by 2025. Thanks enabling factory managers to fix
perform the analysis, inference and to the leapfrogging advance in issues rather than reactively replac-
other critical tasks in real-time, technologies from chipset to appli- ing—at great cost—faltering units.
with lower latency and without cations, the data generated by these This mix of real-time analysis
some of the same security risks. sensors is put to great use. From a and long-term archival monitoring
frees up time for factory workers to
focus instead on productivity, qual-
ity and innovations. This hybrid
approach reaps significant benefits
for factories, employees, their cus-
tomers and the community.
Real-time analysis helps prevent
mishaps and provides context to
factory operations to make perti-
nent decisions on the floor, while
the archival analysis helps factory
owners make strategic decisions to
enhance the layout, machinery, cost
savings and overall output efficiency
of the business over time. Industry
players stay abreast of ever-changing
technologies to construct optimal
floor environments. And as quality
improves and the quantity increases,
the working environment becomes
more positive, productive and opti-
mal for innovation. p

www.S m a r t I nd us t r y.c om
- 5-
 Smart Industry: Tec h n ol ogy REPORT

Insights from time-series data: Context required

By Michael Risse, VP/CMO at Seeq Corporation

p Recently I wrote about the data in an enterprise historian or yield better insights just because it’s
explosion of interest and innova- data lake, they may find they do all in one place.
tion in time-series data storage not accrue the same benefits as To understand this, let’s consider
offerings, including historians, they do with other data types. In scenarios where centralizing data
open source and data-lake options, fact, if they use expectations based does benefit the user—relational
and cloud-based services. The on experience with relational data data, for example. Relational data
plethora of choices ensures indus- as a justification for aggregating has keys that work as handles to
trial-manufacturing customers will their manufacturing data, they the data, i.e. tables, fields and
find a data-management option will be disappointed with the column names, so aggregating
that fits their needs. Whatever the results. With some types of data, or centralizing data yields more
priorities—data governance, data consolidation in a single system possible relationships among the
consolidation, security, analytics or provides advantages for analytics tables, fields, and databases. This
a cloud-first initiative—customers and insights as compared to isn’t new; business intelligence
will have many good choices for distributed data sets, but it’s not solutions (Cognos, PowerBuilder,
where to store data. the same with time-series data. etc.) gained significant traction
At the same time, if an Whether it’s a data historian, lake, with this approach starting in
organization is planning to platform, pond, puddle or silo— the 1990’s. Today, data storage is
consolidate their manufacturing time-series data won’t necessarily so inexpensive vendors can offer

The ability to define the time periods of interest within a signal based on any type of criteria is a
critical “analytics time” component of working with time-series data.

www.S m a r t I nd us t r y.c om
- 6-
 Smart Industry: Tec h n ol ogy REPORT

platforms providing “any to any” see more insights across larger data (relationships) or semi-structured
indexes, enabling complete self-ser- sets and the data is pre-indexed or data (indexes) does. There are no
vice for a business analyst. organized and ready to work. “handles” in a time-series signal,
Another example is platforms so there is no way to add value in
that index all data contained within KEEP YOUR DATA WHERE IT IS pre-processing the data for analyt-
a semi-structured data set—think With structured (relational) and ics. This is a key issue for engineers
web pages, machine logs, and semi-structured (log files, web working with the data as they have
various forms of “digital exhaust.” pages) strategies as success stories to (at the time they do their analy-
Two variations of this approach are for centralizing data, it’s easy to see sis) find a way to integrate “What
used by Google and “document” why one could assume consolidating am I measuring” (the sensor data)
NoSQL databases such as Mon- time-series data into one place might with “What am I doing” (what an
goDB. The idea is the structure of yield equal benefits to end users, but asset or process is doing at the time)
the data doesn’t have to be consis- it doesn’t. IT-centric data solutions and even “What part of the data is
tent or defined in advance as in a may try to convince themselves important to me?”
relational table. Instead, a schema their centralization models apply to
is overlaid on the fly, or after the time-series data, but they fail like CREATING CONTEXT
fact, which enables the user to trying to climb a greasy flagpole: it As an example of the challenges
work with any “handle” created by doesn’t work without handles. in working with time-series data,
the index. Again, this means that Why is this? Time-series data let’s consider a simple time-se-
the more data is centralized and simply doesn’t lend itself to pre-pro- ries data set that has sensor data
indexed, the better. Users get to cessing the way structured data recorded every second for a year,

Contextualization is what
transforms time-series
data from a squiggly line
in a control chart into data
objects of interest.

www.S m a r t I nd us t r y.c om
- 7-
 Smart Industry: Tec h n ol ogy REPORT

In Seeq’s advanced-analytics application, time periods of interest are “Capsules” and marked
by colored bars at the top of the trending pane. In this case they denote areas that need to be
addressed in a data-cleansing effort.

