MODELS FOR PREDICTING WINE FERMENTATION KINETICS

Transcription

1 MODELS FOR PREDICTING WINE FERMENTATION KINETICS Brian O Neill 1, Torbjorn van Heeswijck 1 and Richard Muhlack 2. 1 School of Chemical Engineering, The University of Adelaide, Adelaide, South Australia AWRI PO Box 197, Glen Osmond SA brian.oneill@adelaide.edu.au ABSTRACT Fermentation of grape juice by yeast is a critical stage in industrial wine production. However, the kinetics of the process are poorly understood due to the extreme conditions present during such fermentations. Problematic fermentations occur regularly and result in significant cost as a result of wasted tank capacity and low value of the final degraded product. Control of the fermentation process is important to avoid stuck fermentations (a stuck fermentation is a fermentation that has stopped before all the available sugar in the wine has been converted to alcohol and CO 2 ) and clearly, the fermentation unit operation strongly influences the aesthetic endcharacteristics of the wine. A variety of models have been proposed to predict the dynamic behaviour (kinetics) of the process. Two traditional and simple biochemical models (Monod kinetics and the Gompertz model) are predominantly employed and these were investigated in this work. The primary aim of this study was to determine the ability of both models to predict fermentation behaviour when fitted to data from early stages of fermentation. Initially, the Monod and Gompertz models were fitted to sugar consumption data for laboratory scale wine fermentations using least squares regression. The model produced a reasonable qualitative fit for the kinetic (growth and production) data with a root-mean-squared error (RMS) of 21 g/l or less in each case. Both models were then fitted to sugar consumption data from twenty two industrial fermentations over a varying number of time steps following the initial measurement in each data set. The number of time steps required to produce an RMS error less than 20 g/l was 8 days in 19 cases using the Monod model. However, the Gompertz model (successful in 18 cases) generally required 2 4 fewer time steps (days). No correlation was found between the number of time steps required or the regressed parameters and the volume of fermentation. The values of the regressed Monod parameters maximum growth rate (µ m ), the Monod constant (K S ) and the yield coefficient (Y X/S ) lay between day -1, g/l and g cell X/g substrate (assumed to be sugar content) S in most cases, with deviations loosely correlated to the use of Semillon grapes. Both models exhibited promise for use in industry alongside traditional winemaking techniques, depending upon the specific goals and requirements of each individual winery. Rigorous error analysis was not possible due to a lack of supplied experimental uncertainty data and this will be investigated in future work. 1. INTRODUCTION Australia is consistently ranked as one of the top ten wine producing countries in the world, with approximately 1.23 billion litres including exports valued at $2.87 billion 14/10/2008 Page 1

2 sold in (Australian Government Department of Foreign Affairs and Trade 2008). The industry operates in every state and territory of Australia and has earned a reputation for its development and use of technological advancements in grape growing and wine making. Ongoing research, development and improvement in many areas of the wine making process are vital if the Australian wine industry is to maintain this competitive advantage, and one such area is the dynamic monitoring, simulation and prediction of primary yeast fermentation of grape juice. Fermentation control is a significant, challenging and complex task in industrial winemaking. Deviations from expected fermentation behaviour can significantly alter the flavour and aroma characteristics of the final product, as well as reducing the efficiency of capital equipment use. In severe cases, a stuck fermentation may occur with cell growth ceasing before the available sugar being used. If not avoided or corrected, entire batches of grape juice may be rendered unsuitable for conversion into high quality wine with significant economic loss. 2. VARIABLES AND MODELS The major factors affecting the performance of wine yeast fermentations are the availability of nutrients (including sugars, oxygen and nitrogen), the alcohol content and its subsequent toxic effect on yeasts, the ferment temperature and ph, and the capacity for CO 2 generation under pressure. All factors must be considered in industrial wine fermentation and ideally a fermentation model should incorporate terms accounting for several of these factors (Jackson 2000). 2.1 The Baume and Brix Scales The Baume scale ( o Baume) is widely used as a method of approximating the sugar and alcohol content of wine. It quantifies the specific gravity of a solution and is closely related to the potential ethanol product of a grape must, though in a slightly non-linear fashion (Boulton et al 1998). The degrees Brix ( Bx) is a measure of the sugar content of an aqueous solution. One degree Brix is 1g in 100g of solution. It is traditionally used in the wine, sugar and fruit juice industries. Both parameters are widely used by oenologists to track sugar consumption during wine fermentation typically using either hydrometers or refractometers. 2.2 Monod Kinetics The kinetics of grape juice fermentation are poorly understood, as the behaviour of a fermentation differ markedly for different grape and yeast varieties coupled with the unique conditions under which grape juice fermentation occurs (Boulton 1980). The environment is typically hostile (low ph, high bi-sulfite ion and high ethanol concentrations); substrate inhibition (high sugar levels) and competitive inhibition (the presence of both D-glucose and D-fructose) occur during the progress of fermentation. Kinetic models of grape juice fermentation are typically knowledge based, modifying general mechanistic biochemical growth equations (e.g. Monod kinetics, enzyme inhibition equations) whilst accounting for as many variables as deemed appropriate. Monod s model developed in the 1940s by the Nobel Prize winner Jacques Lucien Monod. It is based on the concept of a limiting nutrient (sugar transport is the rate-limiting step in wine fermentations). These models present problems when used to predict industrial fermentation s 14/10/2008 Page 2

