当前位置:首页 >> 农林牧渔 >>

Ai-based software tools for beer brewing monitoring and control


REVIEW

DOI: 10.2478/V10133-010-0060-0

FB

AI-BASED SOFTWARE TOOLS FOR BEER BREWING MONITORING AND CONTROL
1

S. Vassileva1, S. Mileva2 Institute

of Control and Systems Research, Bulgarian Academy of Sciences, Sofia, Bulgaria 2 Institute of Cryobiology and Food Technologies, Sofia, Bulgaria Correspondence to: E-mail:

ABSTRACT

Brewing fermentation is among the most well studied processes in the food industry. Nevertheless, the process still provides challenges to the brewers. It is subject to alteration stemming from the variation in the yeast, a living organism, and due to the complex raw materials of biological origin. With a reliable predictors- artificial intelligence (AI)-based software tools as well as software analyzers for the fermentation monitoring and control one could manage with fewer measurements especially when an early warning gives the operators time to make corrective actions. Our work summarizes research results obtained in intelligent software tools design for beer manufacturing carried out in the Institute of Control and System Research and Institute of Cryobiology and Food Technologies as an answer of world trend in knowledge-based bio-economy establishment. Biotechnol. & Biotechnol. eq. 2010, 24(3), 1936-1939 Keywords: software analyzer, brewing fermentation, brewing yeast, artificial intelligence, knowledge-based bio-economy The potential of neural nets то monitor and control dynamical systems was discovered two decades ago (3, 13) and many applications in industrial processes exist. Informative measurements are critical for the success of predictive modeling of such processes. Austin et al. (2) present an early neural approach and Beil et al. describe a hybrid system of an extended Kalman filter, a fuzzy model and a neural network (4). Both models, however, rely on parameters not typically available in industrial-scale breweries. In this paper we summarize AI-based application for the prediction of key-process of early stage of beer fermentation, secondary beer fermentation process and beer quality parameters. The system predicts the progress of the fermentation based on properties of the raw material, yeast Saccharomyces uvarum (carlsbergensis) condition and history, fermentation temperature and aeration.

Introduction
Biological processes in food industry, partially in beer brewing, are among the challenging ones to establish control, due to the complexity of the biological raw materials and the use of living organisms, such as yeast, as processing agents. Brewing is among the most well studied processes in the food sector. Nevertheless, the process still provides challenges to the brewers. It is subject to alteration stemming from the variation in the yeast, a living organism, and due to the complex raw materials of biological origin. the variability may manifest in changes in the fermentation process or in the beer quality. Thus, there is a need for tools, both analytical and computational, to assist in the monitoring the process and to keep it in a desired course. The breweries are forced to make daily measurements in order to observe the course of the fermentations. With reliable predictors such as intelligent software tools as well as software analyzers for the fermentation monitoring and control one could manage with fewer measurements especially when an early warning gives the operators time to make corrective actions (11, 14, 17, 18, 19). Many proposals for the modeling of beer fermentation exist in the literature, perhaps due to the strong link to the models of alcoholic fermentation and biotechnological processes (10, 12, 15). Kinetic models have been presented by Gopal et al. and Johnson et al. (6, 7), fuzzy modeling by Vassileva et al. and Cummins et al. (5, 16), and a rule-based system by Kashihara et al. (8). Such models are typically developed and tuned by hand, which is a considerable burden. 1936

Brewery fermentations

The main fermentation process starts with aerating the cooled wort and adding yeast to it. The yeast starts to consume the nutrients contained in the wort, in order to stay alive and grow. At the same time, the yeast produces alcohols and other metabolites. The sugar content of the wort is the primary measurement by which the course of the fermentation is followed; the main fermentation is deemed to end when the sugar content falls below a predefined concentration level, so that only a small amount of fermentable sugars remain. The sugar content of the wort is expressed by the specific gravity of the wort; the wort gets lighter as sugar is converted to alcohol and other compounds. The traditional batch main fermentation phase takes about a week, although faster processes have been developed (9). It is followed by a lagering stage, where some undesirable Biotechnol. & Biotechnol. eq. 24/2010/3

