Sie sind auf Seite 1von 32

EFFECTS OF MODEL PLANT MISMATCH

DIRECTIONS ON IMC PERFORMANCE FOR


FOPTD PROCESS SYSTEMS

Janis K Joseph

Guided by,

Dr.Manju M.S Dr.Arun K Tangirala


Asst.Professor Professor
Dept. of Chemical Engg. Dept. of Chemical Engg.
GEC Thrissur IIT Madras

June 30, 2018

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 1 / 32


Overview

1 Introduction

2 Literature Review

3 Objectives

4 Methodology

5 Observations

6 Conclusion

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 2 / 32


IMC
Internal Model Control (IMC) principle states that “Control can be
achieved if and only if the control system summarizes either implicitly
or explicitly, some representation of the process to be controlled”.[5]

Schematic representation of internal model control

Prediction error:
ˆ = [Gp (s) − Gm (s)]U (s) + d(s)
d(s) (1)
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 3 / 32
Model Plant Mismatch (MPM)
Discrepancies between the model and process in model based controllers
is known as MPM. Performance of any model-based controller depends
on the quality of the model and hence on MPM.

Closed loop transfer function

Gc (s)Gm (s)
Gcl = (2)
1 + [Gp (s) − Gm (s)]Gc (s)

Characteristic equation

1 + [Gp (s) − Gm (s)]Gc (s) = 0 (3)

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 4 / 32


Literature Review
Studies on mismatch directions and thresholds
Suggests a systematic approach to determine the thresholds of
Tufa and Ka, 2016 [19] mismatches. Investigates the effects of direction of mismatch
with respect to gain, time constant and delay.
Studies on root cause of poor control loop performance
Presented a method for uni-variate systems using cross corre-
Stafenlj et al., 1993 [17]
lation to distinguish between disturbances and modeling error.
Proposed methodology uses certain closed loop relations for
Badwe et al., 2010 [2] isolating the root cause of performance degradation. Method
requires setpoint excitation
Developed a new quantity called plant model ratio(PMR).
Selvanathan and Tangi- Proposed method is able to determine the absence or pres-
rala, 2010 [16] ence of MPM even in the presence of poorly tuned controller
or unmeasured disturbances(UD).
Comparison between statistical distributions of nominal out-
put and nominal error are used to capture the source of poor
Batelho et al., 2015 [4] performance(MPM or disturbance). Proposed method is in-
dependent of setpoint, hence flexible to different MPC struc-
tures.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 5 / 32


Contd...

To distinguish between MPM and UD, proposed comparison


Batelho et al., 2016 [3] of statistical distributions of nominal output and modelling
error.
Studies on MPM detection
Introduced three MPM detection indices for detection and iso-
Jiang et al., 2006 [9]
lation of MPM for multivariate systems.
Webber and Guptha, Extended cross correlation approach in [17] to detect specific
2008 [20] mismatched models in multivariate control systems.
Detects MPM from closed loop operating data using partial
Badwe et al., 2008 [1] correlation analysis between model residuals and manipulated
variables.
Using Pmr methodology theroretical conditions to diagnose
Selvanathan and Tangi-
mismatches among gain, time constant, delay and combina-
rala, 2010 [16]
tion of these are developed.
Developed a method for detection and diagnosis of MPM for
Kaw et al., 2014 [10] SISO systems with minimum possible efforts. Hypothesis test-
ing is used to detect mismatch and its causes.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 6 / 32


Contd...

Proposed MPM evaluation based on nominal output sensitiv-


Batelho et al., 2015 [4] ity function. Method is independent of setpoint perturbation
and the data based identification procedure
Proposed method diagnose the presence of MPM and also pin-
Tsai et al., 2015 [18]
points the exact MPM affected elements.
Developed a method for detection and diagnosis of MPM for
Ling et al., 2017 [11] SISO systems with minimum possible efforts. Hypothesis test-
ing is used to detect mismatch and its causes.
Generalized the PMR approach to multivariate control sys-
Yerramilli and Tangi-
tems. Two threshold-based hypothesis tests are presented for
rala, 2016 [21]
diagnosing mismatch in gain and dynamics.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 7 / 32


Objectives

To study the effects of MPM directions on FOPTD systems.

To study which directions of MPM are important.

To investigate whether plant model ratio (PMR) gives better way


of representation of MPM directions.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 8 / 32


MPM Direction
Mathematically MPM can be expressed in two ways:

1 Additive uncertainty

∆G(s) = Gp (s) − Gm (s) (4)

where mismatch in gain, time constant and delay are:

∆K = Kp − Km ; ∆τ = τp − τm ; ∆D = Dp − Dm
2 Multiplicative uncertainty
!
Gp (s) Kp τm s + 1 −(Dp−Dm)s
GP M R (s) = = e (5)
Gm (s) Km τp s + 1

where mismatch in gain, time constant and delay are taken as:
Kp τp Dp
Kα = ; τα = ; Dα =
Km τm Dm
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 9 / 32
y
(GP M R or ∆G)
R
ϕ
x

z
R magnitude MPM representation

GP M R or ∆G is a complex number, which can be represented


using polar coordinate system.

