Development and Application of a Flexible Controller in Yeast Fermentations Using POsub2sub Cascade Control 2010 Engineering in Life Sciences Copia

Research Article

Development and application of a flexiblecontroller in yeast fermentations using pO2cascade control

The development and application of a flexible process controller in fed-batch yeastfermentations using pO2 cascade control was performed. A new algorithm for fed-batch fermentations using pO2 cascade control was developed, the concept ofwhich could be used as a realizable solution in fermentation systems equippedaccording to the basic configuration. The algorithm is based on the combinedinfluence of pO2 and pH on the substrate feeding intensity. To test and developthis algorithm, Saccharomyces cerevisiae DY 7221 and Candida tropicalis CK-4fermentations were carried out. As a result of the use of the combined algorithm,the specific growth rate and productivity grew in both fermentations. In this case,the effect of the use of the algorithm was most pronounced in the C. tropicalisfermentation.

Keywords: Bioreactor / Fed-batch / Fermentation / pO2 cascade control

Received: December 21, 2009; revised: May 31, 2010; accepted: June 15, 2010

DOI: 10.1002/elsc.200900112

1 Introduction

Different tasks of fermentation processes are determined to agreat extent by the potentialities of the process controller. Thetypical potentialities of fermentation control are commonlyincluded in the commercially available basic configuration of abioreactor controller. Hence, the control of the main para-meters, namely, temperature, pO2, pH, foam, level and over-pressure can be ensured, and the possibilities to controlfermentations using additionally the tools of the user interfaceand different protocols can be relatively broad. However, mostof the parameters of real value for the fermentation culturecannot be measured directly and controlled online. In thisconnection, it is important to gain as possible more infor-mation from fermentations not only with the help of speciallydeveloped experimental devices and systems, but also using theversions of sensors, executive elements and process controllersavailable in the market. In this connection, it is important tounderstand the real possibilities of measurement and processcontrollers in different conditions of fermentation processes.For example, it is well known that the application of sensors infermentations is limited. By analyzing the availability of asensor, it can be stated that relatively few additional probes

available in the market, besides the main parameters, are usedin some fermentations (dissolved carbon dioxide, redox andoptical density).

The basic measurements that can actually be performed infermenters are summarized by several authors [13]. Thelimitations of sensors selection in bioprocesses are connectedwith the fact that their properties must be kept after sterilization.Besides, also in sensors used, the materials are limited (stainlesssteel 316, PTFE, borosilicate glass and some special compounds).Due to this, the role of bioreactor control techniques increases.There are articles in which the review of fermenter controltechniques is summarized [2, 47]. It has been demonstratedthat the problem of the lack of available sensors can be solved byusing different approaches of advanced control [7]. Although theadvanced and highly sophisticated control has been muchpublished and is the subject of considerable discussion, theexperience shows that relatively simple control methods areapplied in most industrial-scale bioreactors [8].

Today, also by laboratory-scale fermentations, intermediateor high end control systems involving Programmable Logiccontrollers and the Human Machine Interface (SCADA) fromwell-known brand names are mainly used, because the stabilityand service of a control system is very important taking intoaccount the fact that a fermentation process without inter-ruption must function even many days. Brand name systems

Juris Vanags1,2

Viachaslau Hrynko3

Uldis Viesturs2

1Biotehniskais centrs, JSC,

Riga, Latvia

2Latvian State Institute of

Wood Chemistry, Riga,

Latvia

3Institute of Physical

Organic Chemistry,

National Academy of

Sciences, Republic of

Belarus, Minsk, Republic of

Belarus

Additional corresponding author: Uldis Viesturs

E-mail: [email protected]: PC, personal computer

Correspondence: Dr. Juris Vanags ([email protected]), JSC, Latvian State

Institute of Wood Chemistry, 27 Dzerbenes Street, LV-1006, Riga,

Latvia.

& 2010 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.els-journal.com

Eng. Life Sci. 2010, 10, No. 4, 321332 321

have a relatively wide selection of process control functions.However, if it is necessary to apply an advanced or specializedprocess control algorithm, the user self-develops a personalcomputer (PC) or Programmable Logic controllers-basedprocess control programs, and this software usually is notoffered as a product. attempts to develop a market availableflexible enough bioprocess controller have been demonstrated[9, 10]. A nonstandard algorithm using this controller and anupdated program downloaded in the controller using thee-mail and internet-based remote service of the manufacturerhave been shown [11].

The most frequently found more advanced type of controlare pO2 cascade controllers, in which dissolved oxygen iscascaded to the control loop for agitation, airflow, substratefeeding rate or some other parameters [4]. These options ofpO2 cascade control are available mostly in all brand namebioprocess controllers. However, for example, the control inthe pO2 cascade with substrate feeding very often needsadditional improvements in the algorithm to ensure theoptimal control. This aspect of control is important, because itis usually connected with the fed-batch mode. Generally, thefed-batch mode is most challenging for control. The simplestvariant of fed-batch control is entering the profile beforestarting the feed, and then correcting this profile during thefermentation based on offline glucose and other analyses. theexistence of limitations for the conventional control techniqueof fed-batch control is reported [12]. The challenge arisesmainly because the optimization of the feeding rate is adynamical problem [13]. different fed-batch models for betterpredicting of the profile behavior have been developed[14, 15]. The main task of fed-batch control is to keep theglucose concentration between the minimal and the maximallimits, and it is not always easy in the model to take intoaccount all the factors, which can influence thesubstrate consumption; hence, it is necessary to use alsoonline parameters. For example, adaptive control of theoverflow metabolite has been used by estimation of only oneparameter [16]. Different methods have been proposed bytracking the respiratory coefficient [17], as well as oxygenuptake rate [14, 18], ethanol concentration [19] and pHmeasurements [20]. An alternative method has been devel-oped, in which the substrate feeding rate is periodicallyreduced or stopped and then taken back at the desired value[21]. There are examples of the optimum feed profile in thefeedback form for ethanol fermentations [22]. The improvedproductivity has been obtained with feeding profiles that arecalculated online in a feedback loop [23]. For example, theoptimal control law has been developed, based on thecombination of the feed forward control action to followpredetermined set point trajectories [24]. There are also otherapplication examples of adaptive fed-batch algorithms[25, 26].