compounds are further converted. Depending on the production line the maturation takes place in the same or in a different tank. In the end of fermentation the yeast is cropped, and used for another pitching. Fermentation is controlled by regulating the temperature, oxygen content, and the pitching rate; i.e., the amount of yeast put into the fermentation tank. The temperature has a great influence on the rate and extent of fermentation and on the quality of the final product. The yeast growth can be controlled by the oxygen content in the early stage of fermentation, after pitching. In addition, the course of fermentation is affected by other factors, such as the wort composition and the yeast condition. Ideally, these factors should be constant, so that the predictability of fermentation is maintained. In practice, neither the wort composition nor yeast condition is static. The natural variation of malt induces some variations to the wort composition, although such variations can be diminished by re-planning the mashing recipes (1). The condition of the yeast is a more complicated issue. Traditionally, the breweries observe the vitality, i.e. the percentage of live and dead cells before pitching by laboratory analyses. However, these methods do not tell anything about the viability of the yeast, i.e. the fermentation ability of the cells. The common practice is the yeast to be used many times before disposal. The ability of the yeast to ferment depends to a great extent on the physiological state of the yeast. For example, new yeast typically behaves differently from yeast that has been recycled many times. Also, yeast that has been stored for long periods between fermentation is often less vital. Ideally, the brewery should be able to modify the fermentation process so that the variability of the yeast and wort would be canceled out. So, if the viability of the yeast is low, the brewery could increase the pitching rate or slightly elevate the temperature or oxygen content. A recipe planner will be well suited to the similar tasks (1), although a reliable estimate of the yeast vitality and viability is needed. However, it can be expected from the above introduction, no single analysis exists that would permit prediction of the time of fermentations to any reasonable degree. Monitoring and control of beer fermentations with AIbased software tools The modeling and control of biotechnological fermentations using artificial intelligence approaches as well as fuzzy logic, neural networks and its hybrid intelligence methods is a wellstudied problem (10, 15). However, the models from the biotechnical literature cannot be directly transferred to brewing, as many measurements in the models are not available in the brewery. For example, carbon dioxide of exhaust gas is rarely measured, neither is biomass during fermentations (10, 15). In practice the set of analyses at the modeler’s disposal is very limited, including mostly amounts and timing of transfers of raw materials and yeast, process temperatures and the manually taken gravity (sugar content) measurement. These restrictions also rule out the previous adaptive models developed specifically for brewing fermentations. The model by Austin et Biotechnol. & Biotechnol. eq. 24/2010/3

al. (2) requires wort free amino nitrogen measurement (FAN) that is not a common analysis in some breweries. The model by Beil et al. (4), on the other hand, requires the CO2 evolution rate as input. Our investigations in beer brewing monitoring and control by implementing AI-based software tools and knowledgebased methods are an answer to the world and EU HACCP demand to establish a knowledge-based bio-economy. Recent methods for automatic control and monitoring of technological processes accelerate introduction of new class of knowledgebased system (KBS)- intelligent software analyzer, known also as software sensor systems for inferential or indirect measurements (1). Preparation stage of beer fermentation includes production of optimal content of substrates and pitching yeasts. Industrial yeasts strains are microbial populations with certain morphological and physiological content and features of cells. With the methods of artificial intelligence (AI), implemented on experimental data, software analyzers are synthesized for defining: concentration of various kinds of yeast cells (living, death, weak, budding); budding energy and fermentation activity of brewing yeasts (16). in the early stage of brewing fermentation, obtaining of industrial strains with the required features is significantly dependent on synthesis of ergosterol and dissimilation of glycogen- substances whose content is traditionally defined by biochemical analysis, requiring for preliminary preparation of samples and equipment. In order to shorten time for analysis, for labor and staff savings and mainly to achieve statistical results of higher accuracy in comparison with the accuracy of biochemical analysis, using neuro-fuzzy approach for data approximation, the software analyzers for real-time measurement of ergosterol and glycogen were developed. Synthesized software analyzers show high accuracy and both are sensitive to the cultivation conditions (varying initial concentration of the dissolved oxygen content) (18). Prediction of beer quality on the basis of real process data is carried out by using fuzzy rules, to characterize the influence of beer fermentation on the beer quality indices as well as vicinal diketones (VDK), higher alcohols and fatty acids (17). Some new software analyzers of factors, influencing beer quality, as well as VDK, higher alcohols (MeOH, PrOH, EtAc, BuOH and AmOH) and fatty acids (C5, C6, C8 and C10) by implementing artificial neural networks (ANN) are presented in (19). As input attributes of the analyzer models measured or indirectly evaluated pitching yeast, temperature, oxygen content, free amino nitrogen content, ergosterol initial content, etc. are used. The universal predictors are biomass and limiting substrate content. Fig. 1 depicts diacetil prediction on the basis of biomass and free amino nitrogen content for strain S. cerevisiae (carlsbergensis) 56 at low oxygen content of 0.2 mg/ dm3. Diacetyl and pentane-2,3-dione(VDK) both impart a characteristic aroma and taste to beer; this is variously described as “buttery” ,”honey or toffee-like” or as “butterscotch”. 1937

