Identification and Control During Fermentation Start-Up - a Method to Avoid Synchronous Growth in Continuous Cultures

Copyright © lFAC Computer AppliCalions in Biotechnology. Garrms ch-Partenkirchen. Gennany. 1995

IDENTIFICATION AND CONTROL DURING FERMENTATION START-UP - A METIiOD TO A VOID SYNCHRONOUS GROwn! IN CONTINOUOUS CULTURES

Hanne Msller and Sten Bay Jergensen

Process Design, Dynamics and Control Group Department of Chemical Engineering Technical University of Denmarlc. DK 2800 Lyngby

A control strategy is proposed to avoid development of synchronous growth in carbohydrate limited fermentations. The basic idea is to control the metabolic flux through the pathways using an ethanol measurement and substrate feed rate to keep the ethanol concentration at a low level. An adaptive and a constant parameter controller were investigated experimentally. Both were initialized at the target conditions for the continuous fermentation, where the uncontrolled process is known to be unstable. The latter fact renders it infeasible to attempt open loop operation at the critical dilution rate. The adaptive controller turned out to be superior to the non-adaptive controller. The superiority of the adaptive controller is ascribed to its ability to identify the process under varying cell activity . The experimental results demonstrate that the desired operating point is reproducibly obtainable. However, after prolonged operation under different types of disturbances the yeast seemed to adapt towards an increased respiratory activity for the same low level of ethanol in the fermentation broth. Keywords : closed loop control, fermentation process, reproducibility

I . INTRODUCTION Large-scale continuous fermentations are started up by fITSt performing a batch phase with a volume much smaller than the volume of continuous fermentation. When the cells have reached the early stationary phase a fed-batch phase is initiated. After the desired volume is reached the outlet is opened and the substrate flow rate is flxed. Fed-batch operation is performed by following a set-point table for either the feed rate alone or for both feed rate and substrate concentration. However, the fedbatch phase (Axelsson, et al. 1988) and the start-up operation of the continuous culture are difficult to perform without triggering ethanol production and then possibly also triggering synchronous cell growth. One method to control ethanol production is to perform feedback control of e.g. substrate feed rate (Dairaku, er al. 1983; Axelsson, et al. 1988) using a suitable measurement or estimate. Because available sensors have been limited to pH-electrodes, DOT- and temperature sensors, open-loop control of the feed rate in order to obtain maximal productivity is still very common (Axelsson, er al. 1988). On-line estimation of cell density would render it feasible to perform closed-loop control of the cell mass production. Unfortunately, on-line estimators of the cell mass require a reliable model that it is difficult to provide. Another approach to obtain nearly maximal productivity is to control a metabolite, e.g. ethanol, which can be measured either in the liquid 317

phase (Axelsson er al. 1988), or in the gas phase (Jmgensen, er al. 1992; Andersen, et af. 1995). A reducing gas sensor was selected as reasonable control variable, provided the set point is at a low level. In the present work the start-up of a continuous fermentation of Saccharomyces cerevisiae on simple medium was investigated. The working volume of the tank was relatively small (8.15 I). Thus a batch operation with a volume equal to that of continuous operation could be applied without technical difficulties . The control task was limited to control the process, where the cells adapted their active biomass and increased in number. Hence, the cell composition had to change, and the feed demand would follow a saturation proflie. A conventional constant gain PlO-controller is not expected to be able to manage this control task due to the varying fermentation characteristics, whereas an adaptive controller should be able to cope with such variations. Moreover, start-up procedures of cultures of Saccharomyces cereviciae where the metabolic activity is not controlled often result in synchronous cultures. The non-linear control task and the industrial desire to avoid synchronous cultures renders an investigation of the start-up control task most relevant.