Summarizing the aforementioned and analyzing thecommercially available fed-batch realization potentialities inbioreactors, it can be concluded that users, without theadditional equipment, resources and training in respect to fed-batch, cannot ensure more than the execution of the time-dependent feeding profile and varying the feed rate dependingon pO2.

The aim of this study was to develop the approaches andalgorithms of microorganism cultivation processes, theconcept of which could be used as a realizable solution incontrollers of fermentation processes. The developed solutionsare oriented to the bioreactors equipped according to the basicconfiguration. Due to this, these will be potentially availablefor relatively wide circles of users. algorithms and approacheswere developed and tested in the fermentations using twostrains of yeasts (Saccharomyces cerevisiae DY 7221 andCandida tropicalis CK-4). These strains were selected as well-known and relatively fast-growing cultures of microorganisms.Taking into account the fact that the main goal in thesefermentations was connected with developing of algorithms, aminimal process off-analysis was carried out, e.g. only biomassand saccharose concentrations were determined.

2 Materials and methods

2.1 Bioprocess controller

In order to facilitate the aim of this study, a bioprocesscontroller Bio-3 was used. BIO-3 is a flexible controller, whichis easily customized for different fermentations and bio-reactors. A bioprocess controller BIO-3 was connected withsensors, executive elements of the bioreactor and with SCADAaccording to the process connection scheme shown in Fig. 1.

The controller is based on the following principles:

(i) Easy and demonstrative program environment in thecontroller.

(ii) Application of a separately programmable touch screendisplay.

For this purpose, the controller is based on a Basicprogrammable multitasking microprocessor. As multitasking, theproduct of Wilke Technology, GmbH BASIC TIGER was usedin the Basic programmable microprocessor [27]. This is amodule of a microcontroller, which has all necessary built-inhardware components to develop process control units. Animportant advantage of this module is programming in BASIC.Due to this, programming can be done relatively quickly, anddifferent customers can employ the same program.

Figure 1. Process connection diagram of the bioprocesscontroller BIO-3.

322 J. Vanags et al. Eng. Life Sci. 2010, 10, No. 4, 321332


A touch screen display was used in the controller to ensure aconvenient change of the existing virtual control buttonsconfiguration, and to create new control buttons. separateprogrammability is important to the display for the followingreasons: the control and communication functions program,which is in the controller, and the interface/visualization program,which is in a separate displays microprocessor, are separated. As aresult, the programs become visualized, easily processable and gaina higher degree of flexibility. For example, in this case, if changesare necessary in the displays interface functions, then no correc-tions are necessary in the controllers program.

The colored 5.600 touch screen display under the trademarkWeintek used in the controller is separately programmable,and this is realized by the information exchange with thecontroller via the MODBUS communication protocol. Thistouch screen display has a special environment, which makesprogramming and reprogramming easy and fast. programminglooks more like drawing screens.

In the experiments of this study, the following configura-tion of BIO-3 was used:

(i) The available control loops were: temperature, pH, pO2,foam and level.

(ii) The possibility to carry out time-dependent feedingprofiles (entering before the process and with thepossibility to correct those during the process).

(iii) Cascade control of pO2. The following cascades wereused:(a) stirrer rotational speed n,(b) oxygen enrichment O2,(c) substrate feeding QFeed.(d) Data exchange with the PC visualization program

(SCADA) using an industrial standard MODBUSprotocol. Bioprocess controller BIO-3 can be inter-faced practically with every SCAD using the OPCserver developed by us.

(e) Possibility to easily update the controller program bydownloading the latest version directly from the PC tothe controller through an RS 232 serial port.

In the basic configuration of BIO-3, there are two level accesspasswords. The first level access is for set-point limit adjustment,

whereas the second level access is necessary to change the PIDparameters and conduct the sensor calibration. In the advancedversion, there are three level accesses. The first one, e.g. thehighest level password access, allows the access to the next twoaccess password levels. The lowest, e.g. the third level access,enables to perform PID and limiting parameter adjustment, andinput/output scaling sensor calibration, whereas the second levelaccess additionally allows manual control operations, and thefirst level access also allows the process and system configuration.

2.2 Rules of pO2 cascade control

The cascade control functions in BIO-3 according to thefollowing rules:

(i) pO2 control is started with the first cascade. The processis controlled in the current cascade until the limits of thecontrolled elements are not achieved. If the limits areachieved, then the control is continued with the next orprevious cascade after expiring the cascade delay time.The transition direction (to the next or previous cascade)depends on the trend of the pO2 varying dynamics andthe limit achieved. The transition to the next cascade isnot possible, if the current cascade is the last one, andalso the transition to the previous cascade is not possibleif the current cascade is the first one.