ϕ is the direction of mismatch or MPM direction.

How does performance change with R and ϕ?

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 10 / 32


Methodology
To create mismatches of different directions, spherical coordinate
system or polar coordinate system of mismatch parameters can be
considered.
   
Kα r sin θ cos φ
Mα =  τα  =  r sin θ sin φ  (6)
   
Dα r cos θ
Process transfer function can be written as

(Km × Kα )e−(Dm ×Dα )s


Gp = (7)
(τm × τα )s + 1

If ∆M is considered:

(Km + ∆K)e−(Dm +∆D)s


Gp = (8)
(τm + ∆τ )s + 1

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 11 / 32


Contd...

Performance metrics used are ISE and IAE


Z ∞
ISE = e2 (t)dt (9)
0
Z ∞
IAE = |e(t)|dt (10)
0
Normalized values of integral errors

ISE0
ISEN = (11)
ISE
IAE0
IAEN = (12)
IAE
ISE0 and IAE0 are integral errors of a properly tuned controller

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 12 / 32


Vector Parameters as Multiplicative (Mα )
Gain and Time Constant Mismatch
" # " # (
Kα r cos θ r>0
Mα = = (13)
τα r sin θ θ ∈ (00 , 900 )

(Kα ,τα )
r

θ
X
Mismatch in gain and time-constant as polar coordinates

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 13 / 32


Gain and Delay Mismatch
" # " # (
Kα r cos θ r>0
Mα = = (14)
Dα r sin θ θ ∈ (00 , 900 )

Time-Constant and Delay Mismatch


" # " # (
τ r cos θ r>0
Mα = α = (15)
Dα r sin θ θ ∈ (00 , 900 )

Gain, Time-Constant and Delay Mismatch


    
Kα r sin θ cos φ  r>0


Mα =  τα  =  r sin θ sin φ  θ ∈ (00 , 900 ) (16)
   
Dα r cos θ

 φ ∈ (00 , 900 )

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 14 / 32


Vector Parameters as Additive (∆M )
Gain and Time Constant Mismatch
" # " # (
∆K r cos θ r < min(Km , τm )
∆M = = (17)
∆τ r sin θ θ ∈ [00 , 3600 ]

Gain and Delay Mismatch


" # " # (
∆K r cos θ r < min(Km , Dm )
∆M = = (18)
∆D r sin θ θ ∈ [00 , 3600 ]

Time Constant and Delay Mismatch


" # " # (
∆τ r cos θ r < min(τm , Dm )
∆M = = (19)
∆D r sin θ θ ∈ [00 , 3600 ]

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 15 / 32


Simulations
FOPTD system is considered from [8]
1
Gp = e−5s
10s + 1

Set-point considered for simulation

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 16 / 32


Observations

(a) ∆K% (b) ∆τ %

(c) ∆D%
IAE variation for gain, time constant and delay mismatch %
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 17 / 32
Contd...
Gain and Time Constant Mismatch

Vector parameters as multiplicative (Mα )

ISEN and IAEN variation with GP M R

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 18 / 32


Contd...

ISEN and IAEN variation Kα and τα

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 19 / 32


Contd...
Vector parameters as additive (∆M )

(a) (b)
IAEN variation with (a) GP M R and (b) ∆M
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 20 / 32
Contd...
Gain and Delay Mismatch

Vector parameters as multiplicative (Mα )

(a) (b)
IAEN variation with (a) GP M R and (b) Mα

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 21 / 32


Contd...

Vector parameters as additive (∆M )

(a) (b)
IAEN variation with (a) GP M R and (b) ∆M

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 22 / 32


Contd...

Time-Constant and Delay Mismatch

(a) (b)

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 23 / 32


Contd...
Gain, Time-Constant and Delay Mismatch

IAEN variation
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 24 / 32
Inference

IMC controller allows good set-point tracking and gives good


disturbance rejection especially for the process with a small
time-delay/time-constant ratio.

Filter time constant plays a vital role to obtain better trade off
between robustness and disturbance rejection and stability.

Increase in the filter time constant reduces the overshoot to an


acceptable limit, but however reduces the disturbance rejection i.e.,
desired noise suppression capability.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 25 / 32


Conclusion

Plant model ratio (PMR) gives better way of representation of


MPM directions.

More accurate results are obtained when vector parameters are


considered as multiplicative.

It is observed that the performance is decreasing when ∠GP M R


decreases or increases from zero, but both are not symmetrical.

