On-line monitoring and controlling system for fermentation processes

Biochemical Engineering Journal 7 (2001) 17–25

On-line monitoring and controlling system for fermentation processes

Yen-Chun Liu, Feng-Sheng Wang∗, Wen-Chien LeeDepartment of Chemical Engineering, National Chung Cheng University, Chia-Yi 62107, Taiwan

Received 10 January 2000; accepted 2 September 2000

Abstract

A personal computer-based on-line monitoring and controlling system was developed for the fermentation of microorganism. Theon-line HPLC system for the analysis of glucose and ethanol in the fermentation broth was connected to the fermenter via an auto-samplingequipment, which could perform the pipetting, filtration and dilution of the sample and final injection onto the HPLC through automationbased on a programmed procedure. The A/D and D/A interfaces were equipped in order to process the signals from electrodes and from thedetector of HPLC, and to direct the feed pumps, the motor of stirrer and gas flow-rate controller. The software that supervised the controlof the stirring speed, gas flow-rate, pH value, feed flow-rate of medium, and the on-line measurement of glucose and ethanol concentrationwas programmed by using Microsoft Visual Basic under Microsoft Windows. The signal for chromatographic peaks from on-line HPLCwas well captured and processed using an RC filter and a smoothing algorithm. This monitoring and control system was demonstrated tobe effective in the ethanol fermentation ofZymomonas mobilisoperated in both batch and fed-batch modes. In addition to substrate andproduct concentrations determined by on-line HPLC, the biomass concentration inZ. mobilisfermentation could also be on-line estimatedby using the pH control and an implemented software sensor. The substrate concentration profile in the fed-back fermentation followedwell the set point profile due to the fed-back action of feed flow-rate control. © 2001 Elsevier Science B.V. All rights reserved.

Keywords:Bioprocess monitoring; Control; Ethanol; Fermentation; Sensors; Chromatography

1. Introduction

The development of a fully automated on-line monitoringand controlling system is very important in fermentationprocesses. So far the automatically measurable parametersin the commercial fermenter are pH, temperature, aeration,agitation and dissolved oxygen. To monitor automaticallysubstrate and product concentrations during fermentation isusually difficult. The use of in-line biosensors in a fermenterhas some limitations because of the difficulty of preventingsensor fouling and offering in situ sterilization [1]. In com-parison with biosensors, the HPLC system is advantage ofanalyzing simultaneously many components in the fermen-tation broth. For monitoring the product and by-productsof penicillin V [2] and cephalosporin C fermentations [3],on-line HPLC systems have been developed. These sys-tems involve the use of a homemade or laboratory-builtsampling device connecting the fermenter and the HPLC.An automated, HPLC system for monitoring the productof monoclonal antibody in a hybridoma cell culture hasalso been developed [4]. Other HPLC-based automaticon-line monitoring systems have also been established in

∗ Corresponding author. Tel.:+886-5-2428153; fax:+886-5-2721206.E-mail address:[email protected] (F.-S. Wang).

the fermentation of recombinantE. coli for the productionof engineered proteins [5,6]. In the system proposed byTurner et al. [6], a steam sterilizable sampling device anda high speed microcentrifuge are installed before HPLC.The whole broth sample is pretreated with centrifugation toobtain a clear supernatant that then is injected to an HPLC.

The linkage of the on-line HPLC to a fermenter or biore-actor for controlling purpose is rarely reported. Most ofthe studies in the literature on the HPLC-based automaticsystem focus only on the on-line monitoring. Turner et al.[6] have recently applied the on-line HPLC system for theclose-loop control (feed-rate control) of a fed-batch fer-mentation of recombinantE. coli. In the present work, wedeveloped a low cost, PC-based monitoring and control-ling system for the fermentation of microorganisms. Thesoftware was implemented on Microsoft Visual Basic 4.0.The data acquisition, analysis and fed-back control of manykey parameters could take place simultaneously under thesupervision of this system.

