Transcript The Application of Data Analytics in Batch Operations

The Application of Data
Analytics in Batch Operations
Robert Wojewodka, Technology Manager and Statistician
Terry Blevins, Principal Technologist
Presenters
2

Robert Wojewodka

Terry Blevins
Introduction



Lubrizol Rouen project background and
objectives
Challenges of applying online analytics
Beta project steps
–
–
–
–
–
–
–


3
Collection of process information
Integration of lab and tank property data
Instrumentation and control survey
Historian collection
Model development
Training
Evaluation
Summary
More information - references
The Lubrizol Corporation
A Premier Specialty Chemical Company
Building on our special chemistry, a unique blend of people,
processes and products, Lubrizol:
 Provides innovative technology to global transportation,
industrial and consumer markets
 Pursues our growth vision to become one of the largest and
most profitable specialty chemical companies in the world
4
A special chemistry aligned for financial success
Lubrizol’s Production Facilities






5
Predominantly batch
Some continuous
Full spectrum of
automation
Diversity in control
systems
Both reaction
chemistry and
blending
Online and off-line
measurement
systems
Production Challenges
6

Addressing the required batch data structures

Better addressing process relationships

Characterizing process relationships sooner

Identifying abnormal situations/events sooner

Better relating process relationships to end
process quality and economic parameters

Moving process data analytics online
Online Data Analytics

Through the use of Principal Component
Analysis (PCA) it will be possible to detect
abnormal operations resulting from both
measured and unmeasured faults.
– Measured disturbances – may be quantified
through the application of Hotelling’s T2 statistic.
– Unmeasured disturbances – The Q statistic, also
known as the Squared Prediction Error (SPE), may
be used.

7
Projection to latent structures, also known as
partial least squares (PLS) may be used to
provide operators with continuous prediction
of end-of-batch quality parameters.
Online Data Analytics
PCA – Fault Detection
PLS – Quality Parameter Prediction
Contribution Plot
8
We Feel We Have a Solution

Lubrizol has expertise and a long-standing use of
multivariate data analysis in support of off-line process
characterization and process improvement activities.

Emerson Process Management established a research
project at University of Texas Austin in September 2005 to
investigate advanced process analytics.
– The primary objective of this project is to explore the online application
of analytics for prediction and fault detection and identification in batch
operations.
– Tools for PCA/PLS model development and online application have been
developed.

9
Through the Lubrizol<>Emerson alliance, we are leveraging
these areas of expertise to bring the online analytics to a
reality.
Rouen Beta Installation
Collaborate on the development of Emerson’s tools for online prediction of process, quality and economic
parameters
10
Challenges in Applying Online Data
Analytics to Batch Processes






11
Process holdups. Tools must account for operator and eventinitiated processing halts and restarts.
Access to lab data. Lab results must be available to the online
analytic toolset.
Variations in feedstock properties associated with each
material shipment should be available for use in online
analytic tools.
Varying operating conditions. The analytic model should
account for batch being broken into multiple operations that
span multiple units.
Concurrent batches. The data collection and analysis toolset
and online operation must take into account concurrent
batches.
Assembly and organization of the data. Efficient tools to
access, correctly sequence, and organize a data set to
analyze the process and to move the results of that analysis
online.
Technical Advancements

Two advancements enable batch analysis and
online implementation of online analytics.
1. A new approach known as hybrid unfolding offers
some significant technical advantages in unfolding
batch data for use in model development.
2. A relatively new technique known as dynamic time
warping (DTW) is an effective approach for
automatically synchronizing batch data using key
characteristics of a reference trajectory.

12
However, as with any engineering endeavor, the
success of the project depends greatly on the steps
taken to apply this analytic technology.
The Steps the Project is Following


Our approach at the Rouen plant will be further
refined and followed for future applications. Thus,
considerable thought is being given to project
planning to achieve an installation success.
The 7 project steps are:
1.
2.
3.
4.
5.
6.
7.
13
Collection of process information
Integration of lab and tank property data
Instrumentation and control survey
Historian collection
Model development
Training
Evaluation of performance
Beta Project Execution
14

Most of the time required to apply online
analytics is associated with collecting process
information, instrumentation and control
survey, integration of lab data, setup of
historian collection, and training.

A well-planned project and the use of a multidiscipline team play a key role in the
installation success.
Collecting Process Information
15

Important that the team has a good
understanding of process, the products produced
and the organization of the batch control.

Important to have a multi-discipline team

Project meetings were conducted at the plant to
allow operations to provide input and for the
team to become more familiar with the process.

This formed the basis of what we refer to as the
Inputs – Process – Outputs data matrix.
Defining Analytic Application
Capturing project
meeting discussions
Data matrix defining parameters to
be considered in the project
16
To address this application, a multidiscipline team was formed that
includes the toolset provider, as
well as expertise from Lubrizol’s
plant operations, statistics, MIS/IT,
and engineering staff.
Beta station
mapping modules
Beta Installation


17
Beta station is
layered on the
existing Delta
system using
OPC.
Mapping
modules were
created in the
beta station to
allow process
and lab data to
be collected in a
single historian.
Integration of Lab Data

Key quality parameters associated with the Rouen plant batch
operation are obtained by lab analysis for grab sample. Then,
a company typically enters the lab analysis data into its ERP
system (SAP® software in the case of Lubrizol)

The properties analysis for truck shipments are also entered
into SAP® software.
– To allow this data to be used in online analytics, an interface was
created between the SAP® software system and the process
control system.

