On-Line Data Analytics

1
On-Line Data Analytics
2
Agenda

• Why on-line batch data analytics
• Development approach
• Lubrizol, Rouen beta installation, results
• Customer feedback, Next Steps

3
Bridging the Gap with On-line Analytics

• On-line Decision Support for Operations Personnel
• Product quality predictions
• Early process fault detection
• Embedded On-line Analytics brings quality
  information, fault detection, and abnormal
  situation knowledge to the operator bridging
  the gap between quality and control.
• Differentiating technology that supports the
  objective of DeltaV remaining the undisputed
  leader in batch.
• The PAT Guidelines issued by the FDA emphasized
  the use of multivariate analytics as a means of
  reducing cost, improving product quality.

4
Why Multivariate Analysis?
UCL

5
Challenges in Applying Online Data Analytics to
Batch Processes

• Process holdups. Tools must account for operator
  and event initiated processing halts and
  restarts.
• Access to lab data. Lab results must be
  available to both the off-line and the online
  analytic toolsets.
• Variations in feedstock. The properties
  associated with each material shipment should be
  available for use in online analytic tools.
• Varying operating conditions. The analytic model
  must account for the 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.

6
Agenda

• Why on-line batch data analytics
• Development approach
• Lubrizol, Rouen beta installation, results
• Customer feedback, Next Steps

7
On-line Batch Analytics

• In September, 2005 a research project was
  established at the University of Texas, Austin to
  investigate advanced process analytics
• The primary objective of this project is to
  explore the on-line application of Multivariate
  Analytics for prediction and fault detection and
  identification in batch operations

8
Concepts Stage

• ISA S88.01 defines stage as a part of a
  process that usually operates independently from
  other process stages and that usually results in
  a planned sequence of chemical or physical
  changes in the material being processed
• Analytic models are defined based on the batch
  stage consistent with goal of Best
  implementation of ISA S88 model.
• The inputs and outputs used in analysis may be
  different for each stage.

9
Impact of Stage on Off-line Model Generation

• When creating PCA and PLS models for a selected
  product, the X variables are defined by stages.
• Some X variables only apply to certain stages of
  manufacturing.

Define X variable per stage using stage selection
10
Selection and Pre-processing of Data

• Visualization of the data used in modeling is
  critical in determining if a batch should be used
  in model generation.
• Comparison of data from multiple batches is
  helpful especially if this is done after dynamic
  time warping (DTW) to align batch lengths
• Based on the beta work, more emphasis will be
  placed on this capability in the future.

Before Data Alignment
After Data Alignment
11
3 Step Monitoring Procedure
Analytics Overview

• If a fault is indicted in the analytics overview
  screen, then selecting the batch number will
  bring up the Fault Detection view.

• If either Fault Detection plot exceeds or
  approaches the upper control limit of 1.0, click
  on that point in the trend and
• -gt Select the Parameter in the lower corner of
  the screen that contributed to the fault
• Evaluate the parameter trends from process
  operation standpoint
• -gt take corrective action if necessary
• Inspect impact of fault on quality prediction
  plot to find out how quality may be affected
• Note Use Up arrow to return to the Analytics
  Overview.

Quality Parameter Prediction
Fault Detection
Contribution
Parameter Trend (s)
12
Operator Training - Web Based Saline
Manufacturing Process Example

• Recipe
• Fill with Water
• Add Salt
• Agitate and heat to boiling
• Maintain boiling for 10 seconds
• Discharge to holding tank

• Measured Disturbances
• Bin Level
• Initial Mixer Temperature
• Water Flow Rate
• Water Temperature
• Discharge Flow
• Media Temperature to mixer
• Media Flow rate
• Unmeasured Disturbances
• Bridging of salt bin
• Fouling of heat exchanger
• Bin level impact on salt screw feeder
• Plugged vent

• Lab Samples
• Salt Density (Kg/liter)
• Outlet Concentration (Kg salt/liter)

Lab Sample Concentration
13
Example Low Hot Oil Flow Rate

• When the hot oil valve is opened, the flow rate
  is much lower than normal
• The lower flow rate impacts the time needed for
  the mixer to reach target temperature extending
  batch time

14
Example Low Hot Oil Flow Rate

• Fault shows up in Indicator 2 deviating above 1.
• To find the cause of the fault, select the point
  of maximum deviation and then choose the
  Contribution Tab or select the parameters that
  contribute most to the fault - shown in the lower
  corner of the screen.