3 behaviour due to the large number of influential variables, many of which may be difficult to measure directly (Remedios 1999). Yeast growth kinetics are normally assumed to follow Monod kinetics dx with = µ X µ = µ m S ( KS + S) dt (1) where X is the total (viable) yeast cell mass, µ is the specific growth rate, S is the nutrient concentration, µ m and K s are constants coupled with a differential equation for substrate linked through the yield and maintenance coefficients Y X/S & m, respectively. ds µ X = + mx (2) dt YX/S Various forms of inhibition can also be accounted for. A consensus in the literature is that the product ethanol behaves as a non-competitive inhibitor and its effect is normally modelled by using a modified µ m value in the Monod equation (equation 3). Here, µ i reduced maximum growth rate growth due to inhibition and K i is the inhibition constant. S µ m = µ i (3) Ki + S A broad range of equations are used by the various literature models - a number correlations have been proposed for µ i, and most models ignore substrate inhibition although sugar concentrations above 200 g/l have been shown to affect cell growth rate (Starzak et al 1994). Self-inhibition effects have been observed as increased yeast inoculums can cause the specific growth rate to decrease, and a model incorporating this effect has proven appropriate when dealing with fermentations at varying temperatures or sugar concentrations (Remedios 1999). Product formation kinetics can be expressed either by the Luedeking and Piret (1959) equation (equation 4) or in a form analogous to the cell growth rate (equation 5). The former has only been fully verified for low sugar concentrations, consequently most models use the latter with a number of different equations for the specific product formation rate proposed. dp dx = α + β X = αµ X + β X (4) dt dt dp X dt = υ (5) where P is the product concentration (ethanol), α, β are constants and υ is the specific product formation rate (metabolic quotient). The substrate consumption kinetics in wine fermentation processes are modelled in a number of different ways with many contradicting literature equations. Methods based on an assumed yield from the stoichiometry of ethanol production from sugar have shown that in practise ~85-95% of the theoretical yield was achieved with the remaining sugar consumption attributed to respiration and cell growth. The most comprehensive kinetic model was proposed by Boulton (1980), based upon the assumptions that transfer of sugars (assumed limiting nutrient) into yeast controlled the fermentation rate. This rate was modelled as analogous to an enzyme reaction in growing yeast, and a constant diffusion in non-growing viable yeasts. The sugar, yeast and ethanol concentration were included as parameters along with the liquid temperature via an Arrhenius-type model (equation 6) and an energy balance. 14/10/2008 Page 3

4 Ethanol production was directly related to substrate consumption (equation 7) and was used to predict yeast cell deactivation (equation 8). where [ 14200(T 300) ] [ (T 300) ] 300RT 300RT µ m = 0.18e e (6) d[ E] 92 ds = dt 180 dt (7) E t XV = X (8) R is the universal gas constant, V denotes the concentration of viable cells, X is the total cell mass, T is the liquid temperature and E is the concentration of ethanol in the fermentation. This model was successfully used to predict the progress of model fermentation data (figure 1). The differential equation models were solved numerically for the initial conditions and other parameters then fitted to the first few data points via leastsquared regression. Figure 1 Prediction of Brix Levels in Model Fermentation (Boulton 2005) The effects of nitrogen concentration and mixing effectiveness have not been thoroughly explored in the literature despite their strong effect on the progress of wine fermentation. 2.3 Gompertz Model The Gompertz equation is an empirical function describing growth in species with limited nutrients, based on the assumption that the probability of organism death increases exponentially with time. The Gompertz function was originally formulated in actuarial science for fitting human mortality data but it has also been applied deterministically to microorganism growth, organ growth and cancer mortality. Its differential form is summarised in equation 9. (Kaplan & Glass 1995). dx αt = ( ke ) x (9) dt 14/10/2008 Page 4

