ARC BRIEF

MAY 2008

Multivariate Analysis: Enabling the

Next Killer Apps
By Tom Fiske

Keywords

Advanced Process Control, Multivariate Analysis, Quality by Design,

Process Analytical Technology, Online Monitoring

Summary

There are numerous applications and solutions available to manufacturing

companies that help improve efficiency, lower costs, improve quality, and
increase profitability. Of all the possible choices, one technology – Multiva-
riate Analysis – is emerging that has the po-
The pressure to improve manufacturing
performance is intense. Manufacturers are
tential to dramatically augment the way
finding it difficult to improve product quality companies operate by providing increased
and provide early detection of process upsets understanding and insight to manufacturing
and shutdowns. Enhancing asset performance operations. Multivariate Analysis finds use
with the aid of a real-time process monitoring as a continuous improvement tool, as an of-
system can result in significant cost savings
fline advisory decision support system, and
by providing early warning and diagnostic of
as an online monitoring and prediction solu-
process excursions and upsets as well as
defining optimal operating regions.
tion. In all cases, the benefits are undeniable.
The technology is complementary to other
techniques and solutions employed by most users, including advanced
process control. This document provides a brief overview of the manufac-
turing operations space as related to Multivariate Analysis.

Manufacturers Control Complex Processes and Make

Complex Decisions Daily

Manufacturers use complex processes to make their products. As such,

they must make complex decisions regarding which processes to use, how
to best use them, and how to best control them in an ever changing world
where satisfying demand requires balancing conflicting requirements of
agility and efficiency. Some decisions, such as designing quality into the
process, occur with low frequency and others, such as adjusting operating
parameters, occur at high frequencies over short time intervals. Despite the

THOUGHT LEADERS FOR MANUFACTURING & SUPPLY CHAIN

 ARC Brief, Page 2

frequency, however, all of these decisions have a significant effect on prof-

itability and require a thorough understanding among the relationship of
many interacting variables to ensure an optimal outcome.

To facilitate the decision making process, manufacturers automate

processes and rely upon a host of solutions and applications. In fact, a typi-
cal plant may have more than two-hundred applications. Some of these
applications help organizations plan their activities while others help to
execute those plans. Still others are used for assessing performance, pro-
viding insight and greater understanding, and making improvements.
Despite all the applications and solutions employed by operations, most
companies still struggle with making optimal decisions about how to best
operate and control their processes – especially in real-time.

Multivariate Analysis is particularly well suited to address the growing

need to extract meaningful information from complex data in an offline or
online fashion. Multivariable Analysis can easily determine the magnitude
of each contributing factor. It can also be used to determine the significant
relationships that exist between variables, thus providing the insight and
understanding companies need to improve their operations.

The Need for Speed, Agility, and Tight Process Control

Recent advances in technology are creating a new global playing field
where information, knowledge, and resources are connected without re-
gards to geographic boarders or political boundaries. Improvements in
supply chain planning, management, and execution along with the Internet
continue to make it easier to make, sell, and support products globally.
Competition on a global basis presents new market opportunities, but also
presents greater challenges. Customers demand greater product variety,
improved responsiveness, along with prod-
The value of information is ucts that satisfy local specifications and
inversely proportional to
the time until it can be comply with regulatory requirements.
actionable
Competing in this new arena requires lo-
Value

The value of information

is directly proportional
wering costs and improving operational
to the number of people
collaborating efficiency, agility, responsiveness, and qual-
ity. Many companies are already using
Time
Multivariate Analysis successfully by moni-
Breadth of Collaboration toring processes to ensure products are
Value of Information made to tight quality specifications and to

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 3

mitigate process upsets or faults. In addition, MVA provides the informa-

tion necessary to make faster more knowledgeable decisions in a variety of
areas that help improve efficiency and lower operating costs.

Too Much Data and Not Enough Information

Companies today continue to lose their most valuable asset – people – to
retirement. The exodus of workers severely erodes a company’s knowledge
base and compromises their productivity if not duly compensated by other
means.

Increasing competitive pressure is forcing enterprises to deploy their physi-

cal and human assets more effectively. The greater scrutiny of
manufacturing operations leaves the smaller workforce with less time to
collect data and make insightful analysis, and correspondingly good busi-
ness decisions about operations. With less time for analysis, people must
not only work harder, but also work smarter. Employees must have access
to critical “on-demand” information to make better decisions to improve
the bottom line.

The key to success is not to generate and

Understanding distribute more data, but to improve un-
derstanding of the data and distribute
INDEPENDENCE

Understanding Principles
CONTEXT

Knowledge
timely information that companies al-
ready have in numerous disparate
Understanding Patterns
Information
applications scattered throughout their
organization. It also means using em-
Understanding Relationships
Data bedded “intelligent” applications that
interpret the data and take action. Solu-
UNDERSTANDING
tions that provide manufacturing
Knowledge Is a Key Competitive Advantage
intelligence are a key element in the dis-
tribution and interpretation of information. Staying competitive means
transforming all the data in the plant into information and making that in-
formation available in the proper context to all personnel involved in
operations. It also means adopting methods to better monitor and control
processes in real-time. People spend over half their time looking for and
analyzing data to perform their jobs. This effort is wasteful and unproduc-
tive. In many cases, different people looking at the same multi-dimensional
problem, will make the abstraction based upon their own biases and judg-
ments and come up with different conclusions. Therefore, the abstraction

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 4

should be done automatically and explicitly. MVA is capable of greatly

simplifying the analysis process while providing deeper insight and under-
standing of the data. In addition, real-time process monitoring using MVA
has proven effective in prediction and dealing with critical conditions that
lead to off-spec product or unscheduled downtime.