The ethanol fermentation ofZymomonas mobiliswastaken as the model for the application of this on-line sys-tem. Z. mobilis is a Gram-negative bacterium capable ofproducing almost a theoretical amount of ethanol fromglucose via an Entner–Doudoroff pathway under anaerobicconditions. When compared with commonly used yeasts,

1369-703X/01/$ – see front matter © 2001 Elsevier Science B.V. All rights reserved.PII: S1369-703X(00)00100-5

18 Y.-C. Liu et al. / Biochemical Engineering Journal 7 (2001) 17–25

it has many advantages including a faster specific rate ofethanol production, improved yield and enhanced ethanoltolerance [7–9]. Z. mobilis can be used for the industrialproduction of fuel ethanol by fermentation [10–12]. Thegenetic improvement of industrial strains has already beeninvestigated, using eitherZ. mobilisas the host strain [13]or genes encoding alcohol-producing enzymes inZ. mobilis[14]. The automatic control of ethanol fermentation pro-cesses at their optimal states is of considerable interest sinceit can reduce operating costs and improve productivity.

2. Material and methods

2.1. Apparatus and instrumentation

A homemade fermenter with a total working volume of 5 lwas used. The temperature of the fermenter was maintainedat a set temperature by controlling the circulation of coolingwater. The vessel was equipped with an Ingold probe forpH measurement. Fig. 1 shows a schematic diagram of thefermenter, along with peripheral instruments for the on-linemonitoring and control. An auto-sampling device (Bench-Mate II) from Zymark Corporation was connected to thefermenter and HPLC. The substrate and product (glucoseand ethanol) in the fermentation broth were on-line analyzedby an HPLC (Waters 600) with Bio-Rad Aminex HPX-87C

Fig. 1. On-line monitoring and control system includes an auto-sampling equipment, HPLC, personal computer and control components.

column and a refractive index detector (Waters 410). Thecolumn temperature was kept at 85◦C, and the mobile phasewas the deionized water, which was applied at a flow-rateof 0.6 ml/min. The temperature in the detector was 40◦C.

2.2. Data acquisition, monitoring and controlling system

Output analog signals from the thermocouple, pH andDO electrodes were collected and converted to digital sig-nals by the interface cards, PCL-816 and PCL-726 (bothobtained from Advantech, Taipei, Taiwan), prior to transfer-ring to the personal computer. The analog signals (voltages)from the RI detector of HPLC system were also transferredto the PCL-816, through an RC filter with a capacitance of0.01mF and a resistance of 100� placed before the inter-face card. These voltage signals for chromatographic peakswere smoothed using a software programmed by followingthe method proposed by Savitzky and Golay [15]. To obtainthe HPLC results, an algorithm for the identification and in-tegration of the chromatographic peaks was also developed.Displaying of the chromatogram and HPLC results on thescreen of Microsoft Windows 95 was programmed usingMicrosoft Visual Basic 4.0.

Fig. 2 represents the software structure of the personalcomputer-based monitoring and control system. The masterpanel of the monitoring and controlling system displaysthe options of several settings as shown in Fig. 3. In the

Y.-C. Liu et al. / Biochemical Engineering Journal 7 (2001) 17–25 19

Fig. 2. The software structure of the monitoring and control system.

sub-screens of these options, one could set the controllingpolicies of pH, feed flow-rate, gas flow-rate, stirring speed,time-course profile of substrate (glucose) concentration,

Fig. 3. The master panel for several settings and the sub-panel for feed flow-rate control.

programming of HPLC and starting time of fermentation.For example, a sub-panel for setting the feed flow-rate con-trol is also shown in Fig. 3. The software for each controlwas programmed using Microsoft Visual Basic. The digitaloutputs from the control programs were converted to ana-log signals by the PCL-726 interface card in order to directthe top driven motor, gas flow-rate controller and peristalticpumps (Masterflex, Cole-Parmer) for the addition of feed,acid and base.

