Multimodel Modelization and

Stabilization of a Bioreactor
Abdelmounaim Kallouq , Asma Karama
LAEPT-URAC 28, Faculty of Science Semlalia, P.B. 2390
University Cadi Ayyad, Marrakech, Morocco

Let us define
Abstract 
−(Xt + Xeq2 ) −(Xt + Xeq2 )
 
(µ(S) − Deq2) ) ∗ Xeq2
 
T −1
g(xt ) =  ,G =   

Q i,j = Fi,j P + P Fi,j + P σ i P
max max
−(St + Seq2 ) Sin − (St + Seq2 ) (−k1 ∗ µ(S) ∗ Xeq2 + Sin ∗ D2eq − Seq2 ∗ Deq2 )  = minλ(−Qi, j) (28)
This paper present a modeling approach of a bioprocess 
δ = maxGT σiGi

i
based on a Takagi-Sugeno formulation (TS). Using the re-
sulted TS model, we construct a Parallel Distributed Com-
2.2 The TS model design using 28 the expression 27 can be increased and we have:
pensation control law (PDC) to ensure the convergence of The non-linearities considered in the model (15) are: V̇ (t) < −||x||2 + δ . It follows that V̇ (t) < 0 for:

the state to a specific equilibrium points. The performance (

Qi,j < 0
z1(x) = µ(S) (16) (29)
of this approach is illustrated by means of a simple example 2
−||x|| + δ < 0
of a biomass growth process.. z2(x) = −Xt (17)
z3(x) = −St (18) So we guarantee that the state is bounded andq
converges to a small origin-centered
ball of radius δ .
1. The global model is given by : Qi,j < 0 is equivalent to

r P −1Fi,j
T
+ Fi,j P −1 + σi−1 < 0 (30)
We consider a nonlinear system where the main feature is
X
ẋt(t) = hi(z(t))(Aixt(t) + Biut(t) + Gi) (19)
that the control input u(k) occurs linearly in the equation re- Remark 1 Using the variable changes
lating inputs to states. It can be represented by an equation i=1
(
of the form: To determine the subsystems(Ai, Bi, Gi) the following rules P 1 = P −1
(31)
ẋ(t) = f (x(t)) + g(x(t))u(t) (1) Yi = KiP 1
are used :
f (.) and g(.) are two nonlinear functions.
Condition 30 can be easily linearized. Using Schur’s complement to 30, it becomes
To transform the model (1) to a T–S fuzzy model, the following rules are:
Rule i : • Model Rule i:  −1 T
P Fi,j + Fi,j P −1 I

IF z1(t) is M i1 and . . . and zp(t). is Mip, <0 (32)
IF z1(t)is Mi1 and . . . and zp(t) is Mip , I −σ
(
T HEN
ẋ(t) = Aix(t) + Biu(t)
(2)
then: If 32 is satisfied for all i = 1, ..., r, j = 1, ..., r
y(t) = Cix(t) i = 1, 2, . . . , r and −||x||2 + δ < 0 then V (t) ˙ <0
ẋt(t) = Aixt(t) + Biut(t) + Gi i = 1, 2, ..., 8 (20)
(
ωi(z(t)) = Πpj=1Mij (zj (t)), i = 1, 2, . . . , r From 24,26 with a Schur complement, it follows that
,t ≥ 0 (3)
ωi(z(t)) ≥ 0 1
Where: −Qi,j > I, i = 1, ..., r, j = 1, ..., r (33)
β
The global model is given by:    
0.0800 0−0.2393 0
A1 =   A5 =  1 1
r  Then all the eigenvaluues of −Qi,j are great then . So <  hold and the radius is
β β
X
ẋ(t) = hi(z(t))(Aix(t) + Biu(t)) (4) −5.4286 −0.3000 −0.8671 −0.3000
bounded by β
i=1

where 
−9.1804 −9.1804
 
−9.1804 −9.1804
 3.1 Constraints On Control Input
B1 = B5 =   , B2 = B6 =   We note here that we deal with an real process and we have a constraint on the
ωi(z(t)) max
control law D < µmax and Sin ≤ Sin . To ensure that, we use the following theorem
hi(z(t)) = Pr (5) −140.0000 60.0000 −0.9506 199.0494
i=1 ωi (z(t)) [?] :

where     Theorem 2 Assume that the initial condition x(0) is known. The constraint ||u(t)||2 ≤ d
−3.2106 −3.2106 −3.2106 −3.2106 is enforced at all times t ≥ 0 if the LMIs:
B3 = B7 =   , B4 = B8 = 
(P
r
i=1 hi (z(t)) = 1

(6) −140.0000 60.0000 −0.9506 199.0494 1 xTt0 P 1 YiT
   
≤ ri=1 hi(z(t))
P
0 ≤1 ≥0 ≥0 (34)
Yi d2I xt0 P 1
(35)
The membership function hi(z(t)) indicates the activation degree of the ith associated   
0.5670 −1.6960 (36)
local model.
G1 =   , G5 =  
−8.1000 24.2291 hold, where P 1 = P −1 and Yi = KiP 1.
2. Bioprocess model A1 = A2 = A3 = A4, A5 = A6 = A7 = A8
G1 = G2 = G3 = G4, G5 = G6 = G7 = G8
The dynamic of the biological process is established from the mass-balance and can
be described by the following equations:

dX
= µ(.)X − DX
dt (7)
dS
= −k1µ(.)X + D(Sin − S)
dt

where : X and S are the concentrations in biomass and substrates and are the state
variables, µ(.) is the specific growth rate, k1 is the pseudo stoichiometric coefficient
and D and Sin are respectively the dilution rate and the influent concentrations of S
and which are the system inputs TS fuzzy control bloc diagram