2. EXPERIMENTAL PROCEDURE The experiments were carried out in the setup described by (Andersen, er ai. 1995). Every period of continuous

operation dealt with in this paper was preceded by a batch phase. In the batch phase the culture grew oxidoreductively on glucose. When the glucose concentration dropped below about 40 mgl" the yeast cells started to adapt to assimilate the formed ethanol. Towards the end of the batch phase when the ethanol was assimilated the reducing gas sensor signal dropped. At this point the cell density was about I gl·' . If the substrate feed rate to reducing gas concentration loop was closed at this point in time, which was the ideal case, the controller automatically could start the flow rate when the measurement value dropped below the set point (Y RG.SP = 10 i.e. ca. 40 mg 1-'). The applied adaptive controller (MIMOSC) was designed using a 3rd order ARX-model with a time delay of 3 min. The controller was tuned with a penalty on changes in the manipulated variable (Qu=I ,OOO) resulting in a somewhat sluggish but robust controller. The IMC-PI-controller was initialized with a measurement filter time constant of20 min. When the set point was almost reached filter time constant was reduced to 3 min.

3. RESULTS AND DISCUSSION

Start-up under PI-control. Several start-up experiments were performed under an IMC-designed PI-controller. The resulting start-up times were quite similar and so were the transient behaviour. Thus only one case: D 1911 N is shown in Figure I a-c. The major difference in transient behavior between the cases is that the tuning of the controller in some cases, which are not shown here, was not sufficient to remove the offset before the closed loop time constant was reduced towards the end of the start-up. In experiment DI911N the dilution rate to reducing gas sensor loop was closed at 0 min, and the agitation rate set manually to 100 rpm resulting in an oxygen scarcity, until the RPM to DOT loop was closed at 1:09 h. In partial anaerobiosis yeast is expected to grow oxido-reductively. In agreement with this expectation the reducing gas sensor signal increased when oxygen was depleted, whereby the dilution rate is low until the rpm-DOT loop was closed and DOT approached the set-point around 2:00 h. The dilution rate as well as OUR and CER remained low during the scarcity of oxygen indicating slow growth. When the rpm-DOT loop was closed the agitation rate increased concurrently with the sharply decreased YRG signal. The latter occurred presumably because cells were able to switch directly to grow oxidatively, as indicated by the simultaneous peaks in OUR and CER. Subsequently the substrate feed rate increased and after 4.5 h the reducing gas sensor signal also started to increase and after some oscillations this signal remained around the desired set point. The oscillations were probably due to an overfeeding of substrate. The reducing gas signal was subsequently reduced by an increased capability to consume substrate oxidatively thus an increase in substrate feed rate could be resumed around 8 h to reach closely to the final steady state value after around 13 h. At which time OUR, CER and the two manipulated variables: D and rpm also reached close to their fmal values. The filter time constant was reduced to 3 min at 13 :55 h whereafter the 318

set point was maintained within I unit except when large disturbances were encountered. Start-up under adaptive control. The performance of the start-up procedure with the adaptive controller is seen in Figure 2a-c. The successful experiment with the adaptive controller was preceded by an unintended oversupply of glucose, which first had to be taken up by the cells. At time zero the rpm-DOT loop was closed and at 0: 10 h the D-YRG loop was closed. When ethanol was consumed the controller increased the dilution rate, but initially too much, so that the manipulated variable was decreased again to a low value at t=O:20-0:30. Most likely the loop was closed a little too early since the Y RG-signal had not yet reached a minimal value. Only around 0:30 h did the controller open significantly for the substrate inlet. Then, the cell density and oxidative metabolic activity increased as seen in Figure 2.b. The reducing gas sensor signal settled, if the disturbance at 2:00 h is ignored, at the target value after about 40 min corresponding to a closed-loop time constant of 10 min, while the other variables settled within 7 h. The estimated parameters of the 3rd order model structure and the variables of the controller changed very much initially when the model had not yet adapted to the process. Therefore the forgetting factor was often close to its lower limit as old information had to be forgotten fast when new information was coming in. The behavior of the controller was quite good. It must be noted that during adaptive control the manipulated flow rate showed signs of a ringing zero following disturbances e.g. at 15.7 h. These oscillations can be removed by a different control design which removes this discrete transfer function zero on the negative real axis from the closed loop transfer function (e.g. Zafiriou and Morari, 1989). This was achieved in the IMC-designed PI regulators as is evident by comparing the two . start-up figures . In addition, small perturbations of the set-point may be utilized in order to improve adaptivity during the start-up. Evaluation of the start-up experiments: The single adaptive control run indicated that a start-up process can be successfully carried out using a properly designed adaptive controller. This start-up procedure exhibited reasonable performance and provided a high cell yield on substrate, short operation time, and can be performed automatically. Thereby the possibility for avoiding a synchronously growing culture can be exploited. Obviously, the model identified at the target point did not describe the controlled system during the whole start-up procedure. Consequently the adaptive controller was superior to the constant gain IMC-PI-controller. An optimal start-up operation could require, that the yield on substrate is maximal and that the start-up time is minimal. Thus the pathway for oxidative growth should be exploited maximally during the start-up. Such an optimal start-up could be feasible if a suitably structured cell mass model was available. A single adaptive controller as applied in this work could only be optimal, if the model structure was reasonable, and if the adaptivity was fast enough to follow the cell adaptation during start-up. Reproducibility of DCTir The obtained dilution rate when operating the fermentor at YRG= 10 under steady state conditions is labelled Den,' This point is close to a