(ii) The cascade can be paused or stopped. If the cascadepauses, the current control variable is frozen until theprocess continues again. If the cascade is stopped, thenthe pO2 control is also stopped. The control variablesreturn to the starting conditions, and the process isstarted from the first cascade.

(iii) The concrete cascade process starts with the defined valueof the corresponding controlled parameters. This valuecorresponds to one of the limits (those are definedaccording to Table 1). In the next cascade, the previouscontrolled parameter acts with the last limit value. Thisvalue of this parameter remains in all next cascades. Thecontrol parameters of all included cascades have startingvalues in every other cascade until the other limit of thisparameter is not achieved. If the other limit of the

Table 1. Cascade control conditions.

No. Cascade Controlled executive elements Starting value Transition value

to the next cascade

Remarks

1. Stirrer speed control n stirrer rotational speed n minimal limit value N maximal limit value

2. Oxygen enrichment O2 valve relative to the opening

impulse length (can be varied in

the range 0100%)

Minimal limit of the

O2 valve relative

to the opening

impulse length

Maximal limit of the O2valve relative to the

opening impulse length

3. Feeding control QFeed productivity of the

substrate feeding pump.

There are two variants,

e.g. with the direct and reverse

(used in this study) influence

of the substrate to pO2.

QFeed minimal

limit value

QFeed maximal

limit value

In every other cascade, the

feeding pump is turned

off (e.g. QFeed5 0)

Eng. Life Sci. 2010, 10, No. 4, 321332 Control of Fed-Batch Fermentations 323


controlled parameter is achieved, then this is the value ofthe controlled parameter in other cascades. The exceptionis pO2 control with feeding, e.g. the feeding pump inevery other cascade is turned off.

The cascades realized in this study have the controlledparameters as well as the starting and transition limited valuesas listed in Table 1.

2.3 Bioreactor

The above-mentioned possibilities of the controllerwith application examples of flexibility were demonstratedin fermentations using a laboratory bioreactor EDF-5.3_1 (manufactured by Biotehniskais centrs (BTC),JSC).

EDF-5.3_1 is a glass bioreactor with a working volume of 4 l(total volume 6.2 l). The bioreactor is mounted on stainlesssteel pipeline legs. thermoregulation is carried out by circula-tion of thermostated water trough a U-shaped tube, which isplaced in the bioreactor. In the lower part of the bioreactorvessel, there is a metallic bottom, where the sampling deviceand the connection for air sparger are placed. The bioreactorhas a novel magnetic drive, which is placed in the upper lid.The block diagram of the bioreactor for the followingexperiments is shown in Fig. 2.

It was used for the feeding digital controlled pump.different productivities were ensured with the lengths of pumpacting impulses. The period of the feeding pump in thefermentations carried out in this study was 30 s. The periodwas determined to be ensured with the precision at least 2%from the minimal used productivities.

To realize the oxygen enrichment, oxygen, O2, was intro-duced from the oxygen cylinder parallel to the air flow with thehelp of a digital controlled valve. The degree of oxygenenrichment was determined as the length of the oxygen valve-opening impulse. The period of the oxygen valve action was15 s.

2.4 Process visualization system (SCADA)

For process monitoring, the Basic version of EDF SCADA wasused. The Basic version of SCADA software is speciallydeveloped as a simple in application and easily installable tool.Data exchange between the controller and the PC serial port isrealized with the MODBUS protocol. The software providesthe following functions:

(i) data visualization in graphical and table forms (online orhistorical data),

(ii) archivation,(iii) indication of the status of the bioreactor elements,

Figure 2. Block diagramof the bioreactor EDF-5.3_1 for currentfermentations.



(iv) data export in ASCII for other applications (for example,Excel), offline data adding,

(v) remote monitoring possibility via the internet and(vi) simultaneous monitoring of up to eight bioreactors.

2.5 Materials and methods

Two strains of yeasts, i.e. S. cerevisiae DY 7221 and C. tropicalisCK-4, were used in the fermentations.

A pure culture of the yeast C. tropicalis CK-4, grown on asterile solid agarized medium with the use of a malt extract,with the content of dry matter up to 10%, was used. The strainof the yeast S. cerevisiae DY 7221 was used after the reactiva-tion from the lyophilized state.

The inoculum was grown in 750-mL Erlenmeyer flasks on asterile nutrient medium in a quantity of 100mL, containing amalt extract with the content of dry matter up to 10%. Theculture was incubated on a laboratory rocker with the shakingrate 100min1 and placed in a heat chamber at the tempera-ture 301N for 1218 h.

The rocked culture, under the flame of an alcohol torch,was reseeded into a laboratory fermentation unit. S. cerevisiaewas suspended with activated dry culture. For C. tropicalis, theinoculate was grown in laboratory rockers, and the concen-tration of biomass in the inoculate was limited by the poten-tialities of the shaker. Due to this reason, the seedconcentration of C. tropicalis was lower in comparison with thecase of S. cerevisiae. The nutrient media for S. cerevisiae DY7221 and C. tropicalis CK-4 fermentations were composedaccording to Table 2. The fed-batch media for S. cerevisiae DY7221 and C. tropicalis CK-4 fermentations were composedaccording to Table 3. The yeast suspension biomass was deter-mined using a photoelectrocolorimeter Thermo SpectronicHelios Epsilon at the wavelength l5 540nm, and the opticaldensity of the cell suspension was measured. then the yeastbiomass concentration was determined from the calibrationcurve. The concentration of saccharose was determined by theglucose oxidase method using a photometer after the inversion ofsaccharose to glucose. Determining the glucose concentration inthe molasses medium, before applying the glucose oxidasemethod, the sample was mixed in equal parts with 1M H2SO4,

and heated at a temperature of 801C for 10min. All parameters offermentations are shown in Figs. 36. Only temperatures are notshown in these figures because they were constant in allfermentations, i.e. 301C. The dead zone of temperature controlwas adjusted to 70.21C.

