Large-Scale Controller Performance, Assessment, Diagnosis and Improvement

1
Large-Scale Controller Performance,Assessment,
Diagnosis and Improvement

• Mike Paulonis
• Eastman Chemical Company
• Presentation to East TN AIChE
• Feb 2005

2
Collaboration

• Co-creator of controller performance assessment
  at Eastman

John Cox
3
Agenda

• Productivity The Driver
• Circumstances Leading to Development
• System Overview
• Performance Assessment Details / Screenshots
• Application Results
• Summary

4
Eastman productivity
Volume sold Employees
15 years ago 1 X 19,000
Today 2 X 12,000
Tremendous productivity gains ( 8/yr) !
5
Sources of productivity
Process Improvement
Process Automation
Information Technology
6
A closer look at process automation

• Generating productivity
• Enable stable plant operation at higher rates
• Enable a higher percentage of first-class product
• Enable plants to run safely with fewer operators
• Affected by productivity changes

Year AdvCT employees
1993 17
1999 14
2005 11
7
Focus on 1999

• Increase in process control work on existing
  plants revealed many control loops with poor
  performance
• Partial cause was a decrease in number and
  availability of personnel to monitor/troubleshoot
  control loops
• Control loop/valve maintenance mode was run
  until failure
• The need for a system to efficiently detect and
  diagnose poor control loops became apparent
• Increase control engineer productivity
• Enable plant engineers to do more on their own

8
More from 1999

• Honeywell introduced "Loop Scout" to general
  customers

• The good assess control loop performance from
  historical operating data
• The bad
• Cost 350 per loop per analysis!
• Only accepted data from Honeywell control systems
• Have to send the controller data to Honeywell for
  analysis

9
We can do better (we think)

• Our vision for controller performance assessment
• Interfaces to all Eastman control and data
  systems
• Analyze thousands of loops automatically and
  periodically
• Rank loop performance to quickly find
  opportunities
• Users can pull up current or historical
  performance of any loop (interactive discovery)
• Users can get email reports on controller
  performance specific to their plant (exception
  processing)
• Reports have summaries understandable by anyone
  and details for in-depth analysis by control
  engineers
• Capture controller tuning changes which often
  link with performance issues

10
Build it!

• Read a lot of academic papers
• Implemented assessment algorithms in Matlab
• Tested/improved using real plant data
• Automated the capture of bulk data from control
  systems
• Automated the analysis of bulk data
• Improved some more hand tuned classification
  algorithm from about 100 example loops
• Built a database to house all the analysis
  results and service user queries
• Built a web application using Cold Fusion to
  bring it all together and make it accessible to
  everyone

11
System architecture

• Distributed data collection
• Distributed computing resources
• Server-side web programming
• Linked in to wide-area network

12
Eastmans large-scale system

• Daily tuning history on 14,000 loops
• Weekly performance history on 8,500 loops

• Ten sites
• Forty-nine process areas
• Interfaces to a variety of control systems

13
Use mode 1 - Tell me about a loop
14
(No Transcript)
15
Example help
16
Use mode 2 - Find problem loops
17
Find problem loops
18
Push content for busy people

• Email subscription reports
• Monthly performance detail
• Top 20 losers and gainers
• Rankings worst to best
• Monthly performance summary
• Performance by loop type
• Performance by loop type over the last year
• Performance benchmark Eastman-wide
• Daily tuning changes
• Notification - tuning changes often mean other
  problems
• Monthly tuning detail

19
Email examples
20
Neat pictures, but does it really work?

• Primary question around valve problems
• Controller performance assessment showed many
  problems
• Most seemed to be valve-related, not tuning
• Valves need maintenance - culture change
• Valve repair/replacement costs
• Don't want a lot of false positives on valve
  diagnoses

Open it 100, OK, now close it!
21
Valve travel testing
ValStat
PROTUNER
22
Polymer controller improvement project
Control Loop Diagnosis Repair
HR2 Spray Recirc Temp Hdw/sticking Add positioner
GR2 Spray Recirc Temp Hdw/sticking Add positioner
FR2 Spray Recirc Temp Hdw/sticking Add positioner
H MeOH Column Reflux Hdw/sticking Fix valve diaphragm
H MeOH Column Level Hdw/hysteresis Replace valve
J MeOH Column Level Hdw/unspecified Replace valve
F MeOH Column Level Hdw/sticking Engage positioner
DR2 Level Hdw/unspecified Repair positioner
C Paste EG Temp Hdw/sticking Repair positioner
HR1A Preht Dow Temp Hdw/sticking Replace valve
23
Example performance improvement
24
Second pass - entire classes

• Reactor Dowtherm control valves all have motion
  problems (some up to 15 deadband)
• Why?
• Long stem
• High-temp packing
• Dirty conditions
• No positioner
• Solution?
• Critical service
• Replace all
• A significant milestone!

25
The trend looks good
26
Improvement in every category
December 2000
September 2002
27
Beating the competition
December 2000 12th of 24 areas 50th percentile
28
The payoff (2001 vs. 2000)

• Reduction in off-class production
• Reactor level - 272 klbs (80 reduction)
• Reactor temperature - 52 klbs (80 reduction)
• Vacuum - 119 klbs (35 reduction)
• Product specs - 96 klbs (35 reduction)
• 98.5 class 1 production - best ever
• Reduction in startup and transition off-class
• 8.1 klb/startup (22), 19.3 klb/transition (70)
• Over 1 million lb/year - maybe 15-25 from
  control
• Line run-life
• H-line best ever 222 days (9/10/01 - 4/19/02)
• Customer complaints down 55

29
Long-term monitoring
Improve
Maintain the gain
Improve
30
The bottom line

• A revolution in knowledge for users of all levels
• Find problems that are not obvious
• Troubleshooting time reduced by 2/3 or more
• Monitor and benchmark performance over time for
  continual improvement
• Enables effective approach to controller
  maintenance
• Encourages communication between control
  engineers, staff engineers, area managers, and
  maintenance forces
• Generates significant economic gains

31
Another example

• Another polymers area embarked on a control
  improvement project in 2003 using performance
  assessment to identify opportunities and track
  results
• Focused on problem loops critical to operation
• Added smart positioners to key valves
• Replaced some undersized valves
• Fixed some slave controller wind-up problems
• Implemented measurement filters where appropriate
• Did some control strategy redesign
• Much more

32
Great results!
33
In search of excellence
34
Other areas can do this too

• Engineer signs up for monthly reports
• Engineer receives and reviews monthly reports
• What is bad, what is good?
• Get management buy-in
• Control loops need maintenance
• Controller performance is important
• Engineer prioritizes loops
• Criticality, return on investment
• Enlist help of CSM, EI shops, Reliability
  Technology and AdvCT as needed
• Engineer arranges funding and schedule for fix
• Engineer follows up to identify tangible benefits

35
Summary

• Controller performance can be determined from
  historical operating data
• Eastman's home-grown system does this weekly for
  thousands of controllers world-wide
• Users can get performance information from a web
  site and/or have it emailed to them
• With the system, work can quickly be focused on
  the best opportunities
• Improved controller performance leads to improved
  process performance and greater profitability
• We can now do much more with less