2.3. Ethanol fermentation

The strainZ. mobilisATCC 10988 was cultured on a glu-cose medium in a homemade 5-l fermenter. The fermenta-tion medium contained 2.0 g/l KH2PO4, 1.5 g/l (NH4)2SO4,1.0 g/l MgSO4 and 5.0 g/l yeast extract. The concentrationof glucose for batch culture was 100 g/l, and that in feed so-lution for fed-batch culture was 150 g/l. The composition ofthe fed solution was the same as the fermentation mediumexcept for (NH4)2SO4 concentration (2.25 g/l).

Prior to fermentation, the pH electrode (Inglod Infit764-50) and peristaltic pumps were calibrated on the pan-els of PC control software. The ethanol fermentation was


operated in both batch and fed-batch modes. The fermenta-tion broth was automatically sampled at each time intervals.The cells in the sampled solution were retained by a 0.5mmsyringe filter. The filtrate was diluted and then automaticallyinjected into HPLC for the analysis of glucose and ethanolconcentrations.

2.4. Control strategy for ethanol fermentation

For batch fermentation, a 10% (v/v) inoculum was addedto the fermenter. Initially, the pH of the fermentation brothwas adjusted to be 5. The stirred speed was set at 120 rpm.The purpose of batch runs was to test the on–off control ofpH. The control procedure was as follows: (1) measuringthe output voltage from pH detector and converting it tobe a pH value; (2) comparing this pH value with the setpoint and calculating the pH difference (1pH, i.e. the setpoint subtracted from this pH value); (3) adding 1 ml HCl(1pH > 0.05) or NaOH(1pH <−0.05), or doing nothing(−0.05 < 1pH < 0.05); (4) repeating steps from 1 to 3 forevery 10 s.

The fed-batch fermentation was not only to test the on–offcontrol of pH, but also to test the on-line fed-back controlof glucose concentration in the fermenter via the change inthe feed flow-rate. A 33% (v/v) inoculum was added to thefermenter with a total initial broth volume of 1.5 l. The pHof the fed-batch fermentation was set at 5. The policy offeed flow-rate change during the fermentation course couldbe set at the sub-panel of “feed flow-rate” prior to fermen-tation. If fed-back control of glucose concentration was per-formed, the feed flow-rate was changed according to thedifference between the glucose concentration determined byon-line HPLC and the concentration set on the sub-panel of“glucose concentration profile”. In this study we used theoutputs from model prediction of Rivera [16] under the con-stant feed flow-rate conditions as the setting values of thetime-course profile of substrate concentration. During thecourse of fed-batch fermentation, the feed flow-rate wouldchange in response to the glucose concentration differencebetween the set profile value and the on-line measured.

3. Results and discussion

3.1. Control functions

A linear relationship between the stirred speed (rpm) andthe voltage (V) applied for the motor, directed by the per-sonal computer, was obtained. One could set the stirredspeed by adjusting the delivery of voltage from PC to themotor. Based on this calibration curve, the motor could beoperated with a voltage ranging from 1.7 to 6.2 V in orderto yield a stirred speed of 100–650 rpm. For the aeration ofthe fermenter by using either sterile air or nitrogen through athree-way valve, the flow-rate of gas stream was controlled

by a flow-rate controller, also under the supervision of PC.The pH and feed flow-rate were controlled via the use ofperistaltic pumps. One pump was used for feeding the sub-strate solution and two pumps were employed for the deliv-ery of NaOH and HCl, respectively, for controlling the pHof the fermentation broth. Calibration curves for flow-rate(ml/min) vs. voltage (V) delivery to the peristaltic pump(Masterflex® Model 7518-10) by PC were also obtained. Forexample, when a tube of Masterflex® 9600-14 was used, theflow-rate (denoted byX) was related to the delivered voltage(denoted byY) by Y = 0.419X + 0.030 (r2 = 0.99); whena small tube Materflex® 9600-13 was used, the calibrationwasY = 1.887X − 0.005 (r2 = 0.99). For these two cali-brations, the range of the voltage was from 0 to 10 V. All theabove-mentioned controlling elements were effective underthe supervision of the PC-based controlling program.