25

20

1 ;I~~

unit. (Oo-'OO)

j

15-V-RC

o

5

r,g

1 .0

'0

Ston-up of 0 cont,nuous fermentation (0161''') with Il.o1 C-PI-c:ontrolier (10." 0 mo ett'lonoIIl.V-7.9 ('I.

Fig

controller. O'OO6R. 10·"0 mo

C[R ( mmol/ I/min )

20

20

OUR ( mmo l/ I/ mrn)

.

1

2 .0 . Start-up of 0 continuous _e,-mentot

n w,t,.., OOOpt, .... V_ 7. 9 J

etl'l~ol ll.

CER (mmol/l/""~)

OUR

(mmol/I / ~)

CER

(- - - ) ~

1.5

",

.

:. .;.

eER (--- )

'r -'

.-_. . ... .' , , ' ~. I ,

-,

.,,,

.

-

~ -

OUR

0 .5 -:

0 .0

;,,.r.;

I

-.ro;.·~

__

o f'">q

~

5 l .b

___

~

1C'

15

20

20

25

RPw

'0

,5

20

2~

''9 2 .b 0")'9"" rote 0

Oryoe n uptOk~ rote onO coroon d.o.tOe' ..... olutlO r'I rot • . 0 19 ' , ,.,.

60 - I 00' ( 0 )

o

!

______.......,.____ nllE;....;,..,;....., ('1

r600

r

25 Fig. "c. o,'sotved Oll~n tenSIOn (OOT) and ogltat,on rot. (RPW ). 020068

0 ...0......0 OXYgel"l te"SI()I"I ( DOT ) and 09,tot'On rote (RPfrMI ). OHi111N

critical dilution rate where ethanol production is initiated. To pursue this objective data on D i.e. Dent' OUR, CER and RQ are shown in Figure 3a-c. It is clearly seen, that the dilution rate obtained in closed-loop with YJ.G.sp= I 0 (ca. 40 mg ethanol 1. 1) varied significantly. This variation indicated that the operating point is not uniquely defmed 319

by Dcrit1 or that steady state was not obtained. Postma et al. (1989) noted that in one case another 30 volume changes were required to reobtain steady state. The fermentor variables versus Dart in Figure 3a-c showed that low average values of OUR, CER and RQ at nearly the same low dilution rates were obtained in experiments

state with the lower metabolic activity to the steady state with the higher metabolic activity is perhaps initiated by an oversupply of glucose. The higher load on the glycolytic pathway will cause a higher load on the remaining oxiclative pathway, and if further capacity is made available in the pathway a higher specific growth rate, and a higher metabolic activity results. Thus, it appears that the organism adapted its metabolic activity after prolonged operation at low ethanol concentration.