S
µ(S) = µmax (8)
Ks + S 4. SIMULATION RESULTS
Validation test
where µmax is the maximum specific growth rate; Ks is the saturation constant.
The biomass growth model (7) can be rewritten like an affine model (1) using the Programming theses LMIs with x0 = (5 9) and d = 0.5 we obtain:β = 2868.6
following transformation: If we assumed that:

D = D1 + D2 3. Synthesis Of TS Control Law 

0.0635 −0.0016
 
0.0636 −0.0016

(9) K1 = K2 =
D2
Sin = D1+D2 max
Sin 0.1093 −0.0052 0.1094 −0.0052
The PDC control is the following form: K3 = K4 = K5 = K6 = K7 = K8 = K2
 dX      
dt µ(S)X −X −X D1 r The stabilization around the equilibrium points is ob-
dS = + max (10) X
dt −k1µ(S)X −S Sin − S D2 ut = hi(z(t))Kixt(t) (21) tained and the constraints on the control law are respected as shown in figures ??, ??.
i=1

2.1 equilibrium points Using the quadratic Lyapunov function:

Xeq and Seq are given by resolving these equations: V (t) = xTt P xt, P = P T > 0 (22)
(
(µ − D)Xeq = 0 Lemma 1 Consider two matrices X and Y with appropriate dimensions and σ a sym-
(11)
−k1µXeq + D(Sin − Seq ) = 0 metric positive definite matrix. The following property is verified [?]:

The resolution of these equations lead to tow equilibrium points given by : X T Y + XY T ≤ X T σX + Y T σ −1Y (23)

• Solution 1 : Theorem 1 There exists a time-varying state feedback controller 21 for system 15
( ensuring that the system state converges toward an origin-centred ball of radius
Xeq1 = 0
(12) bounded by β if there exists P = P T > 0, Qi,j and a scalars σi > 0 solutions of The controlled states and the controller outputs
Seq1 = Sin the following optimization problem [?]:

This solution corresponds to the wash of the bioreactor in other words the interruption min β
P,Qi,j ,σi
of the microbial growth. Thus it’s not the good equilibrium points. 5. CONCLUSIONS
• Solution 2:
 
Qi,j I In this paper, we deal with the stabilization problem of a bioprocess. The process is
<0 (24)
(
Ks D
Seq2 = µmax −D I −β described by a nonlinear equations obtained using the mass-balance. This model is
S −S (13)
Xeq2 = in k1 eq2 transformed to a fuzzy T-S model. The stabilization deals are guaranteed by a PDC
with control law respecting the inputs constraints.
Thus, for every doublet D, Sin we have a set of equilibrium points. To avoid that the  −1 T
P Fi,j + Fi,j P −1 I

control law stabilyse the system(7) arround the first equilibrium, we make a variable Qi,j = (25)
I −σi
change as bellow: References
 and
Xt = X − Xeq2

 Gi(t)T σiGi(t) < β (26)
[1] J. Flusser and T. Suk. Pattern recognition by affine moment invariants. Pattern
St = S − Seq2 (14) for i, j = 1, 2, ..., r
 Recognition, 26:167–174, 1993.
ut = u − ueq2

Proof: [2] M. K. Hu. Visual pattern recognition by moment invariants. IRE Transactions on
By derivation of 22 and by substituting 21 in 15: Information Theory, IT-8:179–187, 1962.
we obtain the following system (15):
[3] P. Maragos. Pattern spectrum and multiscale shape representation. IEEE Trans.
V (t) = xTt P xt
ẋt(t) = f (xt) + g(xt)ut(t) + G(t) (15) ˙ = Pr hi Pr hj [(Fi,j xt(t) + Gi)T P xt + xT P (Fi,j xt(t) + Gi)] Patt. Anal. Mach. Intell., 11:701–715, 1989.
V (t) i=1 j=1 t
T T T T
≤ xt (Fi,j P + P Fi,j )xt + Gi P xt + xt P Gi) [4] A. Meijster and M. H. F. Wilkinson. A comparison of algorithms for connected set
where the added term G(t) is a result of the transformation and : openings and closings. IEEE Trans. Patt. Anal. Mach. Intell., 24(4):484–494, 2002.
with :Fi,j = Ai + BiKj
 dXt  
D1 − D1eq2

  [5] P. F. M. Nacken. Image Analysis Methods Based on Hierarchies of Graphs and
dxt   dt µ(S) − D
= ut =   f (xt ) = eq2 Multi-Scale Mathematical Morphology. PhD thesis, University of Amsterdam, Am-
dt −k1 µ(S) − Deq2 ˙ ≤ xT (F T P + P Fi,j + P σ −1P )xt + GT σiGi
dSt
dt
D2 − D2eq2 V (t) t i,j i i (27) sterdam, The Netherlands, 1994.

SEPEX 2017, Semana de Ensino, Pesquisa & Extensão