To control the feed flow-rate, one could set the feedpolicy at the sub-panel of “feed flow-rate” as shown inFig. 3. The fermentation time was sub-divided into 20time intervals for specifying the changes of the feedflow-rate. The optimal feed policy resulted from modelsimulation and optimization could be taken as the settingfeed policy. However, the feed flow-rate was automati-cally subject to change by the software controlling sys-tem, if the option of “fed-back control” on this sub-panelwas chosen. In this instance, the feed flow-rate waschanged according to the difference between the glucoseconcentration determined by the on-line HPLC and thatset on the sub-panel of “glucose concentration profile”.The change of feed flow-rate was based on the follow-ing mass balance of glucose:(set glucose concentration−measured glucose concentration) × fermentation volume=increase in flow-rate× sampling time interval× feed glucoseconcentration.

3.2. On-line chromatographic system

At the beginning of each sampling, the fermentation brothwas sampled by BenchMate II and the cells in the samplesolution were retained by a filter. The filtrate was then di-luted automatically and injected into the HPLC. Wheneverthe HPLC program in PC received the signal of injection,the output signal from RI detector was automatically cap-tured and converted to digital signal, displayed on the panel.After the chromatographic process was concluded, the peakpositions were identified and the peak areas for each peakwere calculated. The concentrations of glucose and ethanolwere then estimated by comparing the peak areas with thecalibration curves and displayed on the panel. A file as-signed for each analytic run was automatically generatedfor saving the chromatographic results. The files for chro-matographic results could be reopened for displaying thechromatogram and peak data on the screen of PC (Fig. 4).

A smoothing process by combining a hardware and asoftware method is implemented in this work to obtain the


Fig. 4. Monitor-display for the chromatogram (a) and HPLC analysis results (b).

chromatographic signal from RI detector. The chromatogramas shown in Fig. 5a was directly acquired from the originalsignal of RI detector. Firstly, the noise was eliminated fromthe original analog signal of RI detector by using an RCfilter. For the present data acquisition system, Waters 410RI detector connecting to PCL-816 A/D interface, we foundthe best values assigned for the resistance and capacitancewereR = 100� andC = 0.01mF, respectively. Since theresolution of the A/D interface was 16 bits, the measurablelimit (sensitivity) of the input signal was 0.0152 mV. Asshown in Fig. 5b, most of the noises were eliminated fromthe chromatogram as shown in Fig. 5a. A smoothing methodproposed by Savitzky and Golay [15] was then used tomodify the digit signal. According to this method, am-pointsmoothing means that one takes signals ofm-points eachbefore and after the specific data point for curve fitting in or-der to yield a smoothing curve passing this data point. If noRC filter was installed, a 9-point smoothing could result insmooth chromatographic peaks as shown in Fig. 5c, but thenoise was still present. It was found that an RC filter plus a9-point smoothing could yield a satisfied chromatogram asshown in Fig. 5d. However, a comparison of the smoothedchromatogram with the original one, we found that the peak

position shifted slightly. The chromatographic results werenot influenced by this shift in peak positions.

