Manufacturing and the data conundrum

By far the most important finding is the increased understanding of how to use process data to improve product quality, but manufacturers are also realising gains in reliability, through

Commissioned by

Too much? Too little? Or just right?

A report by The Economist Intelligence Unit

Case study: Meritor: Towards data-driven production perfection 9

Case study: ABB/Sandvik: Reducing deviations, eliminating imperfections 12

V From monitoring to alerting, predicting and solving 13

About the research

This Economist Intelligence Unit study, commissioned by

Wipro, examines how manufacturers now collect, analyse

and use the complex, real-time data generated in production

processes By far the most important finding is the increased

understanding of how to use process data to improve

product quality, but manufacturers are also realising gains in

reliability, throughput and maintenance practices by tuning

into what their production processes are telling them

This report, a follow-up to our 2013 omnibus report on

data usage, The data directive: How data is driving corporate

strategy—and what still lies ahead, is based in part on a

survey of 50 C-suite and senior factory executives from

North America (50%) and Europe (50%) from companies

that produce a broad range of industrial goods These

include electronics (12%), machinery (12%), chemicals

and gases (12%), vehicle parts (10%), rubber or plastics

(10%) and more Respondents are from intermediate to very

large organisations; 32% have global revenues in excess of

US$5bn, 32% have revenues of between US$1bn and US$5bn

and 36% have revenues of US$500m-$1bn To complement

the survey, the EIU conducted in-depth interviews with senior manufacturing executives and academics, as well as related additional research

The report was written by Steven Weiner and edited by David Line Our thanks are due to all survey participants and interviewees for their time and insights

Interviewees (listed alphabetically by organisation) included:

• Peter Zornio, chief strategic officer, Emerson Process Management

• Stephan Biller, chief manufacturing scientist, GE Global Research

• Joe ElBehairy, vice president, engineering, quality and product strategy, Meritor

• Kent Potts, manager of industrialisation, Meritor

• Daniel W Apley, professor, industrial engineering and management sciences, Northwestern University

• Shiyu Zhou, professor, Department of Industrial Engineering, University of Wisconsin-Madison

Executive summary I

Chief technology officer

Chief quality officer

Other “C-suite” role

Chief safety officer

Chief, supply chain

Key findings from the survey include:

l Manufacturers have significantly ramped up

their shop floor data collection

Some 86% of survey respondents report major

increases in the amount of production and

quality-control data stored for analysis over

the past two years But it hasn’t been easy—

only 14% of those surveyed report no problems

managing the data glut from real-time

production sensors and associated reporting

and analytical models

l A minority of manufacturers has an

advanced data-management strategy

Fewer than half of respondents (42%) have

what they consider to be a well-defined

data-management strategy A further 44%

say they understand why shop floor data is

valuable and, consequently, are putting in

place resources to realise that value There is

no doubting its importance, though: every

single manufacturer surveyed reports that

data collection is a priority concern for their

business

l Manufacturers find it difficult to integrate

data from diverse sources—and to find the

skilled personnel to analyse it

Difficulty integrating data from multiple

sources and formats is the most commonly

cited problem in managing greater volumes

of data, picked by 35% of respondents—no

surprise, given the age of most manufacturing

plants and that technology is transitory while

infrastructure is durable Companies also find

that because of the speed of data-technology

advancement they often lack the internal

expertise necessary to maximize the benefits

of collected information (cited by 33%)

l While data collection from monitoring is common, data analysis to predict issues or solve problems is less so

While almost all manufacturers find it normal

to monitor production processes—for example, 90% or more say their companies have mature data analysis capabilities for such essentials as asset and facility management, safety, process design and supply chain management—less than half have in place predictive data analytics, and less than 40% use data analytics to find solutions to production problems

l Data is delivering stellar quality and production-efficiency gains…

Using insights gathered from data analysis, two-thirds of companies report annual savings of 10% or more in terms of the cost of quality (that is, net losses incurred due to defects) and production efficiencies, and about one-third say their savings on both measures have been in the range of 11% to 25% This may explain why more than three-quarters of respondents identify aggressive data programmes as an important way to boost efficiency and lower costs

production-l …but collecting data doesn’t automatically yield benefits.

