The Un-Tunable PID Control Loop Best-Practices and Innovations for Tuning Oscillatory, Noisy and Long Dead-Time Processes Robert Rice Vice President, Engineering March 2015 PUBLIC www.rockwellautomation.com www.us.endress.com PUBLIC Agenda Economic Drivers Real-World Challenges Tuning Demystified Real-World Successes Closing Thoughts PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Economic Drivers Process Automation: A State-of-the-State Assessment The Amazing Problem-Free Plant Michael Brown Control Engineering 85% of controllers perform inefficiently when operated in automatic mode 65% of controllers are poorly tuned to mask control-related problems 30% of PID control loops are operated in manual mode 20% of control systems are not properly configured to meet their objectives PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Economic Drivers Top Line and Bottom Line Benefits Invest in Control – Payback in Profits Carbon Trust 2 – 5% Production Throughput PUBLIC 5 – 10% Production Yield 5 – 15% 25 – 50% Energy Consumption Production Defects CHICAGO PROCESS SOLUTIONS SUMMIT Economic Drivers Missed Opportunities for Financial Gain Annual Production & Efficiency Losses Control Station, Inc. PUBLIC $7.6 Million $5.0 Million $1.8 Million $8.0 Million Basic Materials Chemicals Power & Utilities Oil & Gas CHICAGO PROCESS SOLUTIONS SUMMIT Agenda Economic Drivers Real-World Challenges Tuning Demystified Real-World Successes Closing Thoughts PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Real-World Challenges The ‘Black Art’ of PID Controller Tuning Limited Education Chemical Engineering curriculum Single semester totaling 16 hours Not covered by most trade schools Focus on PLC programming Limited Experience Few staff tasked with PID tuning Methods handed down No formalized approach or methodology Out-of-the-box parameters applied Limited Emphasis Other projects deemed more important PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Real-World Challenges The Devil is in the Data Noise Wait for it… PUBLIC Oscillations Wait for it… CHICAGO PROCESS SOLUTIONS SUMMIT Dead-Time Real-World Challenges Where to Turn? Economic drivers Clear opportunities for improvement Strong financials: Payback, ROI Training & experience Limited skilled resources Pool of candidates drying up Traditional ‘state-of-the-art’ software Struggles under ‘real-world’ conditions PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Agenda Economic Drivers Real-World Challenges Tuning Demystified Real-World Successes Closing Thoughts PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT PID Controller Tuning Demystifying the Process Find Step Model Tune Test Identify the Controller and Specify the DLO and Control Objective Perform a “Bump Test” and Collect Dynamic Process Data Fit a Model to the Process Data Use Tuning Correlations to Calculate Tunings Based on Model Implement and Test results 11 PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Document Document the Tuning Process Tuning Demystified Tuning Recipe: A Simplified, Repeatable Process How do you identify PID control loops that need to be tuned? Reactive: Respond to the Operator’s Needs Proactive: Analyze Process Data to Identify PIDs that Contribute to Increased Process Variability Proactive monitoring should: PUBLIC Identify Mechanical, Process and Controller Tuning Issues Facilitate Root-Cause Detection Recommend Appropriate Corrective Action Track and Report Findings CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 1: Find Controller, Specify Objective Good Control is “SIMPLE” PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 1: Find Controller, Specify Objective Reflux Drum – Level Control Example What is/are the primary Control Objective(s)? PUBLIC Maintain Liquid Level In the Reflux Drum Maintain Column Stability Prevent Environmental Release by Avoiding Drum Hi Limit CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 2: Step or Bump the Process Data should show “Cause and Effect” A bump test must generate a response that clearly dominates the random (noisy) PV behavior PUBLIC Here the PV moves approximately four (4) times the noise band – a good value CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 2: Step or Bump the Process Good bump tests Open loop tests require the Controller Output to be stepped PUBLIC Closed loop tests require a sharp Controller Output change CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 2: Step or