15
Example Low Hot Oil Flow Rate

• The process measurement that had the highest
  contribution to the fault is shown as the media
  flow.
• The media flow is 2.1 liter/sec less than normal.
• A trend of the media flow can be obtained by
  clicking on media flow parameter in the
  contribution screen.

16
Example Low Hot Oil Flow Rate

• The trend confirms that the media flow rate is
  2 liters/sec which is much lower than the normal
  flow rate of 4 liters/second.

17
Example Low Hot Oil Flow Rate

• The prediction plot confirms that the low oil
  flow rate has no impact on the predicted product
  density

18
Agenda

• Why on-line batch data analytics
• Development approach
• Lubrizol, Rouen beta installation, results
• Customer feedback, Next Steps

19
Beta Test at Lubrizol, Rouen

• Lubrizol and Emerson Process Management have
  worked together over the last two years to
  develop and install a beta version of Emersons
  on-line batch analytics
• This new functionality is currently in field
  trails at the Lubrizol, Rouen, France plant

20
Objectives of the Beta Test

• Emerson and Lubrizol are collaborating on the
  development of batch analytic tools to improve
  plant operations. We are working together to
  demonstrate a beta version of this technology in
  two process areas at Lubrizol Rouen, France.
• The primary objectives of the beta installation
  are
• Demonstrate on-line prediction of quality and
  economic parameters
• Evaluate different means of on-line fault
  detection and identification
• Document the benefits of this technology
• Learn from the beta test to update and improve
  these new and evolving modules and user
  interfaces for Emerson to finalize these modules
  for addition to their product offering

21
Setting a Foundation for Analytics
Capture team input using an input-process-output
data matrix

• Form a multi-discipline team that includes plant
  operations

Integrate Lab and Truck Shipment Data
Survey Instrumentation, tune control loops
Conduct Formal operator training
Calculate Feed Tank Properties
22
Summary of Actual Field Trial Analyses

• 2 units / products
• 18 input variables
• 38 process variables
• 4 output variables (2 initially for the online)

• All data at 1-minute time intervals for the
  analysis
• Total of 172 historical batches used for analysis
  and model development across these two processes

23
Results of the Off-line Modeling Work
24
Immediate Results Achieved On-line

• During training classes where the on-line
  analytics were accessed by the students, a
  previously undetected fault in a key feedstock
  mass flow measurement was detected.
• Going unnoticed with traditional monitoring
  systems.

25
Benefits being realized

• The reactor pump intensity was 2x higher than
  normal during two stages in a batch after one
  units day off.
• The pump was closely checked and maintenance was
  ready to operate in case of breakdown.
• A process fault detection led to identifying a
  regular issue on the reactor heating control
  loop.
• This led to actions to revisit the loop tuning on
  key process control parameters.
• Another problem was detected on the hot oil
  heating system.

These problems were going unnoticed with the
existing monitoring and control systems.
26
Benefits being realized

• Problem identified with an up stream boiler
  negatively impacting operations. A quote from
  the operations personnel follows

thanks to the Beta. An equipment failure was
discovered in advance and avoid losing 5 hours
per batch for the batch in process and also for
the following batches before discovering the
problem with the traditional manner. Probably
some days would have be necessary to discover
that type of mechanical problem without the Beta.
(Boiler combustion air controller located in a
bad accessible zone and thermal oil leakage).
(we would have) discovered this latter with the
periodic update of the indicators of efficiency,
but we saved time earlier thanks to the beta.
Earlier is better than too late!
27
Benefits being realized

• The beta test showed through several batches an
  increasing deviation of the density measurement
  of a critical component.
• This phenomenon was linked to the start of
  plugging which was quickly solved by applying
  steam without time cycle impact.
• The on-line tool indicated a problem going on the
  cooling system of the reactor
• It detected that the component charge was being
  introduced too slowly and that the reactor
  temperature was running a little bit higher. The
  problem was solved on the Aero cooler.