Despite many manufacturers reporting impressive savings from data analysis, 62%

are not sure they have been able to keep up with the large volumes of data they collect, and just 50% are sure they can generate useful insights from it, as it comes from too many sources and in a variety of formats and speeds

Manufacturers have used data to measure production since at least 3000 BC, when the oldest discovered cuneiform tablets were marked with pictographic words and numbers All it took was a reed or stick to mark damp clay, and the number of sheep, bags of grain or output

of spears was readable, but only to the literate overseer

Similarly, today’s industrial data, displayed on computer screens, is understandable and useful only to the trained overseer But there is far more

of it, and it is available instantly, so that as issues arise process adjustments can be made quickly

In today’s ideal digitally networked production environment, complex data can be used far more easily than ever to improve product quality, boost throughput, improve shop floor reliability, enhance safety and predict maintenance requirements, eliminating unscheduled downtime

That is the ideal, at any rate In the past decade,

as more manufacturers have implemented a broader array of digital controls—in the process linking together production machinery that used

to operate independently—it has become an appealing vision of what making things might actually become everywhere

“Today’s integrated operations go above and beyond what has been the traditional realm of process control,” says Peter Zornio, chief strategic officer of Emerson Process Management, a unit of

St Louis, Illinois-based Emerson Electric Company

“We think there are three big ideas at the heart of

Ready or not, here it comes II

it The first is pervasive sensing You can get more and more data points than ever before

“Second, integrated operations means multiple disciplines can analyse and discuss data from the plant together, not just one discipline at a time And third is the realm of big data and equally big analytics.”

Stephan Biller, chief manufacturing scientist for GE Global Research—a group responsible, among other things, for finding ways to make General Electric’s 400 factories as efficient as possible—says the latest iteration of thinking there is called the “brilliant factory.” The brilliant factory idea works together with the industrial internet and software development that GE calls

“Predictivity,” mirroring what theorists believe can be a manufacturing world so all-knowing that it routinely predicts production and product problems and solves them, too

“It’s the entire digital thread from engineering and design, to manufacturing engineering, the factory and our suppliers,” says Dr Biller of the GE factory “What’s new is envisioning the feedback loop from the factory in real time, through factory engineering and from the service shops The amount of data is quite astounding.”

In fact, at GE’s new battery production plant

in Schenectady, New York, 10,000 variables

of data are collected, in some cases every 250 milliseconds “We now have an infrastructure in the plant, data highways, that match what we have in the public Internet,” says Dr Biller

“ Today’s

integrated

operations go

above and beyond

what has been the

traditional realm of

process control ”

Peter Zornio, chief strategic

officer, Emerson Process

Management

The allure of this vision is pervasive In a survey

of manufacturers conducted by The Economist

Intelligence Unit for this paper, 86% say that

during the last two years they have significantly

increased the amount of production and

quality-control data stored for analysis Nearly

two-thirds say they use sensor-generated data from

networked machines—an essential element of

the integrated factory—and 20% say they plan to

use data from networked production machinery

(Figure 1) Equally telling, two-thirds of those

surveyed say they also use sensor-generated

data from external sources, off their shop floor,

for comparison purposes—a move into the more

complex and analytically difficult world now

generally called “big data”

But not everything is settled when it comes

to collection and use of digitised data Most factories are decades old and predate in their design any consideration of this type

of technology The most recently completed complex greenfield oil refinery in the US began operations in 1977, for example Despite the decades of post-World War II quality improvement programs—such as the teachings

of statistical process control guru W Edwards Deming, the scholarship of Joseph Juran,

Japanese kaizen process improvement teams,

Six Sigma programmes, the Toyota Way and Lean

Manifold data sources

What sources of data are used by your company to lower the cost of quality and improve

manufacturing efficiency? Select all that apply

(% respondents)

Figure 1

Sensor-generated data from individual machines

Sensor-generated data from networked machines

Supplier-provided test data

Supply chain management system/supplier data

Enterprise data (ERP)