3 Results

3.1 Cultivation of S. cerevisiae DY7221 with fed-batchand cascade control of pO2

In all fermentations, the nutrient medium was preparedaccording to Table 2.

The initial concentration of dry biomass in all fermenta-tions was 4 g/L, and the initial concentration of saccharose was2 g/L.

A nutrient medium according to Table 3 was used forfeeding. The feeding was started from the second hour ofcultivation. The criterion for the onset of feeding was theevaluation of the residual concentration of saccharose, whosemeasurement dynamics depended on the lag-phase duration.True enough, the results of the analysis of saccharose could beobtained in about 30min. However, using the experience ofthe previous fermentations and knowing the growth dynamics,it was possible to extrapolate the change in the saccharoseconcentration. The condition for the onset of feeding waschosen, when the glucose concentration decreased below 1 g/L.

The feeding profiles were chosen so that the saccharoseconcentration would be kept within 0.1 g/LoNsacho2 g/L. Thechoice of the upper limit was determined from the consid-erations of the elimination of the beginning of the synthesis ofethanol, whereas the condition for the choice of the lower limitwas the elimination of the considerable decrease of growthrate. The feeding profiles were determined from the experienceof the previous fermentations. If necessary, correction of theprofiles was planned, based on the above-mentioned condi-tions (0.1oCo2.0 g/L), performing saccharose concentrationanalyses.

In the first fermentation (Fig. 3), PO2 was controlledaccording to the cascade control principle, with the successiveuse of the following three cascades with the control of:

(i) rotational speed of the stirrer n,(ii) degree of oxygen O2 enrichment and

Table 2. Nutrient media.

Component Value

Sugar-beet molasses with the saccharose

concentration 45% (g/L)

6.0

(NH4)2HPO4 (g/L) 0.85

K2HPO4 (g/L) 0.5

MgSO4 (g/L) 0.05

ZnSO4 (mg/L) 0.25

FeSO4 (mg/L) 0.25

MnSO4 (mg/L) 0.25

D-Desthiobiotin (g/L) 0.031

H2O (tap) up to 1.0L

pH H2SO4 (98% H2SO4 0.10.2mL) 4.24.4

Table 3. Fed-batch media.

Component Value

Sugar-beet molasses with the saccharose

concentration 45% (g/L)

350.0

(NH4)2HPO4 (g/L) 1.2

(NH4)2 SO4 (g/L) 14.5

ZnSO4 (mg/L) 1.0

FeSO4 (mg/L) 1.0

MnSO4 (mg/L) 1.0

D-Desthiobiotin (mg/L) 0.124

H2O (tap) (98% H2SO4 0.10.2mL) up to 1.0L



(iii) substrate feeding rate QFeed (with the effect of the reverseaction of the substrate on the PO2 value).

In the first two cascades, the feeding profiles were setaccording to the above-mentioned principles. Up to the sixthhour of fermentation, the feeding profiles were activated in theform of 30-min impulses with the same pause duration.This was connected with the fact that the precision of the usedperistaltic pumps was sufficient (2%), beginning with theproductivity 1mL/min. Upon the transition to thethird cascade, the set feeding profiles were automaticallyswitched off, and substrate feeding was performed at regularintervals, depending on the value and dynamics of the PO2change.

The following lower/upper limits of control were set in thePO2 control cascades:

(i) nmin5 50 rpmnmax5 350 rpm,

(ii) QO2min5 0%QO2max5 10% from the maximum O2 impulse duration,

(iii) QFeedmin5 0.2mL/minQFeedmax5 4mL/min

All automatic control modes were activated immediatelyafter starting the process.

The concentration of the dissolved oxygen PO2 was kept atthe level 15% from the concentration of the saturation, withthe dead zone range 75%. This choice was grounded by thefact that we conducted fermentations with the given strain ofS. cerevisiae at PO25 15 and 50%, and similar results on thebiomass growth dynamics were obtained.

The rotational speed of the stirrer n, already within40min after the onset of fermentation, reached 220 rpm; thenthe increase in the rotational speed occurred practically line-arly, with the approximate rate of the increase in the rotationalspeed of the stirrer Dn/Dt5 26 rpm/h. n reached theupper limit of control nmax5 350 rpm at the sixth hour offermentation. Then the process proceeded at the same rota-tional speed of the stirrer n, increasing the degree ofenrichment with oxygen O2. In this case, the PO2 fluctuationfrequency increased (Fig. 3). This is obviously connected withthe precision of the recurrence of the effect of the actionof the oxygen impulses. However, the amplitude of fluctua-tions was not above the dead zone limits. This means thatthe PO2 control process occurred with a sufficient precision,although the step of one oxygen impulse was 1%. O2 reachedthe upper limit of the cascade control, namely, 10% at theeighth hour of fermentation. Then the PO2 control with thechange in the feeding rate occurred. The given cascade beginsfrom the minimal limit of the feeding rate QFeedmin5 0.2mL/min. Such a feeding rate remained till the 14th hour of

Figure 3. Fed-batch fermentation of S. cerevisiae with three cascade pO2 control.



fermentation, because it was below the value, at which theincrease in the feeding rate QFeed must begin. According to thelaw of the control of the given cascade, PO2 must beginincreasing if PO2oPO2 (set point)1PO2 (dead zone). Figure 3shows that the increase of PO2 began at the 14th hour offermentation, and then the feeding rate also increased imme-diately. The given action of feeding has a notable action onPO2, and the feeding reached the minimal value again at the15th hour of fermentation. The analyses of saccharosedemonstrate that, from 10th to 14th hour of fermentation, itsconcentration was lowered (o0.3 g/L). This is confirmed alsoby the increase in pH, beginning approximately with the tenthhour of fermentation.