Bump the Process Bad bump tests AVOID Disturbance-Driven Data & Slow Ramping CO Changes PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 2: Step or Bump the Process Types of process behavior Self-Regulating PUBLIC If all inputs are held constant, the process will seek a steady-state Example: Heat Exchanger Non Self-Regulating Process will only reach a steadystate at its ‘balancing’ point Example: Surge Tank CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 2: Step or Bump the Process Simple First Order Models Self-Regulating Non Self-Regulating ∗ · KP ⇨ Process Gain [ PV ] CO ƬP ⇨ Time Constant [time] · PV KP* ⇨ Integrator Gain [ time·CO ] θP ⇨ Dead-Time [time] θP ⇨ Dead-Time [time] “All models are wrong, some are useful” George Box PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 3: Fit a Process Model First Order Plus Dead-Time (Self-Regulating Model) Process Gain How Far How Far does the PV Move for Change in the Output Process Time Constant How Fast How Fast does it take the PV to reach 63% of its total change PUBLIC 63%∆ ∆ ∆ Process DeadTime How Much Delay How much delay is there from when the CO is changed until the PV first moves CHICAGO PROCESS SOLUTIONS SUMMIT ∆ ∆ Tuning Demystified Step 3: Fit a Process Model First Order Plus Dead-Time (Non Self-Regulating Model) Integrating Process Gain How Far and How Fast How Far and How Fast does the PV Move when the CO is moved from its balancing point Process Dead-Time How Much Delay How much delay is there from when the CO is changed until the PV first moves PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 3: Fit a Process Model Tunings are only as good as the model Manual or Auto-Tune Approaches Sufficient for Simplest of Controllers Software Modeling Much More Robust Open Loop and Closed Loop Noisy and Non-Steady State (NSS) Conditions PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 4: Tune the PID Control Loop 1 First compute, ƬC, the Closed Loop Time Constant A small ƬC provides an aggressive or quick response Choose your performance using these rules: Aggressive: Moderate: Conservative: ƬC is the larger of 0.1Ƭp or 0.8θp ƬC is the larger of 1Ƭp or 8θp ƬC is the larger of 10Ƭp or 80θp PI tuning correlations use this and the FOPDT model values: and PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 4: Tune the Level PID Control Loop IMC tuning correlation: Depending PID, Non Self-Regulating Process 1 The Closed Loop Time Constant, , should be as large as possible but still fast enough to arrest or recover from a major disturbance. PI tuning correlations use this and the FOPDT Integrating model values: 2 1 ∗ PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT 2 Tuning Demystified Step 4: Tune the PID Control Loop Closed Loop Time Constant rules of thumb: Flow Loops Pressure Loops 2 to 4 times the Open Loop Time Constant, Temperature Loops PUBLIC 3 to 5 times the Open Loop Time Constant, 1 to 3 times the Open Loop Time Constant, CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 4: Tune the PID Control Loop Expected PI Controller Response: Conservative Moderate Set Point tracking (servo) response as Aggressive changes Copyright © 2007 by Control Station, Inc. All Rights Reserved. PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 4: Tune the PID Control Loop Challenges of PI Control: Self-Regulating Processes Kc*2 Base Case Performance Kc Kc/2 2 Copyright © 2007 by Control Station, Inc. All Rights Reserved. Ti/2 PUBLIC Ti CHICAGO PROCESS SOLUTIONS SUMMIT Ti*2 Tuning Demystified Step 4: Tune the PID Control Loop Challenges of PI Control: Non Self-Regulating Processes Kc*2 Kc Kc/2 Ti/2 PUBLIC Ti CHICAGO PROCESS SOLUTIONS SUMMIT Ti*2 Tuning Demystified Step 4: Tune the PID Control Loop PI vs. PID Set Point tracking response PUBLIC PID shows decreased oscillations compared to PI performance PID has somewhat: Shorter Rise Time Faster Settling Time Smaller Overshoot CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 5: Implement and Test Results Modified tuning parameters must be tested Testing PID Controllers Typically Involve: Adjust Set-Point to ensure adequate tracking Did the Process Variable overshoot? Did the Controller Output move too much? Introduce a Load Change or Disturbance Did the Process Variable recover quick enough? NOTE: PID controllers work off of controller error (SP-PV). If there is no error, there is nothing for the PID controller to do. You MUST introduce controller error and force the