It was important to identify the start and end points, aswell as the height and retention time of each peak in thechromatogram. When the chromatographic signal was pro-cessed, the first and second derivatives of the chromatogramat each data point were also estimated simultaneously. Asmall value was assigned for a threshold of zero slope, calledthe slope sensitivity (S). Whenever the first derivative at oneof the data points was within the value of±S, it was sug-gested that this point could be the start, end or highest pointof a chromatographic peak. This point was the starting timeof a peak when the first derivative was within±S and thesecond derivative was positive. After the start point was iden-tified, the first derivative of the chromatographic peak waskept being watched in order to look for the end point. As thefirst derivative approach to zero from the negative side andthe second derivative was positive again, the end point of thechromatographic peak was asserted. The highest point wasthen located by choosing the highest signal value betweenthe beginning and end points. The baseline for this peakcould be the straight line connecting the peak values of startand end points. The peak area was then obtained as the prod-


Fig. 5. Chromatographic peaks of glucose (5 g/l): (a) with no filter and no smoothing; (b) with an RC filter; (c) with 9-point smoothing; (d) with an RCfilter and 9-point smoothing.

uct of time interval and the sum of peak values at every datapoints. These peak values were the signal values minus thebaseline values at every data points. The chromatographicprocess developed in the present work was proved effective.

The on-line HPLC system was effective for the determi-nation of ethanol and glucose concentrations in the fermen-tation broth. Linear calibration curves with a concentrationrange from 0 to 50 g/l were well obtained prior to fermen-tation. The pressure for the injection of sampled solutionswith or without dilution was well within the operating lim-its of HPLC instrumentation. One of the important featuresof this on-line monitoring and controlling system is theimplementation of commercialized automation equipmentfor auto-sampling. This equipment (BenchMate II) couldperform the pipetting of broth sample from the fermenter,

filtration of the sample, dilution of the filtered sample, and fi-nally injection of the filtered, diluted sample onto the HPLCthrough automation based on a programmed procedure. Incontrast to the auto-sampling device BIOPEM reported inthe literature [5], this equipment needs not a circulation forthe sampling. The BIOPEM has a drawback of large deadvolume (200 ml) outside the fermenter. The BenchMate IIis a robotic-based device and many steps of sample pre-treatment prior to injection to the HPLC can be combinedin a procedure for automation. Besides the solid–liquidseparation by filtration, pretreatment of cell debris is pos-sible by mixing cell debris with reagents for cell lysis. Thebroth sample thus could be treated for analysis of an in-tracellular component such as a protein product or enzymeactivity.


3.3. On-line estimation of biomass using pH control

In addition to the substrate and product concentrationsdetermined by the on-line HPLC, the biomass concentrationcould be estimated by using the data from pH control and asoftware sensor. The pH control was demonstrated effectivein the ethanol fermentation usingZ. mobilis. During thebatch and fed-batch culture, the pH of the fermentation brothwas controlled within the values of 5±0.05 under the on–offmode. Repeated runs with an initial glucose concentrationof 100 g/l resulted in the same batch time course as shownin Fig. 6. The ethanol and glucose concentrations in this fig-ure were determined by the on-line HPLC system, whereasthe biomass concentration was off-line measured (as shownby the data points). The biomass concentration could alsobe estimated, based on the fact that the growth ofZ. mobilisin fermentation culture is accompanied with the produc-tion of acid. The production of acid could be automaticallymeasured by the pH control system. The amount of NaOHadded to the fermenter in order to maintain the pH of thebroth at 5 corresponded to the amount of acid produced. Amethodology for on-line estimation of biomass concentra-tion was thus developed based on the acid production [17].

The production of acid in the growth ofZ. mobilis islargely due to the cellular incorporation of nitrogen fromammonium ions supplied as ammonium sulfate in the nutri-

Fig. 6. Batch culture ofZ. mobiliswith on–off pH control. The data pointsrepresent the on-line measurement of glucose (d) and ethanol (j), andthe off-line measurement of biomass (m). The solid curves represent theon-line estimation of biomass concentration.

ent medium. This process can be represented as [18]

NH4+ → NH3 + H+ (acid)

↓x (cells)

Thus, the rate of acid production is related to the rate ofnitrogen assimilation which, in turn, is related to the rate ofcell-mass production. We therefore have