Analyzing the results of the given fermentation, we decidedto modify the control algorithm in the cascade of PO2 controlwith substrate feeding.

The main principles of the algorithm are as follows:

(i) The given algorithm is active only then, if the cascade ofPO2 control with substrate feeding is activated.

(ii) if pH4pH (set point)1pH (dead zone), then theincrease of the substrate feeding rate proceeds accordingto the PID algorithm, depending on the value anddynamics of the variation in pH. In this case, while thegiven condition is compiled with, the PO2 value isneglected.

(iii) if pHopH (set point)1pH (dead zone), then the changein the substrate feeding rate QFeed occurs only from thevalue and dynamics of the PO2 changes.

(iv) The coefficients of PID control in both regimes, namely,feeding control depending on PO2 and pH, respectively,were different, because the character of the action of thesubstrate feeding QFeed on PO2 and pH was different.

Based on this algorithm, the program in the commercialcontroller of bioprocesses BIO-3 was supplemented. Thesechanges and performing of their testing took 2 days.

The next fermentation with S. cerevisiae proceeded with thealready upgraded program of control (Fig. 4). In thisfermentation, the conditions were similar and had thefollowing distinctions from the previous one:

(i) Two cascades for the control of PO2 were employed,namely, with the rotational speed of the stirrer n and thesubstrate feeding rate QFeed

The following lower/upper limits of control were set:

(i) nmin5 50 rpmnmax5 430 rpm

(ii) QFeedmin5 0.4mL/minQFeedmax5 4mL/min

Figure 4. Fed-batch fermentation of S. cerevisiae with combined pO2 and pH control in the pO2 feeding cascade.



At the beginning of the process, QFeedmin was set to 0, and then itwas changed by 0.4 L/min at the 20th hour of fermentation.

The rotational speed of the stirrer approximately from the30th min to the 12th hour of fermentation grew practicallyevenly at the mean rate 16 rpm/h, and then the increase of therotational speed n started decreasing gradually.

A comparison of both the fermentations (Figs. 3 and 4)shows that the biomass growth occurred approximately simi-larly. In the second fermentation (Fig. 4), at the 19th hour offermentation, n reached the maximal set limit, namely,n5 430 rpm, and a transition to the second cascade of controlof PO2 occurred. At the beginning, QFeedmin was set as 0 (thislimit was changed by 0.4mL/min); therefore, within 20min,the feeding did not act. In this period of time, PO2 increaseddramatically up to 25 units, as a result of which QFeed increasedup to 1.2mL/min. Then PO2 decreased gradually, as a result ofwhich QFeed at the 22nd hour of fermentation decreased to theminimal value, i.e. QFeedmin5 0.4mL/min. Such a value ofQFeed was retained till the 28th hour, when the pH valuestarted increasing; then the increase of QFeed occurred,depending on the value and dynamics of the pH change. As aresult, within about 30min, the pH value was within thenecessary range. Almost simultaneously with the increase inpH, PO2 started increasing, but as shown in Fig. 4, this valuedid not exceed 17 units (namely, set point1dead zone),and if it were not the given amendment in the program, thesubstrate feeding rate QFeed would not grow. At the endof the fermentation, with increasing PO2, QFeed began toincrease.

Figure 5. Dynamics of the specific growth rate and productivity Pin S. cerevisiae fermentations. & Specific growth in the fermen-tation according to Fig. 3. & Specific growth in the fermenta-tion according to Fig. 4. J Productivity P in the fermentationaccording to Fig. 3. Productivity P in the fermentationaccording to Fig. 4.

Figure 6. Fed-batch fermentation of C. tropicalis with three cascade pO2 control.



In this fermentation, it was possible thereby to ensure themaintenance of the saccharose concentration in the range of0.41.2 g/L (except the 14th hour of fermentation, whereCglucose fell down to 0.3 g/L).

The fed-batch fermentation proceeded within 37 h, reach-ing the biomass concentration in terms of dry weight 37 g/L.As shown in Fig. 5, a tendency of decreasing growth ratepractically from the beginning of the fermentation was clearlyobserved. On the one hand, this can be connected with thedecrease in the concentration of sugars sources. the possiblefactor of the inhibition effect can be a reason for the decreasein the growth rate. It is also pertinent to note that the use ofthe molasses feeding medium, owing to the high viscosity ofmolasses, can give no more than 30% of sugar in the feedingmedium. Therefore, the cultural medium in the fermentationprocess is diluted, and the high concentration of the yeastcannot be achieved.

It is pertinent to speak here about the dilution of the yeastmass of the fermentation, because the initial volume was 2.0 L,with additional 1.5 L with the feeding, namely, a total of 3.5 Lof the cultural liquid per final concentration of the yeast.