or 3.6M samples, in the form as when plant or process line is it’s obvious there are an infinite
timestamp:value. creating a certain product number of ways the signal might
Most likely, the user doesn’t want • Data from a business application, be sliced or consumed for ana-
all the signal data for their analysis; for example when energy price is lytics purposes. And, since there
instead they want to identify time greater than x are so many possible options, the
periods of interest “within” the • A multi-dimensional combina- actual choosing of the time peri-
signal. For example, perhaps the tion of any or all of these time ods of interest can only be done at
user needs handles to periods of periods of interest (like where “analytics time” when the user’s
time within the data for analysis they overlap, or where they don’t) intent is clear and the relevant
defined by: • An event, for example if a time segments may be identified.
• Time period, such as by day, by user wants to see data for the In addition, this example is just
shift, by Tuesdays, by weekdays 90-minute period prior to an alarm one signal. Imagine production
vs. weekends, etc. environments of 20,000 to 70,000
• Asset state: on, off, warm up, In other words, time periods of signals such as pharmaceutical
shutdown, etc. interest are when a defined condi- or chemical plants, oil refineries
• A calculation: time periods when tion is true, and the rest of the data with 100,000 signals, or enterprise
the 2 derivative of moving aver-
nd
can be ignored for the analysis. roll-ups of sensor data that include
age is negative Two comments on this example. millions of signals.
• Context in a manufacturing First, even with a simple example Second, while the examples
application like an MES, such of one year of data from one signal, above use the term “defined by”

www.S m a r t I nd us t r y.c om
- 8-
 Smart Industry: Tec h n ol ogy REPORT

to describe the time periods of all of its forms should be included EXCEPTIONS TO THE RULE
interest, we can also call this “con- in its definition. There are possible exceptions to
textualization.” Generally, industry Finally, it is important to “analytics time” context, because
use of contextualization is when remember that any analysis of dynamic contextualization is not
data is merged from different data time-series data involves sam- always a requirement for some
sources, like a Batch system, signal, pling of analog signals with strict work products. A standard report
and quality metric coming together adherence to the challenges of or KPI, such as a weekly status
in a data model. But in the exam- interpolation and calculus, some- summary or OEE score, has the
ples above, the context can come thing that IT-data consolidation/ time periods of interest defined in
from anywhere: a measurement, aggregation efforts typically don’t advance and then used repeatedly
another application, or simply the address. The ability to align signals in the analysis.
user’s expertise and intuition. with different sample rates from On the other hand, if it’s an ad
Contextualization, at “ana- different data sources in differ- hoc investigation such as root-cause
lytics time” and in the hands of ent time zones spanning daylight analysis or understanding variation
the engineer, is what transforms savings time or other changes is in a quality metric, then only engi-
time-series data from a squiggly an absolute requirement prior to neers at “analytics time,” when they
line in a control chart into data enabling the defining of the rele- are doing the work, will be able to
objects of interest for analysis, and vant time periods. define the data dimensions required
for analysis.
Knowing what you are looking
for in advance, or not, is the dif-
ference between fixed reporting
solutions using static definitions
of time periods (even if they
provide a drop-down selection
of time ranges, it’s still defined
in advance)—and high-value
dynamic, ad hoc, analytics and
investigation efforts.

PROCESS EXPERTISE
IS REQUIRED TO
CREATE CONTEXT
One outcome of the need for
analytics-time contextualization
is many data scientists start at a
disadvantage when working with
time-series data. This is because
they can’t run their algorithms
until the data is modeled, which

www.S m a r t I nd us t r y.c om
- 9-
 Smart Industry: Tec h n ol ogy REPORT

means focusing or contextualizing define, assemble and work with for it at “analytics time,” is going
the data they are working with the time periods of interest within to define and enable the advanced
first. But who knows the time time-series data, including access analytics and insights that are
periods of interest and relevant to related data in manufacturing, the focus of smart industry and
contextual relationships needed for business, lab and other systems. Industry 4.0 investments. There-
a particular analysis? The engineer. fore, organizations that align the
The experience, expertise and abil- CONCLUSION contextualization requirements
ity to define the relevant data for There is ever more attention and of time-series analytics with their
investigations and analyses are all pressure on digital transformation data strategy will have a higher
in the same person. and the required IT/OT inte- chance of improving production
Contextualization—and the gration necessary to provide an outcomes through insights.
engineers who understand the integrated view across business and The location of the data—in a his-
assets and processes—is therefore production data sets. Therefore, it’s torian or many historians, in a data
imperative to time-series analytics going to be increasingly important lake, on premise or in the cloud—
regardless of the data-management for IT and manufacturing organi- isn’t going to change the required
strategy. Only the engineer, who zations to recognize the importance analytics effort. There are great
alone has the expertise and under- of contextualization regardless of reasons and offerings for data stor-
stands the needs of their analysis, storage strategy in working with age and consolidation, on premise
will know what they are looking for time-series data. and in the cloud, but the priority for
right at the time of investigation. The engineer who knows what insight is accelerating the expertise
This includes the ability to rapidly they are looking for, and can ask of process engineers and experts. p

