Advanced Process Control: An Overview Sachin C. Patwardhan Dept. of Chemical Engineering I.I.T. Bombay Email: sachinp@iitb.ac.in 1
Automation Lab Plant Wide Control Framework IIT Bombay Long Term Scheduling Market and Planning Demands / Raw material availability On-line Optimization Slow Parameter Setpoints drifts PV, MV Multivariable / Nonlinear Control Advanced Control Regulatory (PID) Control MV PV Fast Load Plant Disturbances 2
Automation Lab IIT Bombay Hierarchy of control system functions 3
Automation Lab Why On-line Optimization ? IIT Bombay � Shift if operational priorities Example: FCC Unit operated under � Maximization of Gasoline / LPG production � Maximization of ATF production � Maximization of profits � Minimization of energy consumption � Changes in operating conditions � Changes in feed quality (refinery: change in crude blend) � Changes in operating parameters � Catalyst degradation � Heat-exchanger fouling � Changes in separation efficiency 4
Automation Lab On-line Optimization IIT Bombay Updated Steady Steady State Model State Model Parameter Estimation Cleaned input On-line Steady Output Data State Optimization Steady State Data Reconciliation Set Points Updated Inputs Outputs Operational Goals PLANT 5
Automation Lab Why Advanced Control ? IIT Bombay � Why advanced control? � Complex multi-variable interactions � Operating constraints � Safety limits � Input saturation constraints � Product quality constraints � Control over wide operating range � Process nonlinearities � Changing process parameters / conditions � Conventional approach � Multi-loop PI: difficult to tune � Ad-hoc constraint handling using logic programming (PLCs): lack of coordination � Nonlinearity handling by gain scheduling 6
Automation Lab IIT Bombay Example: Quadruple Tank System dh a a k γ 2 gh 2 gh v 1 1 3 1 1 = − + + 1 3 1 dt A A A 1 1 1 Tank3 Tank 4 dh a a k γ 2 gh 2 gh v 2 2 4 2 2 = − + + 2 4 2 dt A A A 2 2 2 dh a ( 1 ) k − γ 2 gh v 3 = − 3 + 1 1 3 1 dt A A 3 3 Tank 1 dh a ( 1 ) k Tank − γ 2 gh v 4 4 2 2 = − + 2 2 Pump 2 4 dt A A Pump1 4 4 V 2 V 1 Manipulate d Inputs : v and v 1 2 Measured Outputs : h and h 1 2 7
Automation Lab Multi-loop Control IIT Bombay � Industrial Processes: multivariable (multiple inputs influence same output) and exhibit strong interaction among the variables � Conventional Control scheme: Multiple Single Input Single Output PID controllers used for controlling plant (Multi-Loop Control) � Consequences: Loop Interactions � Lack of coordination between different PID loops � Neighboring PID loops can co-operate with each other or end up opposing / disturbing each other 8
Automation Lab Tennessee Eastman Problem IIT Bombay Primary controlled variables: Product concentration of G Product Flow rate 9
Automation Lab TE Problem: Objective Function IIT Bombay 10
Automation Lab TE Problem: Operating Constraints IIT Bombay 11
Automation Lab Model Predictive Control IIT Bombay � Multivariable Control based on On-line use of Dynamic Model � Most widely used multivariable control scheme in process industries over last 25 years � Dynamic Matrix Control (DMC) developed by Shell in U.S.A. (Cutler and Ramaker, 1979) � Model Algorithmic Control developed by Richalet et. al. (1978) in France � Used for controlling critical unit operations (such as FCC / crude column) in refineries world over � Mature technology � Can be used for controlling complex large dimensional systems 12
Automation Lab Advantages of MPC IIT Bombay � Modified form of classical optimal control problem � Can systematically and optimally handle � Multivariable interactions � Operating input and output constraints � Process nonlinearities � Basic Idea Given a model for plant dynamics, possible consequences of the current input moves on the future plant behavior (such as possible constraint violations in future etc.) can be forecasted on-line and used while deciding the input moves 13
Automation Lab MPC: Schematic Diagram IIT Bombay Disturbances Optimization Outputs Inputs Process Dynamic Dynamic Prediction Model Model MPC Plant-model mismatch Set point Trajectory Dynamic Model: used for on-line forecasting over a moving time horizon (window) 14
