PREDICTIVE

FUNCTIONAL
CONTROL

                                           J.Richalet 2015
 WHO INVENTED

PREDICTIVE CONTROL ?
  ERGONOMY OF FLIGHT
CONTROL

•  In the fifties :
•  How the pilot manages his aircraft ?
•  What operating image ?
•  Comparison between:
•  Human control and Automatic control :

SURPRISE ! ….  
 
  Jean Piaget (1896 / 1980)

•  Swiss Grand Father of Cognitive Psychology

•  How the child learns and controls his

environment

. Four phases ….!

     
 4 PIAGET BASIC PRINCIPLES

–  OPERATING IMAGE –  INTERNAL MODEL

–  TARGET – Sub TARGET –  REFERENCE TRAJECTORY

–  ACTION –  STRUCTURATION OF THE MV

–  COMPARISON BETWEEN
–  ERROR COMPENSATION
PREDICTED AND ACHIEVED
RESULT

• NATURAL CONTROL : “ YOU DO NOT DRIVE YOUR CAR WITH A PID SCHEME”
PREDICTIVE CONTROL

IS NOT

AN INVENTION

BUT A DISCOVERY ….!

 ELEMENTARY TUTORIAL EXAMPLE

G e-θ s y
P= =M= θ = Tsamp . r
1 + Ts u Trajectory expo. : λ
-Tsamp
y(n) = α y(n-1) + (1- α) . u(n-1-r) . Gα = e
T 1 coincidence point: H
Processus with delay yPr (n): = yP (n) + yM (n) - yM(n-r)
1 Base function: step
Target : ΔP(n+H) = (C - yPr) (1 - λH)

Model : Model increment :

Free mode y M (n) α H
Increment = Free(n+h) + Forced(n+h)- ymodel(n)
Forced mode u(n) . GM ⎛⎜1 - α H ⎞⎟
⎝ ⎠ The only mathematical problem for trainees …!
(Liebniz 1674)
Control equation : Δ P (n + H) = Δ M (n + H)
(C - y (n)) (1 - λ H) = y M(n) α H + u(n) GM(1 - α H) - y M(n)
Pr

(C - y (n)) (1 - λ H) y (n)
u(n) = Pr + M
GM(1 - α H) GM
BAYER
reactor
 TEMPERATURE DE MASSE
120
Tmax=108.4°
d°
100
92°

80

60

Tinit=42°
40

20

min
0
0 20 40 60 80 100 120
Figure 4
 ON LINE ESTIMATOR of
DISTURBANCES
Pert B

+ +
Cons1 MV1
R1 P SP
+ +

MV2
SP*
Cons2=SP R2 M≡P

-
Pert*
+
 ESTIMATION DE L‘EXOTHERMICITE
120

100
Tmasse/ Tmasse
estimée

80

60

EXOTHERMIC
40 ITE

TS
20

TS-
TE
0
min
0 10 20 30 40 50 60 70 80

Figure 3
 T2

Rea cta nt Steam

Flow T3
Heat
C ooler Exc hanger
T4

T1 Reactor

BASF
31
BASF
4 CONTROL STRATEGIES OF BATCH REACTORS

R R Reagent dT
r CMV M= UA T - T + DHx
M P M dt e M
r e CeP Ve dTe = r e Fi Ti - Te + UA TM- Te
¥
TM
dt
Fi, Ti Te
.
q(Fi) TM+ TM= Ti

1) Fi =ct /MV= Ti CV=TM / level 0 =Ti ? :PFC

2) Ti =ct (!) MV=Fi Parametric Control non linear :PPC
3) MV : Ti and Fi : Enthaplic control (power) :PPC+
4) MV : Pressure of reactor :PFC
15
Batch Reactor : MV=Qf / CV= TM
DEGUSSA EVONIK