RFID

After sales failure data

Customer feedback system—Compliance/incidents

management data

Operator logs

Manufacturing execution system (MES) process historian

Sensor-generated data from external sources for

comparative purposes

42 8

62 20

18 66

Manufacturing—tens of thousands of factories

in North America and Europe are light years removed from advanced, cutting-edge digital processes

Most of these plants installed control systems along the way, many of them proprietary systems that have been locally customised and continue

to operate—producing, perhaps, batch reports

on operations at the end of the day

“Migrations from systems of this nature are not for the faint of heart,” notes a senior executive

in the control systems industry He tells the story of one large factory, with annual revenue

of US$500m, where production is controlled

by orders written on coloured pieces of paper, one colour for each day of the week If every workstation in the plant is using the same colour, the process is in sync

It is therefore no surprise, in this environment, that only 14% of surveyed companies say they have experienced no problems as they manage increasing volumes of machine-generated process and quality data Companies wrestle with efficiency and quality-improvement data from so many sources that confusion and apples-to-oranges comparisons are easily made The number-one source of data, used by 96% of surveyed companies, is old-fashioned customer feedback, followed by process historian systems (90%), existing enterprise resource planning

systems (88%), accounting and financial data (88%), pre-existing supply chain management systems (86%) and after-sales failure data (82%; Figure 1)

“There is an enormous amount of data, and it’s a challenge to figure out how to integrate it,” says Daniel W Apley, professor of industrial engineering and management sciences at Northwestern University in Evanston, Illinois

“What you would like to use it for is to identify root causes of quality problems and product variation People have been talking about this for decades But the truth is, there are still many open research challenges and no real established methodology that can be used to trace quality problems back to the root causes when there are thousands of upstream process variables that are potential root causes When there are thousands

of variables, you typically need data for hundreds

of thousands, or millions of parts in order to find meaningful statistical associations between problems and root causes.”

As GE’s Dr Biller says, “When you think about all the tasks that people have to do—the maintenance system, scheduling, material handling, incoming material, the machines themselves and their error codes, how much material is in each of the buffers, does the part pass or fail—and each plant has 10 to 15 individual systems This is what makes the task somewhat difficult.”

Respondents to the EIU survey conducted for this

report see product quality management as the

area in which greater volumes of data are most

likely to make the biggest difference Nearly

three-quarters (72%) pick this in their top three

business areas likely to see gains from more

data, a much larger proportion than for any of

the other areas and 28 percentage points more

than the proportion picking process controls, the

number-two area of potential gains (Figure 2)

Shiyu Zhou, a professor in the Department of

Industrial Engineering at the University of

Wisconsin-Madison, says that discussions about

the need for better data analytics are “typically

reactive” to customer queries or complaints—

which often link back to quality issues In fact,

he says, it has become easier than ever to hear

the voice of the customer because of

data-driven product designs that report performance

issues automatically to manufacturer service

departments Examples, he says, are medical

equipment, such as magnetic resonance imagers

or CT scanners, or jet aircraft engines that are

linked to the Internet and communicate on their

own when service is needed Emerging problems,

in turn, lead to an enhanced need to boost

analytic capacity linked directly to shop floor

production processes

The machines themselves, in other words, feed

the need for process data, leading to installation

of more linked machines, and more actionable

data in the factory In this view, products that

ask for service are like the razor blade, which by

steadily growing duller creates the need for more

Quality first III

razor blades, and a strategy for making them

At Meritor, a maker of drivetrains, axles brakes and other commercial vehicle components, customers tend to focus on one metric—the number of rejected parts per million (PPM)—to evaluate suppliers “When you take into account high-level manufacturing processes—we do casting, forgings, stampings, machining, heat treating and assembly—and every truck buyer wants to have the truck the way they want it with specific transmission, axles, and brakes—the variations are in the thousands,”

says Joe ElBehairy, Meritor’s vice president for engineering, quality and product strategy

What’s more, truck demand can swing wildly in volume, which stresses manufacturing systems, where long and stable production runs most often reduce product variation To respond, Meritor has as much as quintupled the amount of data it collects at its 28 manufacturing plants

Meritor began to track defect rates not just

by part, but also by individual production operations It also decided to differentiate between reject PPM of products shipped to customers and supplier PPM, which takes into account quality levels from component suppliers

