Title: Coupling Process Control Systems and Process Analytics to Improve Batch Operations
1Coupling Process Control Systems and Process
Analytics to Improve Batch Operations
- Bob Wojewodka, Technology Manager
- Philippe Moro, Global IS Manager
- Terry Blevins, Principal Technologist
2Presenters
- Robert Wojewodka
- Philippe Moro
- Terry Blevins
3Introduction
- What we will cover
- The need to move beyond process control to
process data analytics coupled with control - Why this is important
- Basic concepts on the analysis methodology
- The Lubrizol ltgt Emerson alliance and
collaborative work to advance these concepts - The beta test field trials
4A Premier Specialty Chemical Company
The Lubrizol Corporation
- 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
5Production in Lubrizol
- Predominantly batch
- Some continuous
- Full spectrum of automation
- Diversity in control systems
- Both reaction chemistry blending
- On-line and off-line measurement systems
6Production Challenges
- 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 on-line
- Continual improvement of Operational Excellence
7Lubrizol / Emerson Alliance
- Alliance agreement
- Pricing
- Conversion to DeltaV / many projects
- Standardize on aspects of PlantWeb architecture
- Collaboration
- Exchanging process optimization and data
analysis, and integration knowledge with Emerson - Emerson sharing knowledge with Lubrizol
- Collaborative development projects
- Lubrizol committed to assist with field trials
and be early adopters
8Analytics Drive the Power of Information
Optimization
The Power of Information
ROI
Predictive Modeling
Descriptive Modeling
What is the Best that could happen?
Ad hoc Reports OLAP
What will happen?
Standard Reports
Raw Data
Why did it happen?
What happened?
Adapted by RAWO from slide courtesy of SAS Inst.
Data
Information
Knowledge
Intelligence
2
93 Levels of Analytics Identified In Lubrizol
Data Analytics
Off-Line
On-Line
Routine data access
Via a Web Page
- Add hoc analyses
- Model development
- Process studies
- Lab studies
- Business studies
- Troubleshooting
- Process improvement
- Interactive analyses
- etc.
- People do their own analyses using the analysis
tools
- Real-time analytics
- Deployment of models
- ASP analytics
- Process analytics
- Monitoring, feedback, control, alerts
- Link back into PlantWeb
- Web interface for the display
- Etc.
- Routine analyses
- Routine reports
- Routine graphical summaries
- Routine metrics KPIs
- Vehicle for data selection by user
- Vehicle to deliver data to the user
- On-line visualization
Applications
Clients
Clients
Clients
10The Challenge of Batch Operation
- The wide range of operating conditions present
challenges in the design, commissioning, and
on-going manufacturing.
11The setting
- Operators work in a highly complex, highly
correlated and dynamic environment each day. Any
help with advanced warning of pending events is
valuable. - Operators manage a large amount of data and
information on a running unit. Even with
automated units, only so much can be monitored
and managed at one time. Any help with automatic
monitoring across many variables is valuable. - The ultimate goal is to prevent the undesirable
effects of an abnormal situation by early
detection or the detection of a precursor to an
undesirable event.
12The Nature Of Batch Data is also a challenge
Each batch has data vectors
Time
Batch 1
Batch 2
Batch 3
Batch 4
Batches
Batch 5
Batch 6
Y - Space
Batches all have variable length time durations
Batch .. bi
X - Space
Quality Measurements
On-line Process Measurements
13The Nature Of Batch Data is also a challenge
Within between batch analysis
Within Batch Variation
Operation Oi, Phases 1 to pi
Operation 3, Phases 1 to pi
Operation 2, Phases 1 to pi
- Analyze data in order to
- Decrease costs
- Reduce time cycle
- Reduce processing problems
- Increase yield
- Improve quality
- Reduce waste
- Reduce variability
- Improve reliability
- Avoid undesirable upsets
Operation 1, Phases 1 to pi
Batch 1
Batch 2
Batch 3
Between Batch Variation
Batch 4
Batch 5
Batch 6
Y - Space
Batch .. bi
X - Space
Quality Measurements
On-line Process Measurements
14What is needed
- Process analysis (off-line, on-line), identify
- Process relationships
- Influential parameters
- Correlation to quality parameters
- Correlation with economic parameters
- Alarming for operators and focused advise
- Process assessment and control
- Monitoring process performance
- Detection of upsets
- Finding assignable causes
- Early detection
- Drill down for explanation of deviations
- Actions taken and feedback
15Also a need to move beyond univariate thinking
Univariate SPC Charts
16The need to move beyond univariate thinking
Variable 2
Variable 1
17The need to move beyond univariate thinking
Multivariate SPC Chart
18But, there are many more than 2 process variables
X2
This relationship of X1, X2, and X3 explains
the next highest amount of the variation between
batches
Even though in 3-dimensions, this relationship
made up of X1, X2 and X3 is the one that
explains the highest amount of the variation
between the batches
Measurements across many batches for these three
variables. We now have a swarm of points in
three dimensions.
X1
Principal Component 1
Principal Component 2
X3
19The analysis reduces this complexity
This then allows us to simplify these complex
process relationships to a much lower dimension
that we can use, interpret, and exploit.
The observations may be projected onto a plane.