5 where x is a dummy variable, k and α are constants. Integrating this equation and then substituting the measured parameters yields equation 10 (Giovanelli, Peri & Parravicini 1996). µ m e ln ( C(t) C ) ( ) ( ) ( ) 0 = ln C C0 exp exp λ t + 1 ln C C0 where µ m is the maximum specific growth rate, e is the exponential constant = 2.718, C is the value of C(t) at steady state and λ is a lag phase constant. This equation can be used to model sugar concentrations in yeast fermentations if C(t) is defined as the sugar consumed at time t: (10) C(t) = S0 S(t) (11) and S 0 is the initial sugar concentration and S(t) is the sugar concentration at time t. The Gompertz equation has also been successfully used to model yeast growth, ethanol production and more with similar definitions of C(t) (Giovanelli, Peri & Parravicini 1996). 3. MATERIALS AND METHODS 3.1 Fermentation Data All fermentation data was provided by Dr. Richard Muhlack from the Australian Wine Research Institute (AWRI) Laboratory Data Three samples each of 0.2 ml and 200 ml Chardonnay juice ferments containing 107 g/l each of fructose and glucose were inoculated with Saccharomyces cerevisiae strain 71B at a concentration of 6x10 5 cells/ml. The glucose and fructose concentrations during each ferment were measured by HPLC at semi-regular intervals over a period of 191 hours according to the method outlined in Castellari et al (2000) Industrial Data The o Baume of 22 industrial wine fermentations from a variety of sources were measured daily (where possible) using a refractometer over a period of 12 to 17 days. Temperature data was also collected over the same time period. This data was collated along with the grape variety, yeast strain, growing region and ferment volume as summarised in Table1. 14/10/2008 Page 5

6 Table 1 Industrial Fermentation Details Ferment ID Grape Variety Yeast Strain(s) Growing Region Ferment Volume (L) T Riesling QA23 Watervale 9,197 T Semillon QA23/Vin7 Barossa 7,832 T Riesling QA23 Watervale 22,400 T SAB VL3/Vin13 Padthaway 29,540 T SAB QA23/Vin7 Mudgee 103,336 T Semillon QA23 Loxton 171,256 T Semillon EC1118 Barossa 138,000 T SAB QA23/Vin7 McLaren Vale 67,239 T Semillon EC1118 Riverland 549,302 T SAB QA23/Vin7 Sunraysia 228,000 T Semillon QA23 Barossa 212,806 T Riesling QA23 Barossa 203,925 T Riesling QA23 McLaren Vale 220,096 T Semillon EC1118 Riverland 581,805 T Semillon EC1118 Riverland 614,200 T Riesling QA23 Clare 43,117 T Riesling QA23 Watervale 9,320 T Riesling QA23 Watervale 9,170 T SAB QA23/Vin7 Adelaide Hills 8,698 T Riesling QA23 Barossa 97,995 T SAB QA23/Vin7 Adelaide Hills 105,981 T Riesling QA23 Watervale 13, Monod & Gompetz Kinetic Modelling Regression to the basic Monod model (equations 1 & 2 to determine µ m, K s, Y x/s & m) and the Gompetz model (equations 10 & 11 to determine µ m & λ) was performed for both the laboratory and industrial fermentation data using the Microsoft Excel with add-in RK4 Version 3.0 ( Michigan Technological University 2006) used for the Monod model. The use of Excel rather that a more sophisticated package (such as Matlab s least squares curve fit algorithm) was a deliberate choice. This work was directed toward wine makers who normally have extensive familiarity with and readily available access to Excel spreadsheets & add-ins but have little experience of more sophisticated curve-fitting software. The goal is to follow a wine fermentation in its initial stages, then, fit a simple model which may be subsequently used to predict future outcomes and to detect likelihood of problematic ferments (e.g. stuck ferments, infestation by spoilage yeasts and bacteria, etc.). Initial guesses for each parameter were required. The software package was then run individually for each data set by entering the sugars-time data to predict the modelled sugar concentration at each data point for the given initial conditions. The squared error at each data points was also calculated and used to calculate the root-meansquared (RMS) error of the model compared to the experimental sugar concentration data. The Excel Solver add-in was then used to minimise the RMS error of the model by iteratively adjusting the parameters given in with all.parameters constrained to be positive ( 0). The final values of each parameter and the RMS error across all data points were collected in each case. 14/10/2008 Page 6