It’s going to be increasingly

important for IT and
manufacturing organizations
to recognize the importance
of contextualization.

www.S m a r t I nd us t r y.c om
- 10-
 Smart Industry: Tec h n ol ogy REPORT

Organizational responses to data overload

By Chris McNamara, Smart Industry content director

p IoT is propelling us from an era where employees are empowered Conference panel titled “Organi-
of information scarcity to an age to change the way they work and zational transformation in an age
of information abundance, but for encouraged to share their knowl- of information abundance,” where
this information to be valuable, edge across the organization. And moderator Trent Salvaggio, exec-
businesses must respond to what that doesn’t always come naturally utive director of the IoT Talent
it indicates. to people, whether they work in the Consortium, began by posing a
This doesn’t always happen. C-suite or on the plant floor. seemingly simple question to the
The reasons are varied. Respond- But there are experts out there— professionals joining him onstage:
ing to new information often guides in leading organizational Have you seen any major changes
requires a change in mindset of transformations through the perils driven by the amount of access we
the entire organization. Adapting and pitfalls. And they gathered now have to information?
to data happens best in workplaces during the last Smart Industry The answers, you’d think, would
be a unanimous, resounding YES!
But the panelists had some surpris-
ing replies.
“Well, there is nothing you
can do with artificial intelligence
without lots of OT [opera-
tions technology] information,”
explained Marie Getsug, program
manager, strategic consulting—
asset management with Jacobs
Engineering Group. “But the chal-
lenge is how we’ve thrown more
and more data at users and opera-
tors over the last decade or two.”
She noted the growing chasm
between this massive amount of
data and the human brain’s capac-
ity for processing all of it. Without
the right structure, she warned,
we panic. We drown in the infor-
mation we’ve worked so hard
to generate.

www.S m a r t I nd us t r y.c om
- 11-
 Smart Industry: Tec h n ol ogy REPORT

EVOLUTION VS. talent to do what it is good at,” provides great, diverse insights, but
TRANSFORMATION added Khosravi. “Providing a it also involves the full workforce
Panelist Magyar Khosravi, director way to accelerate experts making in the process, getting buy-in from
of industry solutions with Canvass their decisions.” the get-go. “And it often helps
Analytics, described the change as That requires change, of course. justify the investment,” he added.
more evolution than transforma- New methods. New mindsets about “The sum of these use cases justifies
tion. “It’s not a dramatic change in getting down to work. The panelist wider efforts.”
culture,” he said. “Small efforts are shared their perspectives on train- Even with the right motivation,
key. We’re evolving machines to ing teams to properly adopt and the correct strategy and the proper
do things cheaper. We’re evolving implement these practices. Wazni training, smart-manufacturing
people to make better decisions and detailed a complete, three-tiered efforts fail at an alarming rate. The
do more skilled things.” approach in which every level of panel was asked why that remains
And that process (whether you an organization is trained for their a reality.
label it evolution or transformation) level of responsibilities. “We create Khosravi believes that sights are
ideally involves wisely, securely a matrix of courses that each posi- often set too high. “They think AI
partnering IT and OT compo- tion needs to be successful in the is magic. It’s magical, yes, but it’s
nents of the enterprise (a sentiment smart-manufacturing space,” he not magic.” He added that failure
that was repeated throughout this said, detailing the top layer of lead- to establish the correct processes for
conference) and bringing in the ers, and middle and lower layers making data insights actionable is
pros—the data scientists according of implementers. also a recipe for failure. “There have
to Arconic’s Hicham Wazni, who “That goal is having management to be clear ROIs in place and people
rounded out the panel. “The key that is inspired rather than fearful. need to know that they must do
is getting them to fit in with your When you have the ability to learn, with the data,” he said. “If you don’t
culture,” the corporate director that fuels the culture.” have those two things, it fails.”
of manufacturing intelligence & “And it’s gone perfectly, right?” As these experts dove deep into
smart manufacturing added. asked Salvaggio, generating laughs the complex topic of organizational
That goals vs. culture divide from the audience. transformation, Getsug made a
informed many of the topics this statement that, taken out of con-
panel touched upon. “Those on MAKING CHANGE text, might seem contradictory to
the plant floor are intimidated That humorous question opened all of these smart efforts. “We must
by IoT and cloud concepts,” the discussion on the common be willing to accept that we don’t
opined Getsug. “And we often unwillingness to make changes know what we don’t know,” she
see a disconnect between strategy that are advised by all of this new said. “The goal is having manage-
and tactics.” data—how actionable insights can ment that is inspired rather than
Those developing the strategy overcome industrial inertia. fearful. When you have the ability
and, ultimately implementing Wazni advised soliciting from to learn, that fuels the culture.”
the tactics are humans, after all, each individual in a plant a use case And the best way to accomplish
requiring structure as much as the for smart manufacturing. Asking those things is education. From
industrial networks at play here. them “How could this improve experts. Like the foursome on this
“The goal is to enable the human your work functions?” This not only Smart Industry panel. p

