Loop-Shaping Controller Design from Input-Output Data Kostas Tsakalis, Sachi Dash, ASU Honeywell HTC Control-Oriented ID: Uncertainty description compatible with the controller design method. (Loop-Shaping for available insight & computations.) Linear-model estimation: Coprime factor uncertainty (forward and feedback) Controller Design: Loop-Shaping (S&T) Final Result: Quick design & implementation, Performance, Reliability CDC December 999
The Application: Control of Paper Machines Inputs: Stock Flow Dryer temperatures (set-points to the local PID s) Machine speed set-point Outputs: Dry weight Moistures Machine speed Disturbances: Operators change set-points in other loops to maintain overall product quality. Feed consistency, especially after paper breaks (re-circulation). Other nonlinear effects Latex Stock Chest Yankee After Dryer Wire/Felt 3% Scanners Reel Figure. Paper Machine Schematic CDC December 999 2
System Identification: Nominal Model and Uncertainty Description Control-Oriented ID: Uncertainty description compatible with the controller design method. Our choice: Loop-Shaping (available insight, computations) based on sensitivity and complementary sensitivity targets. Nominal Model: MISO equation error, yielding a linear estimation model y = N(θ)[u] + D(θ)[y] + e = w T θ Estimated parameters include initial conditions; this is important to handle inputoutput sets that begin on a transient. Left factorization and a coprime factor description of uncertainty: Handling of low frequency perturbations and changes in unstable modes. Indicates required low-frequency sensitivity roll-off (disturbance attenuation) Easy computation of target loop properties Generalities: System ID e u y [ F, θ, q, θ ] 3 Figure 2. Structure of Parameter Estimator [ F, θ 2, q, ] CDC December 999 3
Generalities: Uncertainty Description Error Sources: Plant Input and Plant Output processed by perturbation subsystems (uncertainty) Uncertainty bound computation and target loop selection: Minimization of a robust stability condition σ [ U s C S M - W E ] σ [ N ] + σ [ Y s S M - W E ] σ [ M ] < U S Ν M Y S Controller-independent approximations Input-output scaling and whitening to handle disparate channel-bandwidths and measurement units. u s W E e u y [ F, θ, q, θ 3 ] e w y s [ F, θ 2, q, ] Figure 2. Structure of Identification Uncertainty CDC December 999 4
Identification Honeywell s high-fidelity simulator (TDC3) Perform Identification y = N(θ)[u] + D(θ)[y] + e Simulate Id d system clockwise from top left: dry weight size-press moisture reel moisture machine speed Nominal plant: M - N, (M=I-D) dry weight reel moisture.4.2 -.2 (Identification fit) -.4 5 5 6 4 2-2 -4 5 5 s.p.moisture speed.5 -.5 (s.s. means removed) - 5 5 5-5 - 5 5 time in min CDC December 999 5
Identification 2 Identified plant singular values Eigenvalue spread:.57-5 RHP transmission zeros:.95, 2., 2.4,.7+/-2j, 4.3 Main interactions Stock flow - moisture Machine speed - moisture and dry weight Plant Singular Values - -2-3 -4-2 - 2 3 rad/min CDC December 999 6
Controller Design Target selection (T & S) approx. per-channel contributions Uncertainty constraints (more critical in moisture channels) RHP zeros constraints (more critical in dry-weight) Final check through robust stability condition Achievable specs determined within a few iterations. Closed-loop T & S and Uncertainty bounds 2 - -2-3 2 3 3 2 4-2 - rad/min CDC December 999 7 4
Controller Design 2 Weighted sensitivity minimization (standard software) A computational alternative: Glover-McFarlane Plant-Controller augmentation S&T approximate loop-shaping Avoids g-iteration but increases dimensionality Fairly efficient with scalar targets Other alternatives, 2-DOF compensators (Open) min K W KSM min K SM M, N = ncf ( G) T Θ W n S Θ M m o M SM H TM o o o H Controller Implementation Observers to substitute missing measurements Anti-windup modifications CDC December 999 8
Controller Design 3 Controller singular values Reduction to remove unnecessary modes (low/high frequencies) Nominal Step responses in four directions. Some coupling between channels remains due to RHP zeros 4 Controller Singular Values.2 Nominal Step responses (four directions) 2.8.6-2.4-4.2-6 -2-2 3 rad/min -.2 5 5 2 25 time (min) CDC December 999 9
Controller Design 4 Robust stability condition Tests with different controllers Closed-loop confidence: effective mult. uncertainty for scaled and un-scaled output δ CL,e < {σ [ S M - W E ] σ [ U s C S ] σ [ T - ] σ [ N ] + σ [ Y s S M - W E ] σ [ M ] } α κ(y s ) α = ( σ [ U s C S M - W E ] σ [ N ] - σ [ Y s S M - W E ] σ [ M ] ) -; κ(y s ) = σ(y s ) σ(y s - ) Stability Condition Check (<) 6 Effective Closed-loop Multiplicative Uncertainty - 4-2 2-3 -2-4 -2-2 CDC December 999-2 - 2
Controller Evaluation Step responses at nominal steady-state (identified) Reasonably smooth control activity Overall behavior assessment: excellent Small error definitions: dry weight ~. (lb) moisture ~. -.2 (%) higher for reel moisture. C.L. Steps at ID ss. (s.s. means removed) Control Activity 5 (s.s. means removed) dry weight -. -.2 -.3 s.p.moisture -. -.2 stock flow - -2-3 s.p.temp -5 -.4 5 5 2 -.3 5 5 2-4 5 5 2-5 5 2 5 reel moisture.5 -.5 5 5 2 speed 5-5 5 5 2 time in min r.temp -5 - -5-2 5 5 2-5 5 5 2 CDC December 999 speed s-p 5
Controller Evaluation 2 Consistency Disturbance (unreasonably large) at t=4. Quick recovery without excessive errors. Steady-state transition to a new operating point under closed-loop control. (Normally done in open loop) Overall behavior assessment: excellent 3 Dist+ss transition 2 (s.s. means removed) Control Activity 8 (s.s. means removed) dry weight 2 s.p.moisture.5.5 stock flow 5-5 s.p.temp 6 4 2-5 5 2 Big consistency dist. at 4 -.5 5 5 2-5 5 2-2 5 5 2 5 6 reel moisture 5 speed - r.temp 4 2-2 speed s-p - -5 5 5 2-2 5 5 2 time in min -4 5 5 2-2 5 5 2 CDC December 999 2
Controller Evaluation 3 Step responses at new operating condition More coupling appears but still within acceptable limits Overall behavior assessment: very good.2 C.L. Steps at new ss. (s.s. means removed) 6 Control Activity (s.s. means removed) dry weight -.2 -.4 s.p.moisture -. -.2 stock flow 4 2-2 s.p.temp 5-5 - -.6 5 5 2 -.3 5 5 2-4 5 5 2-5 5 5 2 6 2 6 reel moisture.5 -.5 - speed 4 2 r.temp -2 speed s-p 4 2 -.5 5 5 2-2 5 5 2 time in min -4 5 5 2-2 5 5 2 CDC December 999 3
Conclusions Approach: Coprime factor uncertainty estimation Flexible and reasonable framework (theoretically, computationally) Compatible with established loop-shaping controller design Controller Performance Excellent disturbance attenuation properties (sensitivity minimization) Reliability (minimal iterations, work well the first time) Controller design Quick design turn-around time Very quick refinements! Fast execution cycle Simulation results Full first-principles, Nonlinear, High fidelity, but still a simulator Preliminary results with real plant data support the same conclusions CDC December 999 4