7 3.2.1 Laboratory Data The above procedure was followed for each data set with a modified spread-sheet that simultaneously calculated the squared error of the model solution compared to each of the three experimental data sets for a given fermentation and then combined them to provide a single RMS error. The initial yeast and sugar concentrations were considered constants during modelling. This process was repeated for each ferment with K S = 10, 200 and 500 g/l along with the constraint Y X/S 0.01 g X/g S for Monod model. The Gompertz model was also successfully fitted to this data using non-linear regression with Excel s Solver.. Note: In grapes, a large portion of the soluble solid is sugars. Glucose and fructose are the main sugars in the juice. The sugar content of the juice of ripe grapes varies between 150 to 250 g/l. In unripe berries, glucose is the predominant sugar. At the ripening stage, glucose and fructose are usually present in roughly equal amounts (1:1 ratio). In overripe grapes, the concentration of fructose exceeds that of glucose. In ripe grapes, there is some variation in the glucose to fructose ratio among the grape varieties. For example, Chardonnay and Pinot Blanc are classified as high fructose varieties, while Chenin Blanc and Zinfandel are regarded as high glucose varieties. Glucose and fructose are fermentable sugars. During the course of fermentation, the yeast converts these sugars to alcohol and carbon dioxide. The amount of alcohol produced is related to the amount of sugar initially present in the juice; thus, by controlling the amount of sugar in the juice, one can control the amount of alcohol in the resulting wine. It should be noted that the relationship between sugar content and alcohol formed is not precise. Roughly speaking, the conversion of sugar to alcohol is Brix x 0.55 = % of alcohol in wine Industrial Data Monod Model The Monod kinetic model was fitted as described to fermentations T-017-1, T-004-2, T-235-1, T-243-1, T and T using the initial guess coupled with the constraint Y X/S 0.01 g X/g S. The modelling process was repeated for fermentation T with the constraint that the maintenance term m = 0 g S/g X.day. The Monod model was fitted to fermentation T using the initial guess in table 1 and the constraints Y X/S 0.01 g X/g S, m = 0 g S/g X. day, using only the first 2 data points when calculating the RMS error to minimise. The RMS error including all data points was then calculated, and the process repeated with an additional data point in the minimised RMS error calculation. This continued until the RMS error including all data points was 20 g/l. This process was then repeated for the remaining industrial fermentation data sets. The Monod equations were also modified in the RK4 package to include ethanol (E) production and cell viability terms (equations 2 and 3) and the modelling process repeated for fermentation T using 7 data points and an initial guess of E 0 = 0 g/l. 3.3 Gompertz Equation Modelling Sugar consumption against time was calculated from the experimental sugar concentration data for fermentation T using equation 11 where S 0 was the first recorded sugar concentration, and the theoretical sugar consumption calculated using equation 10. These calculations were set up in an Excel spreadsheet along with 14/10/2008 Page 7

8 formulas to calculate and minimize the RMS error between the model and the experimentally derived values using Solver. The same method for modelling on a minimum number of data points outlined above was then performed. This procedure was carried out for all remaining industrial fermentation data sets. 4. RESULTS 4.1 Monod Kinetics Modelling Laboratory Data K S,i = 500 K S,i = 10 K S,i = [S] (g/l) Time (Days) Figure K S,i = 500 K S,i = 10 K S,i = Time (Days) Figure 3 Figures 2 and 3 show the curves derived by fitting Monod kinetics to the total sugar content of 200 ml and 0.2 ml wine fermentations. Results for three values of K S,i are shown for each ferment volume. 14/10/2008 Page 8

9 K S,i = 500 K S,i = 10 K S,i = [S] (g/l) Time (Days) Figure K S,i = 500 K S,i = 10 K S,i = Time (Days) Figure 3 Figures 2 & 3 Monod Kinetic Model fitted to Total Sugar Concentration in a 200 ml and 0.2 ml Fermentations, respectively. Figures 4 and 5 show similarly derived curves for separately measured glucose and fructose content respectively. A K S,i value of 200 g/l was optimal in both cases. 14/10/2008 Page 9

10 Time (Days) Figure 4: Fermentation data for Glucose in 200 ml fermentation [S] Fructose(g/L) Figure 5 Monod kinetics fitted to Fructose content in 200 ml fermentation Table 2 shows the final value of key parameters for the best-fit Monod kinetic models achieved for each laboratory data set. 14/10/2008 Page 10