www.S m a r t I nd us t r y.c om
- 12-
 Smart Industry: Tec h n ol ogy REPORT

Edge is the new cloud...but only if it’s intelligent

By Keith Higgins, VP of marketing with FogHorn

p No, edge computing is not going is some confusion in the market enormous cost savings, as well
to kill the cloud. regarding the proper definition. as improved efficiencies and data
While the cloud has brought In specific terms, “real” edge insights for industrial organizations
massive constructive disruption to a computing starts with software that looking to embark on the path
broad range of industries, the edge offers a hyper-efficient complex toward digital transformation.
is now the next big lever for digi- event processor (CEP) that cleanses, To explain the value of true edge
tal transformation with the IIoT. normalizes, filters, contextualizes computing, it’s best to consider
According to Gartner, by 2022, and aligns “dirty” or raw stream- five elements:
as a result of digital business proj- ing industrial data as it’s produced. 1. Real-time streaming analytics,
ects, 75% of enterprise-generated Without a CEP, latency is higher, close to the data source. CEP-
data will be created and processed the data remains “dirty” making based, real-time analytics are
outside the traditional, centralized analytics much less accurate, and ideal for industrial applications
data center or cloud, up from less ML models are significantly com- requiring low latency and result
than 10% in 2018. promised. This definition of the edge in greater efficiencies in things
The reasons are clear—there is is use-case driven—it’s the closest to like uptime, yield and energy
simply too much data to move back the source of data you can get. savings. It also provides much
and forth, security is a growing A “real edge” solution includes higher data fidelity than sam-
concern, and more applications integrated ML and AI capabilities, pled batch processing.
require extremely low latency to all deployable into the smallest 2. Iterative edge-to-cloud machine
be effective. Further, executing (and largest) compute footprints. learning. Edge devices, gener-
machine learning (ML) models The CEP function enables real- ating continuous inferencing
and artificial intelligence (AI) at time, actionable analytics on-site on live-streaming industrial
the edge generates higher quality at the industrial edge, with a user data (including audio and
predictive insights, delivering experience and alerts optimized for video) regularly send insights
greater operating efficiencies, fast remediation by OT personnel. back to the cloud. These edge
including uptime, yield and It also prepares the data for optimal insights enhance the model,
energy savings. ML/AI performance, generating significantly improving its
How do we define “the edge” in the highest quality predictive predictive capabilities. The
its truest form? insights to drive asset performance tuned models are then pushed
Truly processing data at the edge and process improvements. back to the edge in a constant
can help companies digitize to closed loop, reacting quickly
drive differentiation and a estab- THE BENEFITS OF ‘TRUE’ to changing conditions, and
lish a clear path toward securing EDGE COMPUTING generating much higher quality
a competitive advantage, yet there True edge computing can yield predictive insights to improve

www.S m a r t I nd us t r y.c om
- 13-
 Smart Industry: Tec h n ol ogy REPORT

asset performance and pro- intelligence, organizations on industrial-control systems

cess adjustments. can process most of their data and other highly constrained
3. Radically lowers data per- locally, eliminating the need edge-computing devices,
sistence and transport to transmit sensitive OT data such as Raspberry Pi-based
requirements. There is just too across networks. Some envi- equipment. This minimizes
much data being produced by ronments, for security reasons, investments in heavy compute
industrial sensors to send it all are not allowed to be connected or new industrial-control sys-
to the cloud. Processing live to the internet at all. This also tems hardware.
data at the source reduces data reduces security infrastructure,
network and storage-resource risk mitigation, and regulato- So, there you have it. Edge will
needs and can reduce cloud ry-compliance costs. unlock the trillions of dollars of
storage and communications 5. Leverages small-footprint value-creation for IIoT and will
costs by 100-1000x. edge computing and controller move many pilots to production
4. Enhances security pos- hardware. Edge-intelligence smoothly and with improved eco-
ture. With the help of edge software should be able to run nomics and business outcomes. p

www.S m a r t I nd us t r y.c om
- 14-