Building Effective Seed Models For Adaptive Process Control. John Campbell Director, APC Product Management AspenTech

Size: px

Start display at page:

Download "Building Effective Seed Models For Adaptive Process Control. John Campbell Director, APC Product Management AspenTech"

Nigel George
6 years ago
Views:

1 Building Effective Seed Models For Adaptive Process Control John Campbell Director, APC Product Management AspenTech Aspen Aspen Technology, Inc. Inc. All All rights rights reserved 1

He has 15 years of experience with AspenTech in services support, and development. John earned his Ph.D.

2 Our Speaker: Dr. John Campbell Dr. John Campbell Director, APC Product Management Aspen Technology Inc. John leads the group that develops and maintains advanced control and sustained value software products. He has 15 years of experience with AspenTech in services support, and development. John earned his Ph.D. in Chemical Engineering from the University of Wisconsin-Madison 2014 Aspen Technology, Inc. All rights reserved 2

3 Outline Review of Adaptive Process Control The cash curve for building new controllers A new way of thinking Sources of Seed models Pretest pretest is same as always! Models from historical data HYSYS testing a dynamic model HYSYS grabbing SS gains What do we need to get new controllers started? Wants: Easier way to add variables Variables that can be OFF but part of the controller 2014 Aspen Technology, Inc. All rights reserved 3

4 Adaptive Process Control The Key Concepts: Do everything required to update the models without the need to turn off the controller. Ensure that the controller exhibits robust behavior during the periods between model updates. Make maintenance a built-in and continuous part of the process and eliminate the need to wait for turnarounds to revamp controllers. Embed knowledge of APC modeling within the products and leave it to the users to leverage their knowledge of the production process Aspen Technology, Inc. All rights reserved 4

Adaptive Process Control Moves controller maintenance from a project methodology to a continuous intrinsic process. Eliminates many of the tasks requiring control engineering resources.

5 Adaptive Process Control Moves controller maintenance from a project methodology to a continuous intrinsic process. Eliminates many of the tasks requiring control engineering resources. Maintains APC benefits during extended periods where performance degradation has to be tolerated. Exhibits robust controller behavior during periods of model calibration. Engineer is always in the loop regarding model acceptance and deployment. Plant Controller Adaptive Process Control 2014 Aspen Technology, Inc. All rights reserved 5

8 Early Adopter Update More than 30 companies participating Some doing controller maintenance Others are doing new controller development Some highlights from the whitepaper 2014 Aspen Technology, Inc. All rights reserved 8

9 Verifying the Value Claims Reduce effort to re-vamp a controller Reduce the skillset required to perform effective controller maintenance Less disruptive step testing Optimize control during testing Ability to work on multiple projects in parallel Robust controller behavior in the presence of significant model mismatch 2014 Aspen Technology, Inc. All rights reserved 9

10 Applied to a Wide Array of Units CDU Stabilizer Hydrogen Recovery Unit Methanol Unit High Purity Separation Columns Olefins Furnaces Crude Unit Fluid Catalytic Cracking Unit Ultra Low Sulfur Gasoline Production Alkylation Unit Aromatics Unit Butadiene Unit Ketones Solvents Production VBU and VDU Revamp Tail Gas Treatment Unit Air Separation Parallel C3 Splitter 2014 Aspen Technology, Inc. All rights reserved 10

11 Sample Outcomes A user with less than 2 years of APC experience was able to successfully complete a maintenance update for an existing controller An experienced control engineer set a production record while in Adaptive (testing) mode New controllers were developed with an estimated 20 25% reduction in effort One user completed a maintenance project with a budget surplus of half the original allotted amount Early success with the technology has led to plans to implement a rollout program across an entire refinery 2014 Aspen Technology, Inc. All rights reserved 11

R&D Drivers Reduce the effort for all phases of the lifecycle Less complicated product footprint Consolidate all APC functionality into cohesive workflows Faster,

12 R&D Drivers Reduce the effort for all phases of the lifecycle Less complicated product footprint Consolidate all APC functionality into cohesive workflows Faster, consistent on-line deployment Reduce application maintenance labor Increase information for operators Optimize the Cash Curve 2014 Aspen Technology, Inc. All rights reserved 12