11 Glucose Fructose Total Sugars Volume (ml) µ m (day -1 ) K S (g/l) Y X/S (g X/g S) m (g S/g X. day) RMS Error (g/l) Table 2 Parameters Derived from Fitting Monod Model to Laboratory Fermentations Fitting of Industrial Data Sets Figures 6 & 7 show the curves derived by fitting Monod kinetics to the total sugar content of three typical industrial fermentations [S] (g/l) Time (days) Figure 6 Monod Kinetics fitted to all data points for fermentation T /10/2008 Page 11

12 [S] (g/l) Figure 7 Monod Kinetics fitted to all data points for fermentation T Ferment ID T T T T T T X 0 (g/l) S 0 (g/l) µ m (day -1 ) K S (g/l) Y X/S (g X/g S) m (g S/g X. day) RMS Error (g/l) Table 3: Key parameters of three typical modelled fermentations Table 3 shows the final values of key parameters for typical modelled fermentations. 4.2 Summary of Results Least squares regression was used to fit a simplified form of the Monod kinetic model described by Boulton (1980) to laboratory data for consumption of glucose, fructose and total sugars in Chardonnay juice fermented by the Saccharomyces cerevisiae yeast strain 71B. The four regressed kinetic parameters µ m, K S, Y X/S and m for total sugar consumption in a 200 ml ferment were 8.9 day -1, 500 g/l, 0.01 g X/g S and 21.7 g S/g X.day respectively. The fitted model s root-mean-squared error was 21 g/l or less in each case. The Monod kinetic model was also fitted to six sets of Baume-time data from industrial fermentations with varying grape varieties, yeast strains and ferment volumes with an RMS error of 12 g/l or less in each case. The maintenance term, m, was regressed to 0 g S/g X.day in several cases. The model without the maintenance term was also fitted to the industrial fermentation data with equal accuracy. The 14/10/2008 Page 12

13 addition of ethanol production inhibition and cell death kinetics produced no change in the ability of the model to fit the data. The model without the maintenance term was next fitted to twenty-two sets of Baume-time data from industrial fermentations at varying time periods from the first measurement in each data set. The number of time steps required for modelling each data set with a RMS error of 20 g/l or less was recorded, with 85% of fermentations modelled after 8 time steps and all modelled after 10 time steps. Fermentations of SAB grapes using the QA23/Vin7 yeast strain combination were found to require the least number of time steps to accurately model, but no correlation between fermentation volume and required time steps was observed. The regressed kinetic parameters µ m, K S and Y X/S were generally in the ranges day 1, g/l and g X/g S respectively, however, in some cases all three parameters values were an order of a magnitude higher. This was loosely correlated with Semillon grape fermentations and fermentations requiring 8 or more time steps to accurately produce the model. No correlation was found between regressed kinetic parameters or the time steps required to produce an accurate and the volume of fermentation. Least squares regression was used to fit the Gompertz equation of Giovanelli, Peri & Parravicini (1996) to the same twenty-two industrial fermentations at varying time periods. The number of time steps required for modelling to produce an RMS error of 20 g/l or less over all data points was generally found to be 2 4 fewer than for the Monod model. However, the Gompertz model often failed to fit the data at all over ranges of 6-10 time steps, but the use of parameters fitted to lower numbers of time steps as the initial guess for modelling was found to prevent this error. Qualitatively, the Gompertz model did not appear to fit data as well as the Monod model during later stages in the fermentation process. 4.3 Industrial Significance and Recommendations The Monod kinetics model without a maintenance term could be applied to the majority of fermentations if measurements were taken consistently over a period of eight days, whilst, the Gompertz model could successfully be applied to many fermentations for measurements taken consistently over a period of five days. Implementing better initial guesses for both models could further improve the accuracy and decrease the number of measurements required for both models. Clearly, from an industrial perspective it is highly desirable that an accurate model of a given fermentation be available after as few time steps as possible, so the results must be further considered in the context of industry expectations such as the length of a typical fermentation or time restrictions on correction of undesirable fermentation behaviour. If these models were to be used in an industrial context it is likely that some calibration of the models to the winery s individual processes and requirements would be necessary, and operator experience would be essential in determining which model is appropriate to use at which stages of fermentation. Clearly, there is a definite potential for the modelling approach to help guide fermentation control alongside traditional methods. 14/10/2008 Page 13