1.35 ] OUR (mmol/I/min)

,

1.30 J

J -j

1.25

~ " ~

1.2G -

~

1.15

~

110~

4. CONCLUSIONS.

~

The performance of the adaptive controller was superior to the constant parameter model based PI-controller due to its ability to adapt to changes in cell metabolism. The start-up time, with this small volume, was not significantly affected by the initial cell concentration. All start-up experiments were performed without any sign of development of a synchronous culture. Thus, for c1osedloop start-up of a continuous fermentation a robust or adaptive controller should be used in order to avoid development of synchronous cultures by manipulating the dilution rate such that the ethanol level is low. A reproducible operating point could be obtained within 2030 h after initiation of the start-up procedure (-10 volume changes). Prolonged culture operation with initial substantial disturbances resulted in operating points with up to 10 % increased metabolic activity at a constant ethanol level presumably due to a higher yield of energy from glucose. Thus, the cells seemed to adapt to the actual ethanol level and perhaps also to the disturbances to which they were exposed over long time. Maintaining the ethanol level combined with DOT, pH and temperature seems nescessary but not sufficient to specify the metabolic state of the cells.

~

1.05 -j

~ :1

1.00

' "",

0.31

0.32

""

, .

0.33

"""I

0.34

[I-crit ( l / h)

t

iiii

I""''''

0 .35

j

,

0.36

Fig . 2 .a . Oxygen uptoke rate (OUR ) versus D-crit obtatn ed tn tne expe rt ments ,n table 3 .

1.60 ,CER ( mmol/I/ m ,n )

~

150 -i

.

j

1

1.40

~

~

j

:::J, 1 10 ~

i

Flq

.3

Ccr:l::,r OIQll',Oe evo l ution rote ( CER ) "'ef"SuS

0

[ ' -cr l!

ootCln e e

In

the

e xp er im en ts In taD le

3.

REFERENCES ~

15 -:: PO

,

1

1 <

"1

,

1 1:

<

1

1

~

08

<

; .00 ...,

0 96

~

rig . .3 .c.

[,·- cr it ( l /h )

;-I~'~"~'TT"TT"TT"~ ' .,.,....,,,~ , ~,,~,,~,"'".CT'''n-,,~ . .,..,.....,....",.;.,..... ",..,..,,.....,... , ,,..,..,,,..,..,,,..;.,,,..,..,,,,

0.3 1

0 .3:

R~5Djrotory

0.33

0 .34

0 .35

0.36

0.37

coefficient versus O-cnt

OOtOjn~d ' ,n tne experiments in toble

3.

on young cultures 20 to 30 h after a start-up following a batch phase. While significantly higher values were obtained in experiments on a much older culture. The increase in cell mass yield with increasing dilution rate could be explained by a falling cell C-content with increasing specific growth rate. The shift from the steady 320

Andersen, M.Y., S.B . Jm-gensen, N.H. Pedersen, H. Brabrand and L. Hallager (1995). Regulation of a continuous Yeast Fermentation near the Critical Dilution Rate using a Productostat. Submitted to Journal of Biotechnology. Axelsson, lP., C.F. Mandenius, O. Hoist, P. Hagander and B. Mattiasson (1988). Experience in using an ethanol sensor to control molasses feed-rates in Baker's yeast production. Bioprocess Engineering, 3, 1-9. Dairaku, K., E. lzumoto, H. Morikawa, S. Shioya and T. Takamatsu (1983). An Advanced Micro-Computer Coupled Control System in a Baker's Yeast Fed-batch Culture using a Tubing Method. Journal of Fermentation Technology, 61, no.2, 189-196. Jm-gensen, S.B., H.E. M"lIer and M. Y. Andersen (1992). Adaptive Control of Continuous Yeast Fermentation, near Critical Dilution rate. Conference paper at ICAFFT 5/ IFAC-BIO 2 March 29 - April 2, 1992. Postma, E., C. Verduyn, W.A. Scheffers and J.P. van Dijken (1989). Enzymic Analysis of the Crabtree Effect in Glucose-Limited Chemostat Cultures of Saccharomyces cerevisiae. Applied and Environmental Microbiology, SS, 468-477.