Innovations Adaptive Process Control Closed- Closedloop Adaptive Control w/ Quality Data Capable Model Modeling Economic Analysis Slicing loop Robust Model Automatic Capable Identificati Model ID

13 Innovations Adaptive Process Control Closed- Closedloop Adaptive Control w/ Quality Data Capable Model Modeling Economic Analysis Slicing loop Robust Model Automatic Capable Identificati Model ID Relaxation on Maintain Replicates Model optimizing Quality best Automatic generation engineer pinpoints control during judgment MVs for step retest. editing. culmination Accurately Eliminate Relieves of a for of candidate models. data The testing. Assessment tools for engineering identifies costs 10 of year lost areas resources effort capacity, of the rapid evaluation of for model reduced more with valuable quality poor fidelity. performance product tasks 2014 Aspen Technology, Inc. All rights reserved 13

14 Innovations Model Identification Closed-Loop Capable Model ID True MIMO structure Less computationally intensive Good balance between steady-state gains and short-term dynamics Works very well with Multivariable Step Testing Model convergence rate increased 2-5 times! Accurate steady-state gain and gain ratio Numerical stability Handles co-linearity 2014 Aspen Technology, Inc. All rights reserved 14

Innovations Auto Data Slicing Eliminates manual process of removing bad data

Innovations Model Quality Analysis Combined with the Test Agent, continually monitors the

Innovations Adaptive Modeling Automatically generate candidate models for

20 Progressive Modeling Adding Complexity Based on Financial Return Starting with the strong handles (most important MV/CV pairs) Commission with minimal model matrix Add next (small) set of MVs and assess improvements Continue to the point of diminishing returns MV1 MV2 Mv3 MV4 MV5 MV6 MV7 MV8 MV9 Pareto Diagram of MV Importance (strong handles) 2014 Aspen Technology, Inc. All rights reserved 20

21 Impact of Methodology on Benefits Starting with the strong handles (most important MV/CV pairs) Commission with minimal model matrix Add next (small) set of MVs and assess improvements Continue to the point of diminishing returns 2014 Aspen Technology, Inc. All rights reserved 21

22 Shaping the Benefits Curve + Cash Flow Accrued APC Benefits Impact of Adaptive Modeling on sustaining benefits accrual Impact of Automation and methodology on initial spend and benefits rates If no proper maintenance benefits are often back to zero after 4-5 years Time - Cash Flow Capital at Risk t Commissioning Points Traditional Revised 2014 Aspen Technology, Inc. All rights reserved 22

24 Adaptive Process Control for New Controllers The Adaptive Control toolset is easily demonstrated with use of an existing controller, but what if you are converting another vendor s controller or you are building a brand new controller? The tools are still useful, but you must first get a seed model! Next you must set up an application s base tuning. After that the workflow is very similar 2014 Aspen Technology, Inc. All rights reserved 24

Seed Models Sources Fundamental models Empirical models from step test data Domain knowledge operators, process engineers Characteristics Gain errors up

25 Seed Models Sources Fundamental models Empirical models from step test data Domain knowledge operators, process engineers Characteristics Gain errors up to 300% Dominant MVs only! Goal is to get enough model fidelity to enable larger test steps on the process 2014 Aspen Technology, Inc. All rights reserved 25

26 Verifying the Value Claims Reduce effort to re-vamp a controller Reduce the skillset required to perform effective controller maintenance Less disruptive step testing Optimize control during testing Ability to work on multiple projects in parallel Robust controller behavior in the presence of significant model mismatch 2014 Aspen Technology, Inc. All rights reserved 26

30 Pre-test MV steps are performed in a pre-test for two reasons: Generate seed models for Smart-Step Check the PID loops response to setpoint changes and eventually re-tune and adequate them for DMCplus. In a two weeks pre-test is possible to cover MVs and introduce at least 10 steps for each. PID Watch can be used to help analyze and re-tune PID loops more efficiently Aspen Technology, Inc. All rights reserved 30