14 5. CONCLUSION A study of the use of two simple models for the prediction of the kinetics of wine production has been completed. The two models are robust and of low order Monod kinetics requiring estimation of four (4) parameters (two of which namely the yield and maintenance coefficients have only minor effects), whilst the Gompertz model involves only two parameters. Both provided reasonable estimates of the dynamics of wine fermentation at laboratory and industrial scales. Significant experimental uncertainties were evident in the fluctuating nature of the industrial fermentation data. This may have arisen due to errors in instrument calibration and operation or may arise from physical and biological variations in the system such as incomplete mixing of fermentations. Rigorous error analysis could not be applied due to a lack of data but it is not expected that this will significantly impede initial application of the models in industrial applications. The Monod and Gompertz models were found to accurately model sugar consumption in wine fermentation and exhibited an ability to predict future fermentation behaviour based on model regression of operating data from a relatively small number of initial samples. Adaptation of one or both of these models to industry use as an aid in control of large scale fermentations appears to be a real possibility. Model choice will depend upon the desired length of fermentation, ability to regularly and accurately monitor sugar levels and the stages of fermentation at which accurate prediction is critical. Future predictions from such models and a careful examination of any deviations from the normal range of key modelling parameters will help to rapidly detect potential problematic ferments (e.g. stuck ferments, presence of odour compounds created by the rogue yeast Brettanomyces bruxellensis [simply "Brett" in most conversations] and its anamorph relative Dekkera bruxellensis, etc). Consultation with the wine industry and trialling of the models in an industrial context is a future goal particularly if cheap, robust and reliable online sensors for sugar and alcohol become available. Further research should include investigation of the effect of different yeast strains and grape varieties, increased attention to and reduction of sources of error and finally improved understanding of the causes behind large changes in regressed kinetic parameters. 6. NOMENCLATURE 6.1 Symbols C Change in sugar concentration g/l E Ethanol concentration g/l K s Saturation Constant g/l m Maintenance coefficient g S/g X.day P Product concentration g/l R Universal gas constant J/mol.K RMS Root-mean-squared - S Substrate concentration g/l t Time days T Temperature K x General Gompertz variable - X Yeast concentration g/l Y Growth yield coefficient g X/g S 14/10/2008 Page 14

15 Subscripts, Superscripts and Others Steady-state value (at t = infinity) - 0 Initial value (at t = 0) - [ ] concentration -mol/l i inhibited - m Maximum - t Value at time t - V Viable - Greek Symbols α Luedeking constant - General Gompertz constant - β Luedeking constant day -1 λ Gompertz lag phase days µ Yeast specific growth rate day -1 Gompertz growth rate g/(lxday) υ Specific product formation rate day -1 (metabolic quotient) 7. REFERENCES Australian Government Department of Foreign Affairs and Trade 2008, The Australian Wine Industry, viewed 20 May Boulton, R 1980, The prediction of fermentation behaviour by a kinetic model. Am. J. Enol. Vitic., vol 31, pp Castellari, M, Versari, A, Spinabelli, U, Galassi,S, Amati, A 2000 An improved HPLC method for the analysis of organic acids, carbohydrates, and alcohols in grape musts and wines, Journal of Liquid Chromatography & Related Technologies, 23, Chapra, S C 2004, Applied numerical methods with MATLAB for engineers and scientists, McGraw-Hill Professional. Giovanelli, G, Peri, C, Parravicini, E 1996, Kinetics of grape juice fermentation under aerobic and anaerobic conditions, Am. J. Enol. Vitic., vol 47, no. 4, pp Luedeking, R, Piret, E L 1959, A kinetic study of the lactic acid fermentation: batch process at controlled ph, J. Biol. Microbiol. Technol. Engr., vol 1, no. 4, pp Remedios, M 1999, Alcoholic fermentation modeling: current state and perspectives, Am. J. Enol. Viticult., vol 50, no. 2, pp Starzak, M, Krzystek, L, Nowicki, L, Milchaski, H 1994, Macro-approach kinetics of ethanol fermentation by Saccharomyces cerevisiae: experimental studies and mathematical modelling, Chem. Engr. Jnl., vol 54, pp Zoeklein, B 1995, Wine Analysis and Production, Chapman and Hall, New York. 14/10/2008 Page 15

16 BRIEF BIOGRAPHY OF PRESENTER Please include the biography here Associate Professor Brian O Neill is currently a member of staff in the School of Chemical Engineering at the University of Adelaide. His current research interests span a broad spectrum including modelling and simulation, wine processing, production of bio-diesel from wastes, minerals processing and geothermal energy.