Performance is highly affected when ∠GP M R i.e., direction of


MPM decreases from zero.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 26 / 32


Thank you

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 27 / 32


References I

[1] A. S. Badwe, R. D. Gudi, R. S. Patwardhan, S. L. Shah, and S. C.


Patwardhan. Detection of model-plant mismatch in mpc applica-
tions. Journal of Process Control, 19(8):1305–1313, 2009.
[2] A. S. Badwe, R. S. Patwardhan, S. L. Shah, S. C. Patwardhan, and
R. D. Gudi. Quantifying the impact of model-plant mismatch on
controller performance. Journal of Process Control, 20(4):408–425,
2010.
[3] V. Botelho, J. O. Trierweiler, and M. Farenzena. Diagnosis of
poor performance in model predictive controllers: Unmeasured dis-
turbance versus model–plant mismatch. Industrial & Engineering
Chemistry Research, 55(44):11566–11582, 2016.
[4] V. Botelho, J. O. Trierweiler, M. Farenzena, and R. Duraiski.
Methodology for detecting model–plant mismatches affecting model
predictive control performance. Industrial & Engineering Chemistry
Research, 54(48):12072–12085, 2015.
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 28 / 32
References II
[5] T. Elakkiya, R. Priyanka, S. Kiruthika, and R. P. Priya. Compar-
ative study of pid, imc and imc based pid controller for pressure
process. Journal of Chemical and Pharmaceutical Sciences ISSN,
974:2115.
[6] C. E. Garcia and M. Morari. Internal model control. a unifying
review and some new results. Industrial & Engineering Chemistry
Process Design and Development, 21(2):308–323, 1982.
[7] W. Hong, X. Lei, and S. Zhihuan. A review for model plant mis-
match measures in process monitoring. Chinese Journal of Chemical
Engineering, 20(6):1039–1046, 2012.
[8] M. Jelali. Control performance management in industrial automa-
tion: assessment, diagnosis and improvement of control loop perfor-
mance. Springer Science & Business Media, 2012.
[9] H. Jiang, W. Li, and S. L. Shah. Detection and isolation of model-
plant mismatch for multivariate dynamic systems. IFAC Proceedings
Volumes, 39(13):1396–1401, 2006.
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 29 / 32
References III
[10] S. Kaw, A. K. Tangirala, and A. Karimi. Improved methodol-
ogy and set-point design for diagnosis of model-plant mismatch in
control loops using plant-model ratio. Journal of Process Control,
24(11):1720–1732, 2014.
[11] D. Ling, Y. Zheng, H. Zhang, W. Yang, and B. Tao. Detection
of model-plant mismatch in closed-loop control system. Journal of
Process Control, 57:66–79, 2017.
[12] J. M. Maciejowski. Predictive control: with constraints. Pearson
education, 2002.
[13] L. E. Olivier and I. K. Craig. Development and application of a
model-plant mismatch expression for linear time-invariant systems.
Journal of Process Control, 32:77–86, 2015.
[14] S. Padhee, Y. B. Khare, and Y. Singh. Internal model based pid
control of shell and tube heat exchanger system. In Students’ Tech-
nology Symposium (TechSym), 2011 IEEE, pages 297–302. IEEE,
2011.
Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 30 / 32
References IV
[15] D. E. Seborg, D. A. Mellichamp, T. F. Edgar, and F. J. Doyle III.
Process dynamics and control. John Wiley & Sons, 2010.
[16] S. Selvanathan and A. K. Tangirala. Diagnosis of poor control loop
performance due to model- plant mismatch. Industrial & Engineer-
ing Chemistry Research, 49(9):4210–4229, 2010.
[17] N. Stanfelj, T. E. Marlin, and J. F. MacGregor. Monitoring and
diagnosing process control performance: the single-loop case. In-
dustrial & Engineering Chemistry Research, 32(2):301–314, 1993.
[18] Y. Tsai, R. Gopaluni, D. Marshman, and T. Chmelyk. A novel
algorithm for model-plant mismatch detection for model predictive
controllers. IFAC-PapersOnLine, 48(8):746–752, 2015.
[19] L. D. Tufa and C. Z. Ka. Effect of model plant mismatch on mpc
performance and mismatch threshold determination. Procedia En-
gineering, 148:1008–1014, 2016.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 31 / 32


References V

[20] J. R. Webber and Y. P. Gupta. A closed-loop cross-correlation


method for detecting model mismatch in mimo model-based con-
trollers. ISA transactions, 47(4):395–400, 2008.
[21] S. Yerramilli and A. K. Tangirala. Detection and diagnosis of model-
plant mismatch in MIMO systems using plant-model ratio. IFAC-
PapersOnLine, 49(1):266–271, 2016.

Janis K Joseph (GEC Thrissur) MPM DIRECTIONS June 30, 2018 32 / 32

Das könnte Ihnen auch gefallen