Event Scheduler (Hysys Dynamic) Manually configure a sequence of

38 Case Study (Steady-State) The case study tool lets you monitor the steady state response of key variables to changes in your process. You designate the independent and dependent variables for each case study. For each independent variable, you specify a lower and upper bound and a step size. HYSYS then varies the independent variables one at a time. With each new change, the dependent variables are calculated and a new state is defined. HYSYS shows the number of states that are calculated as you define the bounds and step sizes of the independent variables Aspen Technology, Inc. All rights reserved 38

40 New Controllers: What do we need? DCS configuration MV points/intermediate points for mode switching WatchDog Shedding logic Scope the DCS work based on the ultimate size of the controller. Do your best to consider all possible MVs Can add more later as needed 2014 Aspen Technology, Inc. All rights reserved 40

41 New Controllers: What do we need? A seed model (more later ) Basic tuning: LP costs Calibrate works best with LP costs of the same order of magnitude. Make sure the directions are correct (negative values drive increase in MV value) Make use of simulation! Set the dynamic tuning Key tuning factors Calibration Ratio: Used to set the degree of optimization Step Fraction: Used to set the overall test aggressiveness 2014 Aspen Technology, Inc. All rights reserved 41

42 Wish List Add a MV or a CV from the web page Simple wizard to gather key information DCS tag information Subcontroller/test group participation Tuning Better support of MVs and CVs without models DMCplus has strict validation to prevent this Enabling blank models will allow controllers to grow into their ultimate scope Manual steps from within SmartStep Allow simple stepping from SS interface Calculate smooth initial curves 2014 Aspen Technology, Inc. All rights reserved 42

43 AspenTech s Community Rachel Lowell Associate Product Marketing Professional Aspen Aspen Technology, Inc. Inc. All All rights rights reserved 43

Register for AspenTech s Online Customer Community To register, please visit www.aspentech.

Q&A Go to the questions tab on the right

49 Advance Process Control Application Development and Online Deployment Advance Process Control Application Development and Online Deployment (APC2400) April 15, 2014 Reading, UK April 30, 2014 Houston, TX June 12, 2014 Reading, UK Explore fundamental concepts of client/server communication and implementation details using Aspen CIMIO TM software. Configure, and maintain Aspen DMCplus Online, Aspen Watch and Aspen Production Control Web Server software. Perform all steps of an Aspen DMCplus project: data collection and extraction, loading and starting a controller, commissioning the controller, and updating configurations as operating objectives change. Explore the capabilities of the Production Control Web Interface 2014 Aspen Technology, Inc. All rights reserved 49

50 APC Calibrate and Aspen Adaptive Modeling APC Calibrate and Aspen Adaptive Modeling (APC2500) May 26, 2014 Reading, UK June 19, 2014 Houston, TX July 28, 2014 Reading, UK Examine the fundamentals of Calibrate mode for APC applications Examine the methodology of implementing new APC applications using Calibrate and Adaptive Modeling/Control. Compare and contrast it to the traditional method for APC implementation Learn how to use Adaptive Modeling through PCWS Learn how to use Calibrate and Adaptive Modeling to maintain and improve existing APC applications Commission the new model developed with Calibrate and Adaptive Modeling for both new and existing APC applications 2014 Aspen Technology, Inc. All rights reserved 50

51 Advanced Process Control Performance Monitoring Advance Process Control Performance Monitoring (APC2600) April 30, 2014 Reading, UK May 12, 2014 Houston, TX July 2, 2014 Reading, UK Learn how to use Aspen Watch KPI to analyze and troubleshoot controller performance, and how to detect and repair model mismatch in Aspen DMCplus controllers. Learn to sustain the benefits achieved by your control applications. Learn how Aspen Adaptive Control can help improve performance of the controller through generating and deploying updated models Aspen Technology, Inc. All rights reserved 51

Tuning interacting PID loops. The end of an era for the trial and error approach

Tuning interacting PID loops. The end of an era for the trial and error approach Tuning interacting PID loops The end of an era for the trial and error approach Introduction Almost all actuators and instruments in the industry that are part of a control system are controlled by a PI(D)