The material properties associated with truck shipments are
used to calculate the properties of material drawn from
storage
– It is important to characterize both the quality characteristics of
incoming raw materials and the quality of end of batch
characteristics.
18
Integrating Lab and Truck Shipment Data


19
Lubrizol and
Emerson
developed
applications to
integrate lab data
contained in SAP®
software
Online analytic
results will also be
supplied to SAP®
software through
this Web service
interface
Accounting for Feed Tank Properties

Tank
Design 1
Storage Tank Design
20
Tank
Design 2
Tank
Design 3

Storage material
properties are
calculated using
multicompartment
tank model.
Using the
configuration of
the mixing and
point of entry
parameters, the
tank behavior
can be modeled
as fully mixed
(CSTR), plug
flow or short
circuiting.
Tank Properties (Continued)



21
The tank property
calculations are
implemented as a
linked composite
block.
The truck or lab
material properties
(max. of 7 per tank),
timestamp and
transfer quantity are
wired as inputs to
composite block.
Outputs of the
composite block
reflect the
calculated material
properties.
Instrumentation and Control Survey
22

A basic assumption in the application of analytics to a
batch process is that the process operation is very
repeatable.

If there are issues associated with the process
measurement or control tuning and setup, then these
should be addressed before data is collected for model
development.

Parallel to the initial project meeting, an instrumentation
and control survey was conducted for the two batch
process areas addressed by the project.

Also, changes in loop tuning were made to provide best
process performance.
DeltaV Insight for Loop Tuning
23

Beta station
modules were
created to shadow
control loops.

DeltaV insight was
used to examine
loop and get tuning
recommendations.
Loop Tuning (Continued)
24

Process loop
dynamics and
gain were
automatically
identified based
on normal batch
operation.

Recommended
tuning is based
on the identified
process
response.
Historian Collection

When the Rouen plant’s process control system was
originally installed, all process measurements and
critical operation parameters associated with the batch
control were set up for historian collection in 1-minute
samples using data compression.

However, for analytic model development, it is desirable
to save data in an uncompressed format.

This information is collected using 10-second samples
and saved in uncompressed format.
– This allows the data analysis to be done at a finer time
resolution and to also further define a more appropriate
resolution for future implementation.

25
Analysis of the data will then define if the resolution
needs to remain at a fine resolution or if it may be
reduced.
Historian Collection (Continued)
DvCH data extraction utility developed
to create initial datasets for model
development
26

Emerson developed a special
application as part of the
project to create the initial
data sets needed for model
development.

Functionality of this
application is being
incorporated into the model
development tools. The
design allows for data files to
be exported for use in other
offline applications.
Model Development




27
The model development tools are designed to allow the
user to easily select and organize from the historian a
subset of the data associated with parameters that will be
used in model development for a specified operation(s)
and product.
The tool provides the ability to organize and sequence all
of the data into a predetermined data file structure that
permits the data analysis.
Once a model has been developed, it may be tested by
using playback of data not included in model
development.
Since the typical batch time is measured in days, this
playback may be done faster than real time. This allows
the model to be quickly evaluated for a number of
batches.
Interface for PCA and PLS Model
Testing


28
Historian data files
may be played back
faster than real time.
Testing is done with
data not used in
model development.
Training
29

The plant operator will primarily use the
statistics provided by online analytics.
Therefore, operator training is a vital
part of commissioning this capability.

Also, separate training classes on the
use of the analytic tool will be
conducted for plant engineering and
maintenance.
Evaluation



30
During the first three months of the online
analytics, operator feedback and data collected on
improvements in process operation will be used to
evaluate the savings that can be attributed to
analytics.
It also will be used to obtain valuable input to
improve user interfaces, displays, and the
terminology being used in the displays.
This will allow the project team to further improve
the analysis modules to maximize operators’ and
engineers’ use and understanding.
Business Results Achieved



31
At Lubrizol’s Rouen, France plant online analytics
are being applied to batch processes for fault
detection and prediction of quality parameters.
This application in the specialty chemical industry
contains many of the batch components commonly
found in industry.
The analytic toolset Emerson with Lubrizol are
collaboratively developing for this installation is
specifically designed for batch applications and
incorporates many of the latest technologies, such
as dynamic time warping and hybrid unfolding.
Summary
32

The use of statistical data analytics will likely cause
people to think in entirely new ways and address process
improvement and operations with a better understanding
of the process.

Its use will allow operational personnel to identify and
make well-informed corrections before the end-of-batch,
and it will play a major role in ensuring that batches
repeatedly hit pre-defined end-of-batch targets.

Use of this methodology with allow engineers and other
operations personnel to gain further insight into the
relationships between process variables and their
important impact of product quality parameters.

It also will provide additional information to help process
control engineers pinpoint where process control needs to
be improved.
Where to Get More Information
33

Robert Wojewodka and Terry Blevins, “Data Analytics in
Batch Operations,” Control, May 2008

Video: Robert Wojewodka, Philippe Moro, Terry Blevins
Emerson - Lubrizol Beta:
http://www.controlglobal.com/articles/2007/321.html

Emerson Exchange 2008 Short Course: 364 – Process
Analytics In Depth - Robert Wojewodka & Willy Wojsznis

Emerson Exchange 2008 Workshop: 367 – Tools for
Online Analytics - Michel Lefrancois and Randy Reiss

Emerson Exchange 2008 Workshop: 412 – Integration of
SAP® Software into DeltaV - Philippe Moro & Chris Worek