Disparate Plant Systems and Applications Hinder Performance

Most manufacturers acquire applications and solutions on an ad hoc
project-by-project basis. This situation has left them with disparate systems
and islands of information that make it difficult to get a real-time unified
perspective on manufacturing operations.

With the aging workforce and the loss of knowledge and production know-
how, companies are looking for ways to simplify their manufacturing envi-
ronments by reducing the number of applications it needs to support and
by standardizing on applications across their enterprise. They also need to
make a smaller workforce more productive. One of the stated top priorities
for manufacturers is to increase manufacturing performance visibility. This
is more difficult than it appears as the many non-integrated applications
and systems hinder performance visibility. Consequently, users are focus-
ing their efforts on solutions that automate production control and assist in
making knowledgeable production decisions through improved perfor-
mance visibility and manufacturing intelligence. Many companies today
are focusing on aggregating and disseminating production data. This effort
provides significant benefit, but it does not go far enough. Analytical me-
thods, such as MVA are required to bring an additional level of intelligence
to the data.

Making Sense out of the Manufacturing Application MES

As stated earlier, manufacturing operations use complex processes and

make complex production decisions. They use a host of manufacturing or
production applications that cover a broad range of functionality. There are
many ways to classify these applications. One method is to use the ISA-95
standard, which categorizes information flows and major activities within
the production environment. The major production activities include:

• Detailed Production Scheduling

• Production Resource Management

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 5

• Production Dispatching

• Product Definition Management

• Production Execution

• Production Data Collection

• Production Tracking

• Production Performance Analysis

• Maintenance Operations

• Quality Assurance Operations

• Inventory Operations

• Inventory Operations/Transfer Tracking

• Management of Security

• Management of Configuration

• Management of Documents

• Management of Regulatory Compliance

Product Production Production Production

definition capability schedule performance

Detailed
production
scheduling
Production
Production
resource
tracking
management

Production Performance
dispatching Analysis

Product Production
definition data
management collection

Production
execution

Equipment and Process Operational Operational Equipment and Process

Specific Production Rules Commands Responses Specific Data

Level 2 Process Control

ISA 95.03 Manufacturing Operations Functions

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 6

Many of these activities can be broken down further. For instance, Produc-
tion Performance Analysis covers production unit cycle times, resource
utilization, equipment utilization, procedure efficiencies, and production
variability. The information is normally fed to other systems and used to
optimize production and resources. Production performance analysis is a
continuous process that requires reexamining policies as raw materials,
equipment, process configurations, and business objectives change. Simi-
larly, Production Execution can be broken down into several elements that
include advanced monitoring and advanced process control.

ARC closely follows the production management application environment

used by manufacturing operations. ARC expands upon the ISA- 95 stan-
dard by incorporating other functions and activities and by grouping
similar functions in order to obtain greater insight into the production op-
erations environment. ARC has
Quality, and
Production performance data Management created the Collaborative Manu-
Tracking Policies facturing Management Model to
and Models
help users extract greater value
Production from their production assets.
Resource
Availability
Performance The model promotes greater col-
Analysis
laboration and information
Physical sharing between people, organi-
Process and
Production Production zations as well as applications.
data Operating data, Models
collection Equipment Status,
Resource usage
The applications based upon
Changes to Multivariate techniques fall into
Processes, Procedures, several categories within the Col-
and Production
laborative Manufacturing
Activity Model for Production Performance Analysis
Management model. For in-
stance, MVA, DOE (Design of Experiments), PAT (Process Analytical Tech-
nology), and online monitoring have are important components of Real-
time Process Optimization ((RPO) and includes Advanced Process Control
(APC)), Human Machine Interface (HMI), Plant Asset Management (PAM),
Quality Control, Manufacturing Intelligence (MI), Manufacturing Execution
Systems (MES), Continuous Improvement (CI), etc. MVA works with
many other applications in this collaborative environment to provide valu-
able real-time monitoring capabilities and an additional level of
understanding to many domains within the operations management space.

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 7

CMM Model Depicting the Relationship of OM and Other Applications

Benefits of Multivariate Analysis

Manufacturers in the process and hybrid industries are facing a major chal-
lenge to improve product quality and provide early detection of process
deviations, excursions, and potential breakdowns. Preventable anomalous
or atypical process conditions account for losses approaching 3% to 5% of a
plant’s total capacity each year. This translates into hundreds of thousands
if not millions of dollars in lost revenue each year per plant. Losses accrue
not only from reduced throughput from unplanned shutdowns, but also
from off-spec production, equipment damage, reduction in asset availabili-
ty, disruptions to schedules, safety hazards, and environmental
remediation. Offline data analysis can help define optimal operating condi-
tions. In addition, real-time process performance monitoring systems
capable of predicting and preempting process excursions help to ensure
safe and reliable operations while maintaining quality and profitability.