1

Ap

dAp

dt= 1

V

dV

dt+ 1

x

dx

dt= α (1)

whereAp denotes the acid production,V the working vol-ume andx the biomass concentration. The proportionalcoefficientα is similar to the specific growth rate. Accord-ing to Eq. (1), an adaptive software sensor for the biomassestimation is introduced as follows:

dAp

dt= αAp + ω1(Ap − Ap) (2)

dα

dt= ω2Ap(Ap − Ap) (3)

dx

dt= αx − Dx + ω3(Ap − Ap) (4)

whereAp, α andx are on-line estimates ofAp, α andx, andω1, ω2 andω3 are positive tuning parameters. This adaptivealgorithm does not require any analytical expressions for thespecific growth rate. Since a pH control loop was employedin the fermentation process, the acid production data couldbe directly used for the adaptive sensor (Eqs. (2)–(4)) to esti-mate the biomass concentration. An algorithm for the on-lineestimation of biomass concentration was implemented in thiswork. The acid productions for batch and fed-batch fermen-tation were, respectively, shown in Fig. 7. The results from

Fig. 7. Acid productions for fermentations: (1) batch; (2) fed-batch withfed-back control; (3) fed-batch with constant feed flow-rate.


Fig. 8. Fed-batch culture ofZ. mobiliswith monitoring glucose concen-tration and fed-back controlling the feed flow-rate (triangle points). Thedashed curves represent the simulations using the model proposed byRivera under the constant feed flow-rate conditions. The solid curves rep-resent the on-line estimation of biomass concentration. The results fromthe fed-batch fermentation under the same conditions except for a con-stant feed flow-rate are shown in the circle points.

the on-line estimation of biomass concentration for batchand fed-batch fermentations were, respectively, shown in thesolid curves of Figs. 6 and 8. The on-line estimated biomassconcentrations agree very well with the off-line measureddata. The pH control system thus was not only useful formaintaining the pH of the fermentation broth, but also foron-line monitoring the biomass concentration.

3.4. Fed-back control of glucose concentration infed-batch fermentation

The fed-back control function of the on-line monitoryand controlling system was tested using a specified sub-strate (glucose) concentration profile as the objective. Thefeed flow-rate was regulated in the real time by the con-trolling system in order to direct the variation of glucoseconcentration in the broth following exactly the set profile.For a demonstration, the simulation using a kinetic modelproposed by Rivera [16] was set as the glucose concen-tration profile (dashed curve in Fig. 8) and input to thecontrol panel. The results from the on-line determining andcontrolling system are shown in Fig. 8. For a comparison,the results from the fermentation using a constant feedflow-rate (0.35 l/h) are also displayed. It was observed thatthe measured glucose and ethanol approached the set valuewhen the fed-back control was turned on. The fed-batchfermentation with fed-back control of glucose concentra-tion profile resulted in higher yields of ethanol and biomasswhen the feed of glucose was stopped at 9.5 h. The actualfeed flow-rate profile as shown in Fig. 9 was a result from

Fig. 9. Feed flow-rate for fed-back control (solid curve) and constant feed(dashed curve).

the regulation by the PC supervised system. Instead of aconstant value, the feed flow-rate varied in a manner ofa sequence of stepwise changes, with higher flow-rates atthe early time and lower flow-rates at the later time. Theexperimental data including ethanol, glucose and biomassconcentrations as shown in Fig. 8 did not exactly follow themodel simulation using the model proposed by Rivera [16].It was suggested that the actual glucose uptake rate wasgreater than the model prediction. However, we ignored thisproblem of model satisfaction and placed our attention onthe function of fed-back control by the present PC super-vised system. Undoubtedly, this monitoring and controllingsystem worked well. The fermentation results from fed-backcontrol as shown in Figs. 8 and 9 proved to be reproducibleand the same time courses were obtained from replicate runs.