TM

Qf

Tof

Tif
  Control PID Degussa / EVONIK

temperature [°C]
+/- 2 °C

39 different batches with PID

time

Datum | Name der

Seite | 17
Präsentation
Control PFC Degussa / EVONIK

32 different batches with PPC

+/- 0,2 °C
temperature [°C]

time

Datum | Name der

Seite | 18
Präsentation
CONTINUOUS CASTING
 poche de 335 t d’acier liquide à 1550 °C

consigne de poids
busette rotative

A PFC
quenouille t

mesure de poids
mesure de position
distributeur de 60 t
consigne=70% vérin busette à 2 ouïes
PFC PI A
amplificateur
de puissance
capteur
mesure de niveau radio-actif

lingotière à largeur variable

850-2200 mm,
refroidie à l’eau

cages d’extraction
   IDENTIFICATION
   -­‐    Ultra-low level test signals ! ! ….

     «    I do want to see your test signal on the level ..!«    

           
 -­‐  Set of harmonics signals
Eigen function

 -­‐  High level Parallel filtering

 .-­‐  Different metal alloys

   
COMPLEX ALGEBRA PREDICTIVE CONTROL ?!
BULGING COMPLEX CONTROL

70

60

50
without with
40

30
STOPPER
20

10
Amp sinus

0
Bmp cosinus
-10
Frequency reference
  
   -­‐
Bench test of pump gaskets
 
-­‐      Flow : 35000 m3/h ….! ?
- Pressure : 17.50 MPa
- Temperature : 330 d°
 
 CONTROL ENVIRONMENT

Pressuriser
Steam generatorr

REACTOR
PUMP

PFC controller
 :
" Size : 10 m
" Mass : 100 tons
" Flow: ~35000 m3/h

Engine

Pump

Support

PFC control joints

 Prise en tendance de la
température d’entrée
F(T0)

Débit
T5 T0 T31 source T0
chaude
Convexité Echangeur Chaud

TEQ = L x T31 + ( 1 – L ) x T0
Régulation débit
PFC CHAUD
chaud T31
Logique de
Consigne de choix de T51
température l’action
(Chaud ou
Froid)
Régulation débit
PFC FROID
froid
Convexité Echangeur Froid T30

TEQ = L x T30 + ( 1 – L ) x T51 T5

Débit
T5 T51 T30 source
froide
Prise en tendance de la
température d’entrée
F(T51)
 PID
position vanne (%)
Température (°C)

Essai A2si du 17 avril 2012 (673)

65,0

55,0

45,0

35,0

25,0

Temps (heures)
15,0
18:14:24 18:21:36 18:28:48 18:36:00 18:43:12 18:50:24 18:57:36 19:04:48
 PFC
position vanne (%)
Température (°C)

Essai A2si du 18 avril 2012 (676)

100,0

80,0

60,0

40,0
T Inj

Consigne

20,0 retard +3 mn

Temps (heures)
0,0
14:31:12 14:45:36 15:00:00 15:14:24 15:28:48 15:43:12 15:57:36 16:12:00 16:26:24
F1, T1
Convexity Theorem

F, T

F2, T2

F.T =F1.T1+F2.T2 (enthalpic balance)

T=λ.T1 +(1- λ).T2 with 0≤λ≤1

Hyperbolic function ! λ = F1/(F1+F2)

 tsf

tsp qp, tep

qf, tef

1 1
1 − exp[−U.A( − )]
Fp Ff
Γ(Qf ) =
Fp 1 1
1− exp[−U.A( − )]
Ff Fp Ff
Fp, Ff : Thermal flows
Fp = (ρ.Cp)p.Qp Ff = (ρ.Cp)f.Qf.
  
SANOFI  
VERTOLAY  
VITRY sur SEINE
 
ARAMON

MONTPELLIER

KÖLN

Training of staff: Transfer of Know How

 DIFFUSION  and  PROMOTION  

•  Training  in  many  countries  

•  Universi@es,  technical  schools  (IRA  in  France)  
•  Training  of  professors  
•  Con@nuous  educa@on  of:  
–   Teachers  of  technical  schools  
–   Industrial    operators  
•  Documenta@on:  “Techniques  de  l’ingénieur”  
 Prac@cal  implementa@on  with  
Scilab  

•   The  Scilab  PFC  book:  