At the execution level, a real-time process performance monitoring system

provides a standardized approach for performance feedback and control.
For a process performance monitoring system to be effective, it must pro-
vide current performance information and analyses in a time frame that is
consistent with the production cycle so that corrective actions can be taken
to improve product quality, avoid process excursions, and enhance plant
performance.

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 8

Plant automation and information systems collect, process, and store thou-
sands of real-time measurements that provide the basis for real-time
process performance monitoring. It is possible to use this data to determine
a region of optimal performance and to maintain operations within this re-
gion through diligent process monitoring and control.

Multivariate Analysis provides many benefits to users in all industries. The

pharmaceutical industry is already embracing Multivariate Analysis as a
means to help achieve its Process Analytical Technologies (PAT) initiatives.
PAT is a major component of the risk-based manufacturing approach
promulgated by the FDA in order to encourage pharmaceutical companies
to adopt the latest manufacturing techniques and practices. Multivariate
Analysis is well suited for designing quality into the process using DOE
and monitoring the process by making timely measurements of critical
quality parameters during processing to assure acceptable end product
quality at the completion of the process.

Other industries have achieved significant benefits using both offline and
online Multivariate Analysis techniques. Still, the adoption level is not at
the level commensurate with the benefits. One of the biggest misconcep-
tions is that multivariate analysis is superfluous because many
manufacturers already use APC. The two technologies are not mutually
exclusive. While it is true that APC improves process stability and
throughput, it does not have early fault detection capability or the ability to
aid in root cause analysis and continuous process improvements.

Multivariate Data Analysis (MVA) techniques,

One of the biggest misconceptions is
that multivariate analysis is superfluous such as Principle Component Analysis (PCA) and
because many manufacturers already Partial Least Squares (PLS), are better suited to
use APC. The two technologies are not provide early indications of conditions that lead to
mutually exclusive. While it is true that impending faults and to determine the root cause
APC improves process stability and of problems. In manufacturing operations, there
efficiency, it does not have early fault
are typically many variables that together deter-
detection capability or the ability to aid
mine the final product results. In most situation,
in root cause analysis and continuous
process improvements. these variables are not independent of each other.
Since the manufacturing process is not driven by
the number of arbitrarily measured variables, but by the underlying
process itself, the relationship between each variable must be taken into
account using Multivariable Data Analysis. A key feature of PCA is its abil-
ity to model a process or a quality data set with a smaller, simplified set of

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 9

variables. Using available process data, a control region can be created

representing the optimal run conditions. Real-time data is compared to this
optimal control region to detect and predict potential faults. In addition,
the decomposition of the PCA model is useful for providing process insight
and determining the cause of a process upsets.

Multivariate Analysis tools are well beneficial for a wide range of plant per-
sonnel including operators. The value of Multivariate Analysis tools for
operators cannot be understated. These front line workers are responsible
for sustaining performance in the face of changing requirements, process
disturbances, and equipment degradation. The fact that monitoring normal
process data with traditional techniques provides little early warning of
potential problems and the underlying causes, seriously threatens opera-
tional and business performance of manufacturing companies. These
threats often go unnoticed while the process quietly progresses towards a
critical condition because the cause is hidden by the relationship among a
large number of variables.

Multivariate Analysis is not only a powerful tool for operators, but also for
engineers and maintenance personnel. It provides engineers with a struc-
tured methodology to define normal operating ranges, perform detailed
analysis of process faults, and establish a better understanding and interac-
tion of complex process variables. Multivariate Analysis is also helping
operations and maintenance personnel monitor the condition of important
pieces of equipment that are difficult to interrogate by ordinary asset man-
agement techniques.

Conclusions

From ARC’s perspective, process manufacturers’ current challenge lies in

improving Return on Assets. There is a large amount of unrealized per-
formance remaining in manufacturing assets. Offline Multivariate Analysis
tools and real-time process performance monitoring applications can help
improve product quality, reduce costs, and increase productivity while
helping to maintain the process in a safe state.

• Companies need to develop a strategic plan for the adoption and per-
vasive use of Multivariate Analysis tools. Organizations must consider
using Multivariate Analysis applications where major process upsets
have a detrimental effect on asset availability, schedules, and cost.

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com

 ARC Brief, Page 10

• Companies need to also identify which processes can benefit from hav-
ing a more rigorous methodology for monitoring, controlling, and
troubleshooting processes, both in advanced of upsets and afterwards
to diagnose the problems.

• Organizations should consider using real-time measures of perfor-

mance to make decisions about the process and provide the necessary
tools down to the operational level where products are made and
processes are adjusted.

This paper was written by ARC Advisory Group on behalf of Umetrics. The opi-
nions and observations stated in the paper are ARC's. For further information or
to provide feedback on this paper, please contact the author.

©2008 • ARC • 3 Allied Drive • Dedham, MA 02026 USA • 781-471-1000 • ARCweb.com