Principal Component 2
Principal Component 1
The analysis extends beyond 3 variables to k
variables (k dimensional space)
20Extending the analysis to correlate process
relationships with outputs (the Y-space)
Each observation in the x-space corresponds to a
measured result in the y-space
Multivariate process relationships defined (PCA)
Final batch quality and output relationships
defined (PLS PLS-DA)
21Batch Analytic Challenges
- Linking data with the product identifier
- Operation / phase / state data from the unit
- Many sources of data need to be combined, not all
is within DeltaV - Dimensionality (large in both the x-space
y-space) - Collinearity and autocorrelation
- Noise and missing data
- Multivariate relationships are prevalent
- Addressing process dynamics
- Having access to historical data
- On-line requirements and on-line challenges
- The need for dynamic time warping
- Etc.
22We 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 also established a
research project at University of Texas (UT),
Austin in September, 2005 to investigate advanced
process analytics. - The primary objective of this project is to
explore the on-line application of Analytics for
prediction and fault detection and identification
in batch operations. - Emersons research grant given to UT is funding
the work of a PhD graduate student, Yang Zhang,
under the supervision of Professor Tom Edgar. - Through the LubrizolltgtEmerson alliance, we are
leveraging these areas of expertise to bring the
on-line analytics to a reality.
Preparing For Field Trials
23Summary of Research
- There are many texts available on these topics
-
- Also reference chapter 8 of the book New
Directions in Bioprocess Modeling and Control
Maximizing Process Analytical Technology Benefits
24The Multivariate Analyses Being Used
- PCA Principal Components Analysis
- Provides a concise overview of a data set. It is
powerful for recognizing patterns in data
outliers, trends, groups, relationships, etc. - PLS Projections to Latent Structures
- The aim is to establish relationships between
input and output variables and developing
predictive models of a process. Also used to
help put atypical process variation into a
context of what is important. - PLS-DA PLS with Discriminant Analysis
- When coupled, is powerful for classification.
The aim is to create predictive models of the
process but where one can accurately classify
future unknown samples
25Are these methods new?
- NO, they have been around and in use for quite
some time - These are proven methods for use for
characterizing process relationships - They are just new to some areas such as
- Pharmaceutical companies with their PAT
initiatives - New to many long standing industries such as the
process automation and control and many others - New to certain fields of study such as
engineering, process engineering, control
engineering, etc. - Used very effectively in off-line process
improvement studies - Historically many limitations to move these
methods on-line. But times have changed, we
are now ready!
26For this to work, there needs to be a standard
architecture
SAP
Analysts Chemists Engineers
Analysis servers
Recipe Schedule
.net Web services
Consumption
XML
Data Transfer
Data Transfer
via XML
Embedded Analysis
Operator interface
Batch Exec.
Pro
Data Historian
27A strong level of security to protect the flow
of data
Microsoft Internet Information Service Server
Lubrizol Web Service
Lubrizol Windows Service
Lubrizol Domain
Firewall
Delta V Domain
XML
Microsoft Internet Information Service Server
Lubrizol Web Service
Lubrizol Windows Service
28Integration with data analytics
SQL Batch Historian data
Web Client
Continuous Historian data
Analysis servers
Tags
- Makes the marriage (matrix) between
Batch/Tags/Lab - Utilizes Web services
- Organizes this for easy access from web client
Lab Data -Device -Manual
DeltaV Systems
29Analytics for Batch Processes
30Model Development Aligning Batches
- Data for different length of Batches is aligned
using dynamic time warping - The aligned data is processed using hybrid
unfolding before using this to train the
multi-way PCA or batch-wise unfolding for
PLS/PLS-DA model.
31Support for Process Analytics
History Collection of Lab and Spectral Analyzer
Data
Processing of Sample Data for Use in Analytics
Controller
32Operator Interface for Beta Test
PLS Quality Parameter Prediction
PCA Fault Detection
Contribution Plot
33Planned Beta Installation
- Demonstrate on-line prediction of quality and
economic parameters - Evaluate different means of on-line fault
detection and identification i.e. multi-way
PCA/PLS. - Show value of high fidelity process models for
testing fault detection and alternate control
strategies. - Discuss and explore extension of the
methodologies into other aspects of the process
unit data - Refine the approach, user interfaces, and
integration with other systems
34Beta Installation
35Summary
- Fieldbus Provides Infrastructure for Improved
Diagnostic Capability - Improvements Require New Capability in Field
Devices and DCS Integration of These Features - Multivariate statistical analysis methodology is
needed to correlate relationships within and
between devices with that of operational data - Multivariate statistical analysis methodology is
needed to correlate this with product quality and
other parameters of interest - The analyses need to be coupled with DeltaV
- The analyses need to be available for both
off-line process studies as well as on-line
process diagnostics and control - Its Time to Implement!
36Where To Get More Information
- Many excellent texts on multivariate analysis
- Also reference New Directions in Bioprocess
Modeling and Control Maximizing Process
Analytical Technology Benefits - Bob Wojewodka Robert.Wojewodka_at_lubrizol.com
- Philippe Moro Philippe.Moro_at_lubrizol.com
- Terry Blevins Terry.Blevins_at_EmersonProcess.com
- At the 2008 Emerson Users Conference
- We will be presenting an overview of the field
trials - What was done
- What was found
- What were the benefits
- What are the next steps