An analysis of the dynamics of the specific growth rate andproductivity in both fermentations shows (Fig. 5) that boththe parameters in the fermentation with the pO2 and pHcombined pO2 control variant are higher by 10 and 15%,respectively, from the 6th to the 18th hour of fermentation.The course of specific growth rate m has a character of a

Figure 7. Fed-batch fermentation of C. tropicalis with combined pO2 and pH control in the pO2 feeding cascade.

Figure 8. Dynamics of the specific growth rate and productivity Pin C. tropicalis fermentations. & Specific growth in the fermen-tation according to Fig. 6. & Specific growth in the fermenta-tion according to Fig. 7. J Productivity P in the fermentationaccording to Fig. 6. Productivity P in the fermentationaccording to Fig. 7.



declining function. This is testified also by the fact that thegrowth of biomass is almost linear over the main fermentationtime (Figs. 3 and 4).

Regarding the yield of the conversion of saccharoseconsumption to biomass, no clear influence of the pO2control strategy was observed. The value of the yield during thefermentations varied in the range of 0.030.15, and it wasrather connected with other factors, for example, thesaccharose concentration in fermentations media (yield wasevaluated approximately as an instantaneous value).

3.2 Fed-batch cultivation of C. tropicalis CK-4 withcascade control of PO2

Similarly, fermentations using strains of the yeast C. tropicalisCK-4 were carried out (Figs. 6 and 7).

The inoculation and preparation of the inoculum occurredaccording to the method described in this article. The initialfermentation of biomass was about 1.2 g/L. The initialconcentration of saccharose was 3.8 g/L. A cycle of fermenta-tion with the given strain was carried out, and two fermenta-tions are reflected in this article. In both these fermentations,PO2 was controlled according to the cascade control principleusing two cascades, namely, with the control of the stirrerrotational speed n and with substrate feeding QFeed. In the firstfermentation, the feeding rate in the cascade was controlledonly from PO2, and in the second fermentation from pH incombination with PO2, according to the modified algorithmdescribed above.

Although the first cascade of PO2 control was acting, theprinciple of the determination and input of the profiles wassimilar to that in the fermentations of S. cerevisiae of the givenpaper.

The following lower/upper limits of control were set in thePO2 control cascade:

(i) nmin5 50 rpm,nmax5 600 rpm.

(ii) In the first fermentation (Fig. 6), Feedmin was firstly set as0, and then, at the 15th hour of fermentation, it was resetto 0.2 L/min.

In the second fermentation (Fig. 7), Feedmin5 0.8 L/min.Feedmax5 5 L/min in both fermentations.

In both fermentations, PO2 was set at the level 20% fromsaturation with the dead zone 75%.

In the fermentations, the exponential growth with theapproximate growth rate m5 0.2 h1 was observed till the 14thhour of fermentation, then the growth rate started decreasinggradually. Such a growth dynamics is explained to a greatextent by the decrease in the saccharose concentration andpossible inhibition factors.

It was observed in both fermentations that the rotationalspeed of the stirrer n increased up to 185 rpm within about30min. Then the increase of n was slow (approximately till thefourth hour of fermentation), and then the growth rate of nbegan to increase again, retaining such a tendency till the 12thhour of fermentation, when nmax was reached, and the transi-tion to the substrate feeding cascade occurred, depending on

the value and dynamics of the PO2 change. In the firstfermentation, in the initial region of the given cascade,QFeed5 0, and pH and then also PO2 began to increase again.The saccharose concentration also testifies the lowered value ofsugar sources. The increase in the feeding rate began onlyin 45min, because the condition: PO24PO2 (set point)1PO2(1dead zone) was met by that time. Then PO2 starteddecreasing; hence, QFeed decreased automatically, reachingQFeedmin. Already in about 30min, PO2 started increasing,thereby, QFeed increased up to 1.8 L/min. Then PO2 startedstabilizing around the set value. However, beginning with the18th hour of fermentation, pH began to increase, i.e. to alkalize.The increase in the feeding rate decreased the pH, but becausePO2oPO2 (set point)1PO2 (1dead zone), the further increasein the feeding rate did not occur. The analyses of saccharosealso confirm that, beginning with the 18th hour of fermenta-tion, its concentration was below 1 g/L.

It can be also concluded from this fermentation that, first, itis not enough to control the feeding only depending on PO2; itis also necessary to take into account the pH change dynamics;second, in the cascade of PO2 control with feeding, it isdesirable to set a higher QFeedmin limit.

The distinction of the next fermentation was that thecontrol in the feeding cascade was realized according to thecombined algorithm described above (pH1PO2).

As a result, it was possible to eliminate the increase of pH aswell as the decrease in the saccharose concentration till the 28thhour of fermentation. Then, although QFeed constantly increased,it was not possible to eliminate the increase of pH. It should bementioned that the corresponding increase of PO2 was notobserved. At the end of the fermentation, QFeed reached the setmaximal limit QFeedmax and, according to the cascade controllaw, the transition to the previous cascade (namely, with controlof n) occurred, and the feeding rate was switched off. Continuingthe fermentation, it was possible to conduct the feeding with theinput of the corresponding profile. However, if the target wouldbe to continue the fermentation, then it would be wore expedientto increase Feedmax.

The analysis of the dynamics of the specific growth rate andproductivity in both fermentations shows (Fig. 8) that both theparameters in the fermentation with the pO2 and pH combinedpO2 control variant are higher by 18 and 22%, respectively, fromthe 12th to the 23rd hour of fermentation. It should bementioned that the growth of biomass has an exponentialcharacter up to the 15th17th hour of fermentation.