In 2013, Meritor’s reject rate was 139 PPM

During the first quarter of 2014, with more plants working to improve the traceability of production issues, the rate fell to 67 One plant, producing

an entirely new type of air brake, achieved perfection—zero PPM (see case study on page 9)

Where data can make the difference

In which of the following areas do you see greater volumes of data yielding the biggest gains? Select the top three

(% respondents)

Figure 2

Product quality management Process controls

Supply chain management/sourcing

Targeted capital spending Safety & facility management

Process design and improvements Predictive maintenance/asset management Operations management

72 44

42 36 30 30 20

12 10

Throughput improvement

“A key element is that we realise, as a company, that quality is valuable to our customers,” says Mr ElBehairy “Some of the principles we applied are not new or earth-shattering, but we’ve been able

to apply them to the complexity that we provide

in our products.”

With the proper analysis of complex production data potentially yielding such dramatic gains in quality and efficiency, it is perhaps no surprise that the rush to collect it still outpaces planning

to use it Indeed, just 42% of companies responding to the EIU survey say they have

a well-defined data management strategy, although a slightly larger proportion (44%) say they understand the value of shop floor data and are working to capture that value This is despite the fact that all realise the paramount importance of data: every single company surveyed places a priority on data collection

GE’s Dr Biller emphasises that an important part

of any change in data strategy, and consequent

alterations to production processes, is careful and considered planning “It’s a step process,”

he says “First you gather data and network it Then you give the people in the plant the ability

to operate the system using the data You need

to go through the steps rather slowly so that people in the plant understand what we’re trying

to do, and so that we can work with them as a collaborator Most of the time, the people in the plant know far more about it than you do.” Equally important to realising value from this kind of initiative, says the senior executive from the control systems industry, is absolute commitment from senior management, especially the CEO, to building an integrated data process

“You can put in all the components to make it work, the computers and software, but if you don’t have leadership skills and trust, it can lead

to failure no matter what system you have,” he says “Having commitment from the CEO is an absolute prerequisite.”

“ First you gather

data and network

it Then you give

the people in the

plant the ability to

operate the system

using the data You

we’re trying to do,

and so that we can

work with them as a

collaborator ”

Stephan Biller, chief

manufacturing scientist for

GE Global Research

Like most manufacturers, Meritor, of

Michigan, has been on a long and determined

drive to improve processes and products at

its 28 factories in a dozen countries The

company makes drivetrain, braking and other

components for trucks, trailers, off-highway,

defence and speciality vehicles

The latest iteration of company strategy,

dubbed M2016, made operational excellence

a renewed priority “We adopted as our top

metric reject parts per million [PPM],” says

Joe ElBehairy, vice president for engineering,

quality and product strategy In 2013,

companywide this figure was 139 reject PPM

With a goal of lowering that to 75 PPM by 2016,

Meritor has turned, in part, to carefully heeded,

real-time shop floor data

“Our data collection is an order of magnitude

larger than it was several years ago,” says

Mr ElBehairy “Some of it is related to safety,

but a lot of our data gathering is related to

traceability.” If something goes wrong, Meritor

wants to know where and why it happened

“And it’s not just collecting data, but

real-time acting on that data,” says Kent Potts,

industrialisation manager and leader of a

quality improvement push at Meritor’s factory

in York, South Carolina

At York, three workstations assemble calipers

for Meritor’s EX+ air disc brake from start to

finish More than 40 steps are required for the

basic brake, but that’s only the beginning of the

product’s complexity Originally launched with

14 different specifications of weight, stopping

power, pads, packaging and the like, EX+

assembly ballooned to 169 specifications after

sales volume rose sharply following a contract

award two years ago A major customer wanted

the brakes, but insisted that rejects had to be

10 PPM or less

To comply, the York plant added sensors, monitoring gear, a programmable controller system and its own custom programming

Employees were trained on an error-proofing system that verifies that the correct parts and processes are applied for each brake Bar codes are used to keep track of parts, and Meritor devised a system called “fit to light”, in which a computer keeps track of the assembly steps for each brake and turns on lights over the correct bin for the next component Reach for the wrong component, and a red light flashes