28
Benefits being realized

• Quote from the corporate operations dispersant
  team leader responsible for the manufacturing of
  the products associated with this field trial
  following a presentation given to them at the
  start of the on-line trial

my team was really chatting up the capability
at my group meeting afterwards. You can see
where this capability is headedwish we were
further along, but this is a great start.
29
Agenda

• Why on-line batch data analytics
• Development approach
• Lubrizol, Rouen beta installation, results
• Customer feedback, Next Steps

30
Customer Feedback

• Life science customers that have made a request
  to be a beta site for Emersons on-line data
  analytics.
• Amgen West Greenwich, RI
• Lonza Visp, Switzerland Emerson could charge
  50-100K per system for on-line data analytics
  Leander Hertli, Lonza, Head of engineering and
  maintenance
• Genentech Your discussion has created quite
  the buzz within Genentech so much so that they
  would like to move forward as a test site
  Cynthia Gillis, Emerson Business Development
• Martek Beta Proposal Submitted in October, 2009
• Talecris
• WebEx or TIE Presentation on data analytics
• ARC - John Blanchards group April, 2008
• BiogenIdec July, 2009
• CPI/NIBRT July, 2008
• Eli Lilly Web Seminar Sept, 2008
• Martek Feb, 2009, July, 2009
• Novartis Sept, 2008
• Pfizer March, 2008
• Chemical Industry
• Lubrizol
• Celanese Chemicals They have been tracking
  Emersons progress, very interested and excited
  in learning more. We have gotten the attention
  of Celanese engineering team.
  Greg Carr, Control Dynamics

31
John Campbell, BiogenIdec, Manager, Automation,
Raleigh-Durham, NC
32
Sydney Seymour, Talecris
33
TIE Meeting Feedback, Talecris
34
Next Steps

• The Lubrizol, Rouen will be updated and used by
  Lubrizol until a standard data analytics product
  is available. Lubrizol will be an on-going
  reference site for the chemical industry plant
  visits by other customers will be allowed.
• At Marteks request, a beta version of data
  analytics will be installed in early 2010
  providing a life science reference site. Results
  of installation will be published.
• Development of a standard batch data analytics
  product is currently targeted by DeltaV marketing
  as a top priority for v12.
• It will deliver mindshare and further market
  participation in LSFB.
• It will deliver significant incremental
  stand-alone revenues to be quantified by
  marketing.
• Plays a key role in the architecture for
  manufacturing excellence.
• Note Wyeth just spent "high 6 figures"
  with Umetrics, they charge customers around
  20-30K per model where a model per process per
  unit.

35
Architecture for Manufacturing Excellence
Integrated Order and Campaign Information /
Visualization
Web services and OPC
Modeling Prediction Management
Batch Control and Historian
Production Management
Analyzer Management
Process Automation Asset Management
Machinery Manager
Lab Info Mgmt
Busses
Machinery
Analyzers
Instrumentation
36
Where to Get More Information

• Interactive demonstration of data analytics
  applied to the saline process http//207.71.50.196
  /AnalyticsOverview.aspx
• 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.htm
  l
• Emerson Exchange 2009 Workshop 1254 - Benefits
  Achieved Using On-Line Data Analytics - Robert
  Wojewodka, Terry Blevins
• Emerson Exchange 2008 Short Course 366 The
  Application of Data Analytics in Batch Operations
  - Robert Wojewodka, Terry Blevins
• 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, Randy Reiss
• Emerson Exchange 2008 Workshop 412 Integration
  of SAP Software into DeltaV - Philippe Moro,
  Chris Worek
• Emerson Exchange 2007 Workshop 686 Coupling
  Process Control Systems and Process Analytics to
  Improve Batch Operations Bob Wojewodka,
  Philippe Moro, Terry Blevins