4. Conclusion

A personal computer supervised system for on-line mon-itor and control a fermentation of microorganism was suc-cessfully developed. The system consisted of a fermenter,automatic sampling equipment, an on-line HPLC, manycontrol elements directed by PC, A/D and D/A interfacesand a supervised software programmed on Microsoft VisualBasic. The results from the on-line HPLC were captured,processed and used for the control in the real time. On-lineestimation of biomass concentration was also possiblethrough the use of pH-control coupling with a softwaresensor. This system was demonstrated to be effective forfed-back control of glucose concentration in the fed-batchfermentation of ethanol usingZ. mobilis. Various fermen-tation conditions could set for control through options onthe manus panel, including pH, stirred speed, gas flow-rate,


feed flow-rate and glucose concentration profile. More so-phisticate control strategies could be implemented on thissystem for fermentation of microorganisms or cell lines forspecific proposes. This monitoring and controlling systemcould be used as a research tool. It could also be appliedfor quality and process control in industrial processes.

Acknowledgements

Financial support from the National Science Council ofROC under grants NSC86-2745-E194-001R and NSC87-2214-E-194-003 is gratefully acknowledged.

References

[1] C. Turner, N.F. Thornhill, N.M. Fish, Biotechnol. Tech. 7 (1993)19–24.

[2] J. Moller, R. Hiddessen, J. Niehoff, K. Schugerl, Anal. Chim. Acta190 (1986) 195–203.

[3] K. Holzhauer-Rieger, W. Zhou, K. Schugerl, J. Chromatogr. 499(1990) 609–615.

[4] S.K. Paliwal, T.K. Nadler, D.I.C. Wang, F.E. Regnier, Anal. Chem.65 (1993) 3363–3367.

[5] H. Lundstrom, M. Brobjer, B. Osterlof, T. Moks, Biotechnol. Bioeng.36 (1990) 1056–1062.

[6] C. Turner, M.E. Gregory, N.F. Thornhill, Biotechnol. Bioeng. 44(1994) 819–829.

[7] P.L. Rogers, K.J. Lee, D.E. Tribe, Biotechnol. Lett. 1 (1979) 165–170.

[8] K.J. Lee, M.L. Skotnicki, D.E. Tribe, P.L. Rogers, Biotechnol. Lett.2 (1980) 334–339.

[9] P.L. Rogers, K.J. Lee, M.L. Skotnicki, D.E. Tribe, in: A. Fiechter(Ed.), Advance in Biochemical Engineering, Springer-Verlag, Vol.23, 1982, pp. 37–84.

[10] J. Baratti, J.-D. Bu’Lock, Biotechnol. Adv. 4 (1986) 95–115.[11] M. Beavan, B. Zawzdski, R. Droniuk, H. Lawford, J. Fein, Appl.

Biochem. Biotechnol. 20 (1989) 319–326.[12] H.W. Doelle, L. Kirk, R. Crittenden, H. Toh, M. Doelle, Crit. Rev.

Biotechnol. 13 (1993) 57–98.[13] G.A. Sprenger, M.A. Typas, C. Drainas, World J. Microbiol.

Biotechnol. 9 (1993) 17–24.[14] L.O. Ingram, H.C. Aldrich, A.C.C. Borges, T.B. Causey, A. Martinez,

F. Morales, A. Saleh, S.A. Underwood, L.P. Yomano, S.W. York, J.Zaldivar, S. Zhou, Biotechnol. Prog. 15 (1999) 855–866.

[15] A. Savitzky, J. Golay, Anal. Chem. 36 (1964) 1627.[16] S.L. Rivera, Neural networks and micro-genetic algorithms for state

estimation and optimization of bioprocesses, Ph.D. Dissertation,Colorado State University, Fort Collins, CO, 1992.

[17] F.S. Wang, W.C. Lee, L.L. Chang, Bioprocess Eng. 18 (1998) 329–333.

[18] U. Veeramallu, P. Agrawal, Biotechnol. Bioeng. 31 (1988) 770–782.

Documents

On-line monitoring and controlling system for fermentation processes