Meritor used tools that communicate with the programmable controllers and socket trays

so that the tools could be used for multiple assembly operations and brake specifications

The programmable controllers verify that the correct socket and torque gun recipe is used for each assembly process; each piece receives the correct customised treatment

“Additionally, process data are stored in our manufacturing genealogy database for each air disc brake that’s assembled The data includes the brake serial number, who assembled it, the component parts installed and process data such as fastener torques,” says Mr Potts

The result of this application of real-time networked data to improve shop floor processes has been better than any manufacturer usually expects During the year from March 2013 to March 2014, the York factory had a zero defect rate No product rejects Error-free production also permitted improvement of the on-time delivery rate to 98%—the best of any Meritor plant

Techniques like these and tighter attention to quality lowered the company’s overall reject PPM

in the first quarter of 2014 to 67, below the 2016 goal Now, the goal is to sustain the progress

Case study: Meritor: Towards data-driven

production perfection

For every brilliant idea in building real-time, system-wide data collection and analysis, and for every example of skillfully fine-tuning

a complicated production process using the industrial Internet and sensor data, there’s a real-life story that brings you back to actual shop floors

The control systems industry executive describes

a steelmaker where, after sophisticated, automated process controls were installed, operators replaced intricate plant-wide readouts—temperatures, process adjustments

to compensate for feedstock variations and so forth—with data of greater personal interest (In the background, the integrated digital system continued to monitor and collect vast amounts

of production information)

Where theory meets reality

IV

“There are just two numbers [on display],” he says

“One is variable income being produced right now, and the other is how much bonus money will be made at the end of the month.” Precisely calibrated automated readouts, displaying thousands of variables every second, were turned off because machine operators from each shift preferred to compete manually with operators from other shifts rather than with an automated system

“One of the guys at an oil refinery told me, ‘We’re staffed to run; we’re not staffed to change,’” says Emerson’s Mr Zornio “The capital-spending priority, and the manpower priority, is to just keep the place going There is no manpower

or capital to put in place the next generation

of stuff that gets the plant to the next level of improvement.”

Internal siloes make it difficult to employ data effectively

35 33 21

21 9

We have experienced no problems 14

“ One of the guys

Peter Zornio, chief strategic

officer, Emerson Process

Management

Companies that responded to the EIU survey

raise a series of impediments to the enhanced

use of complex, machine-generated data to

improve their processes Number one on the

list, cited by 35% of manufacturers, is difficulty

integrating data from multiple sources and

formats (Figure 3)

“The bottleneck is not in sensing; we

have incredible sensing technology,” says

Northwestern’s Professor Apley “It gets back

to the thousands of variables and identifying

which is the root cause of the problem.” Older,

proprietary systems, including some enormously

popular ERP systems, produce only summary,

batch reports; newer ones may crank out data, in

different formats, four times each second

“Most of these older factories are not

networked,” says Dr Biller of GE “The data stays

within the production machinery If you want

to improve a system’s performance, you have to

get the data out of the machine, then integrate

it into an IT system—some kind of intelligent

platform.”

But simply installing that platform isn’t the

whole answer, either Thirty-three percent of

surveyed companies say an important issue is

finding highly trained people to use it A related

problem—asking the right questions of your

systems to generate the right answers, was cited

as an issue by 21% of surveyed companies Also

problematic: companies organised into feuding

siloes that don’t share essential information

(also cited by 21%)

Talent, says Northwestern’s Professor Apley, is

thin on the ground “Relative to 20 years ago, it

is more difficult now to find young people who

are highly trained in analytics and data sciences

and who want to go into manufacturing,” he

says “They are often more drawn to financial companies, or companies like Google and Facebook.” Even so, Northwestern is among the universities that have recently launched

an engineering-oriented masters of science in analytics program The student body is roughly one-third international and two-thirds domestic students, and so far, all have received multiple job offers “There are just so many companies looking for people who have the skills to analyse large amounts of data,” Professor Apley says