Regarding the yield on saccharose consumption to biomass,no clear influence of the pO2 control strategy was observed.The value of the yield during the fermentations varied in therange of 0.080.2, and it was rather connected with otherfactors, for example, the saccharose concentration in fermen-tations media (yield was evaluated approximately as aninstantaneous value).

4 Discussion

As can be seen in this study, employing the bioreactorcontrollers flexibility, the relatively prompt and demonstrativevisualization of fermentations in the process control program



was possible. Employing this approach, there are potentialitiesof the further upgrading of process control. For example, ascan be seen from the presented fermentations with a modifiedalgorithm of feeding control, depending on PO2 and pH, it ispossible to partially eliminate the decline in the concentrationof sugar sources, in comparison with the case, when feeding iscontrolled only depending on PO2. This also promoted theincrease in the specific growth rate and productivity. Theinfluence of the modified algorithm was not observed.However, employing the modified algorithm, it was notpossible to eliminate fully the decline in the glucose concen-tration below the set threshold. in this case, however, variantsof the improvement of the given algorithm manifest them-selves in this aspect, namely:

(i) Correct choice of QFeedmin and QFeedmax.(ii) The optimization of the PID coefficients, selecting their

values in such a way that the action of the substrate on thevalues of PO2 and pH would be less sensitive and inertthan in the conducted experiments.

(iii) possible improvement of the algorithm, envisaging theincrease in the rotational speed of the stirrer n and thedegree of enrichment with oxygen O2 during the action ofthe substrate feeding control depending on the values ofPO2 or qI.

During the feeding, the biomass grows; in this case, toimprove the conditions of mass exchange, it is necessary toincrease the rate of oxygen supply to the cells of micro-organisms. Therefore, in the cascade of the PO2 and/or pHcontrol with feeding, it is possibly necessary to increase n orO2.

Considering the dynamics of pO2, changes in all thefermentation processes analyzed in this study (Figs. 3, 4, 6 and7), it can be seen at the first moment that their retention in therange pO2 (set point 7 dead zone) is not sufficientlysuccessful. However, each pO2 deviation beyond this range hasits substantiation.

(i) in the first fermentation of S. cerevisiae (Fig. 3), pO2 isbelow the standard from the 10th to the 14th and from the15th to the 18th hour of fermentation, because the pO2control parameter substrate feeding rate QFeed has aminimal limit (i.e. QFeed5QFeedmin). In this fermentation,from the 18th to 20th hour and from the 22nd to the 23rdhour, pO2 fluctuates around the higher limit of pO2(20%).

(ii) The second fermentation of S. cerevisiae (Fig. 4) from the25th to the 28th hour and at the 32nd hour is below thestandard, because QFeedmin is reached. In its turn, after the35th hour of fermentation, pO2 begins to grow consider-ably, hence, also QFeed grows. Although the QFeed growthrate cannot keep pace with the pO2 growth (because thePID coefficients are set based on the relatively slowgrowth), pO2 from the 35th to the 37th hour is beyond theupper limit range.

(iii) In the first fermentation of C. tropicalis (Fig. 6), pO2 isbelow the standard range by the 18th hour of fermenta-

tion, because then QFeedmin is reached. For the relativelyshort instant, pO2 is above the standard range (by the14th, 15th, 18th and 23rd hour), because the changes ofthe feeding rate QFeed do not keep pace with the pO2changes relatively fast.

(iv) In the second fermentation of C. tropicalis (Fig. 7), pO2 isbelow the standard range from the 15th to the 21st hour,because then QFeedmin is reached. For a short time, pO2 isbeyond the standard range (by the 25th hour), becauseQFeed cannot keep pace with the pO2 changes. After the29th hour of fermentation, pO2 is below the standardrange, because, as the pH in this period of time is abovethe standard range, the QFeed regulation occurs based onthe pH variation dynamics.

Thus, it has been demonstrated in this study that, despite thepresence of the relatively numerous different algorithms ofcontrol with feeding, using a flexible controller, it is possible tofind new methods for control of fed-batch fermentations. Theflexible controller enables an easier realization of the ideas basedon the fermentation observations in the process control program.The present solutions to the control of fermentation processes arepotentially available for wide circles of users, because they can berelatively easily implemented in the commercially availablecontroller of bioprocesses. Employing this approach, it is possibleto make also other improvements in the bioreactor processcontrol programs, because the possibility of the prompt realiza-tion of technological ideas in the program enables to reachevolutionary the optimum, even currently unknown solution.

5 Concluding remarks

(i) The control of the dissolved oxygen PO2 with differentcascades extends the possibilities of optimizing fed-batchprocesses.

(ii) By feeding control, using the combined algorithm of theeffect of the PO2 and pH values, it was possible to avoidto a great extent the decline in the concentration of thesources of sugars below the critical threshold for allstrains used. As a result, also the specific growth rate andproductivity increased. In this case, in the C. tropicalisfermentation, the effect of the algorithm on the indices ofthe microorganism growth was higher.

(iii) Despite the existence of the relatively numerous differentalgorithms of control with feeding, it is possible to find newmethods for control of fed-batch fermentations using theflexibility of the developed controller. The flexible controllerenables an easier realization of the ideas based on thefermentation observations in the process control program.

Acknowledgements

This work was supported by the FP7 project (WOOD-NET)under the agreement 203459 and by the Latvian Council ofSciences Project 09-1177.



Conflict of interest

The authors have declared no conflict of interest.

References

[1] M. Ghoul, E. Ronat, J. M. Engasser, An automatic and

sterilizable sampler for laboratory fermentors: application to

the on-line control of glucose concentration. Biotechnol.