controller to respond before it can be determined if the tuning changes actually improved the system. PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Step 6: Document, Document, Document Who: Who is accountable for the change(s)? What: Which loop was tuned? What were the ‘As Found’ and ‘Recommended’ tuning values? When: When was the loop adjusted? Why: Why was this particular loop tuned? PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Tuning Demystified Industrial-Grade Software for Real-World Applications How do you identify PID control loops that need to be tuned? Reactive: Respond to the Operator’s Needs Proactive: Analyze Process Data to Identify PIDs that Contribute to Increased Process Variability Proactive monitoring should: PUBLIC Identify Mechanical, Process and Controller Tuning Issues Facilitate Root-Cause Detection Recommend Appropriate Corrective Action Track and Report Findings CHICAGO PROCESS SOLUTIONS SUMMIT Agenda Economic Drivers Real-World Challenges Tuning Demystified Real-World Successes Closing Thoughts PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Case Study: Praxair Continuous Improvement & Process Optimization Praxair, Inc. The largest industrial gases company in North and South America and one of the largest worldwide. Over 400 Cryogenic Plants Worldwide On-stream reliability of 99% Standardized on Rockwell Automation Process Controllers Standardized on LOOP-PRO TUNER PID tuning software across all regions The following 2 PID controllers alone contributed between $75K-$100K USD / year of savings PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Case Study: Known Underperformers Continuous Improvement & Process Optimization Impact Stable control at lower value Savings: ~1% higher process efficiency PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT BEFORE 0:01 1:37 3:13 4:49 6:25 8:01 9:37 11:13 12:49 14:25 16:01 17:37 19:13 20:49 22:25 100 90 80 70 60 50 40 30 20 10 0 100 90 80 70 60 50 40 30 20 10 0 AFTER 0:01 1:31 3:01 4:31 6:01 7:31 9:01 10:31 12:01 13:31 15:01 16:31 18:01 19:31 21:01 22:31 Example #1: LIQUID LEVEL CONTROL Instability occurred at lower levels making PID tuning difficult Control the level at a reasonable value (i.e. lower is better) Before: Highly noisy PV Process safety and efficiency impact Case Study: Known Underperformers Continuous Improvement & Process Optimization Change PID loop from Manual to Auto; Stabilize control at higher SP Savings: >2% product recovery increase PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT 0:01 0:24 0:47 1:10 1:33 1:56 2:19 2:42 3:05 3:28 3:51 4:14 4:37 5:00 5:23 5:46 100 90 80 70 60 50 40 30 20 10 0 100 90 80 70 60 50 40 30 20 10 0 SP PV OT 0:01 0:24 0:47 1:10 1:33 1:56 2:19 2:42 3:05 3:28 3:51 4:14 4:37 5:00 5:23 5:46 Example #2: MIXING VALVE CONTROL Mix two flows with different specifications (higher is better) Before: Poor tuning. Once in Auto, nearly tripped the plant. As a result, most of time in Manual, with low PV. Process safety and low product recovery impact Impact PlantESP – TuneVue™ Continuously Watches for Suitable Data For Analysis and Recommends Tunings Parameters Including SP Changes, Manual Bump Tests No configuration required for setting noise limits, minimum step size or window length Model Fits are Generated using full Non Steady State (NSS) Modeling Innovation Tuning Parameters Generated for each loop based on the criteria specified by the user (Fast/Slow, Slider Bar) Reports/Alerts Generated based on Deviation from Recommended Tunings PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Case Study Models and Tuning Range Automatically Determined Level Control of Medium Pressure Steam Separator TuneVue Used Existing Set-Point Changes to Identify A Suitable Tuning Parameter Range PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Agenda Economic Drivers Real-World Challenges Tuning Demystified Real-World Successes Closing Thoughts PUBLIC CHICAGO PROCESS SOLUTIONS SUMMIT Closing Thoughts Demystify PID controller tuning Apply a proven, repeatable recipe Integrate the procedure with existing processes Apply ‘industrial-grade’ technologies Eliminate the steady state requirement Leverage advanced heuristics Proactively address performance issues PUBLIC Improve plant-wide awareness Identify problems, isolate root-causes CHICAGO PROCESS SOLUTIONS SUMMIT Questions Robert Rice, PhD Vice President, Engineering November 2014 PUBLIC www.rockwellautomation.com www.us.endress.com PUBLIC