Text  +  Diagram  +  Program  Code  
•  45  programs  in  Scilab  implemen@ng  PFC  
control:  
–  Elementary  
–  Ordinary  
–  Advanced  
 Elementary  example:  control  of  an  integrator  process  with  delay  

// E_com_intg12
// process H(s) :integrator with delay : H(s)=exp(-R.s)/s
clear; xdel(winsid());clc
tf=1200;w=1:1:tf;
u=zeros(1,tf); MV=u;CV=u;
SP=u;sm=u;DV=u;
tech=1;//sampling period
r=50; // dealy R=r*tech
G=0.01; // proportional gain
tau=1/G; // closed loop time constant
am=exp(-tech/tau); bm=1-am; //model parameters
trbf=250; lh=1-exp(-tech*3/trbf);
SP=100; // set point
for ii =2+r:1:tf,
// perturbation
if ii>600 then DV(ii)=-0.2; end
// MV proportional loop
ec(ii)=MV(ii-1-r)-CV(ii-1); // error
MVp(ii)=G*ec(ii);
CV(ii)=CV(ii-1)+MVp(ii)+DV(ii);
// pfc
sm(ii)=sm(ii-1)*am+bm*MV(ii-1); // model
spred=CV(ii)+DV(ii)*1+sm(ii)-sm(ii-r);
d=(SP-spred)*lh+sm(ii)*bm;
MV(ii)=d/bm; // MV pfc
end
scf(0)
plot(w,SP*ones(w),'k',w,CV,'r',w,DV*100, 'k',w,MV,'b')
a=gca(); a.grid=[1,1]; a.tight_limits="on";
a.data_bounds=[1,-40;tf,140];
xlabel('sec');
xtitle('Control of a process : integrator + delay CV r / MV b /
DV*100 k')
 Ordinary  example:  control  of  a  loop  with  constraint  transfer  

// O_pfc_transparent
// back calculation, transparent control of level
clear,xdel(winsid()),clc
tf=1000; //duration of test
w=1:1:tf; //time;
tech=1;//sampling period
u=zeros(1,tf);
CV=u;//process output
MVp=u; MV=u;//manipulated variable
eps=u; //error
sm=u; //output of model
k=0.1; // gain of integrative process level
Kp=0.07; // gain of P(ID)
tau=143; // =1/(k*Kp)
am=exp(-tech/tau);
bm=1-am; //time constant of internal closed loop system by P(ID)
trbf=200; //desired closed loop system by PFC: only tuning factor
h=1; lh=1-exp(-tech*3*h/trbf);
fl=input('fl (1 with back calculation / 0 without back calculation) = ');
MVmax=5; //maxvalue of mv
SP(1:tf)=100;
for ii=2:1:tf,
eps(ii)=MV(ii-1)-CV(ii-1); // error of P(ID)
MVp(ii)=Kp*eps(ii); // manipulated variable of P(ID)
if MVp(ii)>MVmax then MVp(ii)=MVmax;end // constraint on the mvp
mvbc=(MVp(ii)/Kp)+CV(ii-1); // back calculation
CV(ii)=CV(ii-1)+k*MVp(ii-1);// output of process
sm(ii)=sm(ii-1)*am+bm*(fl*mvbc+(1-fl)*MV(ii-1));// internal model of
PFC with/without transfer of constraint
d =(SP(ii)-CV(ii))*lh+sm(ii)*bm; // computation of mv
MV(ii)=d/bm; // computation of mv
end
scf(0)
plot(w,SP','k',w,MVp*10,'m',w,CV,'r',w,MV,'b');
a=gca(); a.grid=[1,1]; a.tight_limits="on"; a.data_bounds=[2,0;tf,230];
xlabel('sec')
if fl==0 then title('without back calculation SP k / MVp*10 m / CV r / MV
b'),end
if fl==1 then title('with back calculation SP k / MVp*10 m / CV r / MV
b'), end
 Conclusion
 
•  Dissemination? : where is the problem?
•  The technical and economic efficiency of
PFC is clearly demonstrated on many different
processes World Wide
•  Implementing PFC: the Scilab PFC book