Bioeng. 1986, 28, 119121.

[2] K. Sch .ugerl, Progress in monitoring, modeling and control

of bioprocesses during the last 20 years. J. Biotechnol. 2001,

85, 149173.

[3] C. F. Mandenius, Recent developments in the monitoring,

modelling and control of biological production systems.

Bioprocess Biosyst. Eng. 2004, 26, 347351.

[4] J. S. Alford, Bioprocess control: advances and challenges.

Comp. Chem. Eng. 2006, 30, 14641475.

[5] Y. K. Rani, V. S. R. Rao, Control of fermenters a review.

Bioprocess Biosyst. Eng. 1999, 21, 7788.

[6] J. Lee, S. Y. Lee, S. Park, A. P. J. Middelberg, Control of

fedbatch fermentations. Biotechnol. Adv. 1999, 17, 2948.

[7] S. J. Qin, T. A. Badgwell, An overview of industrial model

predictive control technology. AIChE Symp. Ser. 1997, 93,

232.

[8] S. Gnoth, M. Jenzsch, R. Simutis, A. Luebert, Control of

cultivation processes for recombinant protein production: a

review. Bioprocess Biosyst. Eng. 2008, 31, 2139.

[9] T. Pencheva, M. Petrov, M. Ilkova, O. Roeva et al., in U.

Viesturs, S. Tzonkov (Eds.), Bioprocess Engineering, Avangard

Prima Sofia, Bulgaria 2006, p. 255.

[10] J. Vanags, U. Viesturs, M. Vishkins, Development of a flexible

and user-friendly system for fermentation processes control.

J. Biotechnol. 2007, 131, 144.

[11] J. Vanags, M. Rychtera, S. Ferzik, M. Vishkins, U. Viesturs,

Oxygen and temperature control during the cultivation of

microorganisms using substrate feeding. Eng. Life Sci. 2007,

7, 247252.

[12] R. Berber, Control of batch reactors: a review. Trans. Inst.

Chem. Eng. 1996, 74, 320.

[13] A. Johnson, The control of fed-batch fermentation processes.

Automatica 1987, 23, 691705.

[14] M. A. H. Hisbollah, K. B. Ramachandran, Comparative

evaluation of various control schemes for fed-batch

fermentation. Bioprocess Biosyst. Eng. 2002, 24, 309318.

[15] M. Jenzsch, S. Gnoth, M. Beck, M. Kleinschmidt et al.,

Open loop control of the biomass concentration within the

growth phase of recombinant protein production processes.

J. Biotechnol. 2006, 127, 8494.

[16] S. Valentinotti, B. Srinivasan, U. Holmberg, D. Bonvin et al.,

Optimal operation of fed-batch fermentations via adaptive

control of overflow metabolite. Control Eng. Pract. 2003, 22,

665674.

[17] C. Bideaux, S. Alfenore, X. Cameleyre, C. Molina-Jouve et al.,

Minimization of glycerol production during the high-

performance fed-batch ethanolic fermentation process in

Saccharomyces cerevisiae, using a metabolic model as a

prediction tool. Appl. Environ. Microbiol. 2006, 72,

21342140.

[18] D. Levisauskas, R. Simutis, D. Borvitz, A. Lubbert, Automatic

control of the specific growth rate in fed-batch cultivation

processes based on an exhaust gas analysis. Bioprocess Biosyst.

Eng. 1996, 15, 145150.

[19] Ch. Cannizzaro, S. Valentinotti, U. von Stockar, Control of

yeast fed-batch process through regulation of extracellular

ethanol concentration, Bioprocess Biosyst. Eng. 2004, 26,

377383.

[20] K. Y. San, G. Stephanopoulus, Studies on on-line bioreactor

identification. IV. Utilization of pH measurements for

product estimation. Biotechnol. Bioeng. 1984, 26, 12091218.

[21] V. S. Whiffin, M. J. Cooney, R. Cord-Ruwisch, On-line

detection of feed demand in high cell density cultures of

Escherichia coli by measurement of changes in dissolved

oxygen transients in complex media. Biotechnol. Bioeng.

2004, 85, 422433.

[22] J. M. Modak, H. C. Lim, Feedback optimisation of fed-batch

fermentation. Biotechnol. Bioeng. 1987, 30, 528540.

[23] J. D. Boskovic, Novel feeding strategy for adaptive control of

bioreactor processes. J. Proc. Contr. 1997, 7, 209217.

[24] S. Park, W. F. Ramirez, Optimal regulatory control of bio-

reactor nutrient concentration incorporating system identi-

fication. Chem. Eng. Sci. 1990, 45, 34673481.

[25] J. Staniskis, D. Levisauskas, An adaptive control algorithm

for fed-batch culture. Biotechnol. Bioeng. 1984, 26, 419425.

[26] N. Besli, G. Ensar, M. T .urker, Application of Fuzzy Control to

Fed-Batch Yeast Fermentation, Computational Intelligence

Theory and Applications, Springer, Berlin/Heidelberg 1997,

1226, pp. 207216.

[27] P. Bernd vom Berg, Groppe Pack den Tiger in die Schaltung.

In: Messen Steuern und Regeln mit dem BASIC-TIGER

SYSTEM, Franzis Verlag GmbH 1999, p. 423.



Documents

Development and Application of a Flexible Controller in Yeast Fermentations Using POsub2sub Cascade Control 2010 Engineering in Life Sciences Copia