Mr Zornio has found siloing can be a significant issue because when “every facility makes their own decision” about which efficiency controls

to put in place, interplant uniformity becomes impossible Nonetheless, centrally controlled manufacturers may make decisions about best practices that individual facilities resist because

of inevitable local variability

“In this big-data world, you may know that you don’t have the people who can look at all the data and figure out what needs to be done,” he says

But suppose you do have the people The next tripwire, says the senior executive from the control systems industry, is “information overload There’s not enough intelligence in software to sort out this overload

“For example, let’s say there’s a machine that sends out an alarm to the operator It needs grease or whatever But what if three or four machines do this? Then suddenly you have alarm overload, and then you have to have alarm management It is easy to find 500 things to do

in a plant But it’s damn tough to find the 497 things we are not going to do That’s the real challenge.”

“ It is more difficult now to find young people who are highly trained in analytics and data sciences and who want to go into manufacturing They are often more drawn to financial companies, or companies like Google and Facebook ”

Daniel Apley, professor of industrial engineering and management sciences, Northwestern University

ABB, based in Zurich, Switzerland, makes power, automation and electrical products and provides a range of industrial control services

One of its recent successes came in helping Sandvik Materials Technology of Sandviken, Sweden (about 190 km north of Stockholm), which makes specialty stainless steel, titanium and alloys for equally specialised uses

Sandvik faced the problem that as the uses

of steel grew more intricate, with greater precision required from each delivery, production equipment had to keep pace

In 2013, as part of a long-term process improvement effort, Sandvik’s attention turned

to an important component of the production system—a bidirectional rolling mill, or Steckel mill, used to make metal strips thinner and thinner with each pass through the rollers

Sandvik had no plan to replace the equipment

Instead, it opted to improve how it used the mill with a few new sensors feeding digital controls

Based on a tightly defined model of perfection, these would continually adjust rolling speeds, pressures and the number of passes through the mill to compensate for variation

First, Sandvik installed additional sensors

to measure precisely the width of the rolled metal and its temperature, which changes during rolling as the metal interacts with the machinery In some factories, dozens or even hundreds of sensors might be required, but Sandvik made do with just nine To control the process, the company needed more data, but not a flood of it

Every rolling job begins with a detailed model

of what the exactly right outcome should be This means the system—provided by ABB—must take multiple factors into account, including the material being rolled; its thickness, width and grade; the target thickness; the number

of passes through the mill that should be required; and the adaptations to rolling pressure, temperatures, rolling speed, torque and flatness that must be made Increasingly, customers want thinner steel, but thinner steel strips can easily be brittle and prone to deformation and in-process separation

“We run all kinds of special steels, the entire range from stainless to high-alloy,” says Patrick Högström, hot rolling mills production manager at Sandvik “The variety of steel grades and sheet dimensions make production very complex and knowledge intensive The model makes it possible to optimise rolling in

a completely different way from what a human being is capable of It becomes smoother, and with noticeably less scrap, which means increased yield.”

Thanks to the new sensors and related process controls, Sandvik can now roll specialty metal

to thinner tolerances while maintaining the metallic properties, such as strength and formability, required for the final use Compared with its old process, the new controls have reduced the degree of deviation from perfection by 35%, and the average volume of imperfections has dropped by 80%

Case study: ABB/Sandvik: Reducing deviations, eliminating imperfections

Even with the many complications between

shop floor data theory and practice, companies

surveyed by the EIU have found a number of

comfort zones where the benefits of real-time

machine-generated information are accessible

More than 80% of companies report “mature”

data analysis capabilities when it comes to

everyday issues of safety, facilities management,

supply chain management, formulation of

capital spending plans, process design, the

use of process controls, asset maintenance

and generalised product quality management

In other words, when production processes

From monitoring to alerting, predicting and solving

to alerts about problems and their causes, half

or more of surveyed companies lack mature capabilities Two-thirds of companies report analytical weakness when it comes to dealing with asset maintenance and throughput alerts, and 76% lack mature capabilities to analyse potential process design issues (Figure 4)

Reporting yes; alerting/predicting/solving—not yet

For which of the following functions and areas does your company have mature data analysis

Capital spending

Supply chain management

Safety and facilities management

Reporting normal operations Alerting about problems

Predicting future problems Prescribing solutions to problems