Automation Lab IIT Bombay CSTR Example Consider non-isothermal CSTR dynamics dC feed flow rate = A f ( C , T , F , F , C , T ) 1 A c A 0 cin dt coolant flow rate dT = f ( C , T , F , F , C , T ) 2 A c A 0 cin dt [ ] [ ] T States ( X ) C T Measured Output ( Y ) T ≡ ≡ A Manipulate d Inputs ( U ) [ F F ] T ≡ Feed conc. c [ ] Unmeasured Disturbanc es (D ) C ≡ u A Cooling water 0 [ ] Measured Disturbanc es ( D ) T ≡ Temp. m cin If model is known, can we estimate C A from measurements of T ? 15
Automation Lab IIT Bombay CSTR: Multi-Loop PI Performance Controlled Outputs PID Pairing 0.4 C A - F c Conc.(mol/m3) 0.35 T - F 0.3 0.25 Linear 0.2 Plant 0 5 10 15 20 25 Time (min) Simulation 400 = k 6 . 34 Temp.(K) 395 c 1 τ = 0 . 2 I , 1 = 390 0 . 0028 k c , 2 τ = 0 . 3 , 2 I 385 0 5 10 15 20 25 Time (min) 16
Automation Lab CSTR: Multi-Loop PI Performance IIT Bombay Manipulated Inputs and Disturbance Coolent Flow (m3/min) 30 20 10 0 5 10 15 20 25 Time (min) 1.5 Linear Inflow (m3/min) Plant 1 Simulation 0.5 0 5 10 15 20 25 Time (min) 2.5 Inlet Conc. (mol/m3) 2 1.5 0 5 10 15 20 25 Time (min) 17
Automation Lab CSTR: LQG Performance IIT Bombay Controlled Outputs 0.45 0.4 Conc.(mol/m3) 0.35 Linear 0.3 Plant 0.25 Simulation 0.2 0 5 10 15 20 25 (No Plant Time (min) Model 398 Mismatch Case) 396 Temp.(K) 394 392 390 388 0 5 10 15 20 25 Time (min) 18
Automation Lab CSTR: LQG Performance IIT Bombay Coolent Flow (m3/min) Manipulated Inputs and Disturbance 30 20 10 Linear 0 5 10 15 20 25 Time (min) Plant Inflow (m3/min) 3 Simulation 2 (No Plant Model 1 Mismatch 0 0 5 10 15 20 25 Case) Time (min) Inlet Conc. (mol/m3) 2.5 2 1.5 0 5 10 15 20 25 Time (min) 19
Automation Lab Linear MPC Applications (2003) IIT Bombay 20
Automation Lab IIT Bombay Industrial Application: Ammonia Plant 21
Automation Lab State Feedback Controller Design IIT Bombay � Step 1 (Model Development) : Develop a discrete time dynamic model for process under consideration � Step 2 (Soft Sensing) : Design a state estimator (soft sensor) using dynamic model and measurements � Step 3 (Controller Design): Assume the states are measurable and design a state feedback controller � Step 3: Implement state feedback controller using estimated states 22
Automation Lab IIT Bombay Models for Plant-wide Control Aggregate Production Layer 4 Rate Models Steady State / Dynamic Layer 3 First Principles Models Dynamic Multivariable Time Layer 2 Series Models SISO Time Series Models, Layer 1 ANN/PLS/Kalman Filters (Soft Sensing) 23
Automation Lab IIT Bombay Mathematical Models Qualitative � Qualitative Differential Equation � Qualitative signed and directed graphs � Expert Systems Quantitative � Differential Algebraic systems � Mixed Logical and Dynamical Systems � Linear and Nonlinear time series models � Statistical correlation based (PCA/PLS) Mixed � Fuzzy Logic based models 24
Automation Lab IIT Bombay White Box Models First Principles / Phenomenological / Mechanistic � Based on � energy and material balances � physical laws, constitutive relationships � Kinetic and thermodynamic models � heat and mass transfer models � Valid over wide operating range � Provide insight in the internal working of systems � Development and validation process: difficult and time consuming 25
Automation Lab IIT Bombay Example: Quadruple Tank System dh a a k γ 2 gh 2 gh v 1 = − 1 + 3 + 1 1 1 3 1 dt A A A 1 1 1 Tank3 Tank 4 a dh a k γ 2 gh 2 gh v 2 = − 2 + 4 + 2 2 2 4 2 dt A A A 2 2 2 dh a 1 k − γ ( ) 2 gh v 3 3 1 1 = − + 3 2 dt A A 3 3 Tank 1 dh a 1 k − γ Tank ( ) 2 gh v 4 = − 4 + 2 2 2 4 1 dt A A Pump 2 Pump1 4 4 V 2 V 1 Manipulate d Inputs : v and v 1 2 Measured Outputs : h and h 1 2 26
Automation Lab IIT Bombay Data Driven Models Development of linear state space/transfer models starting from first principles/gray box models is impractical proposition. Practical Approach • Conduct experiments by perturbing process around operating point • Collect input-output data • Fit a differential equation or difference equation model Difficulties • Measurements are inaccurate • Process is influenced by unknown disturbances • Models are approximate 27
Recommend
More recommend