Their formation is linked to yeast protein synthesis and they are formed from keto-acids, which in turn may be formed by transamination or deamination of the amino acids in the wort, or synthesized from wort carbohydrates. The yeast strain used for fermentation is of great significance in determining the level of higher alcohols in beer. ANN-software analyzer evaluation test demonstrates its high accuracy for MeOH prediction for the strain S. cerevisiae (carlsbergensis) 44 at low oxygen content (Fig. 2). During the fermentation process different fatty acids (C6C10) are synthesized as a product of yeast metabolism and they are excreted into the fermenting wort. They are recognized as common flavor characteristics in both lagers and ales but are more prevalent with lager yeast strains. their concentrations in beer depend on yeast activity and wort composition. The accuracy of the ANN-software analyzer regarding the fatty acid C6 prediction for strain S. cerevisiae (carlsbergensis) 75 at high oxygen content 7.0 mg/dm3 is shown on Fig. 3.
Fig. 1. Diacetil prediction by ANN-software analyser for strain S. cerevisiae (carlsbergensis) 56 at low oxygen content of 0.2 mg/dm3. The best predictors are biomass and limiting substrate content. The analyzer mean square error MSE=9.99539e-005 is reached by the 3-layered NN with 12 log-sigmoid neurons in the first and the second layer and one neuron with the linear transfer function in the output layer. The analyzers accuracy evaluation is shown by the best linear fit A=0.999T+0.00102 and regression coefficient R=1

Fig. 3. Fatty acid C6 prediction for strain S. cerevisiae (carlsbergensis) 75 at high oxygen content of 7.0 mg/dm3. The best predictors are biomass content and free amino nitrogen. The analyzers mean square error MSE=9.3629e-005 is reached by the 3-layered NN with 20 log-sigmoid neurons in the first and 24 log-sigmoid neurons in the second layer and 1 neuron with the linear transfer function in the output layer. The analyzers accuracy evaluation is shown by best linear fit A=1.08T-0.0108 and regression coefficient R=0.97 Fig. 2. High alcohol MeOH prediction by ANN-software analyzer for the strain S. cerevisiae (carlsbergensis) 44 at low oxygen content of 0.2 mg/dm3. The best predictors are limiting substrate content and free amino nitrogen. The analyzer mean square error MSE=9.99187e-05 is reached by the 3-layered ANN with 20 log-sigmoid neurons in the first and 24 log-sigmoid neurons in the second layer and 1 neuron with the linear transfer function in the output layer. The analyzers accuracy evaluation is shown by best linear fit A=0.996T+0.00227 and regression coefficient R=0.999

The main higher alcohols that are present in beer include: MeOH-methanol, PrOH-propanol, BuOH-butanol and AmOHamyl and isoamyl alcohol, and etc. They have a higher boiling point than ethanol and can leave a “fusel” flavor in the beer. 1938

Recent investigations on factors of aging, taste and flavor of beer are oriented to the Total Antioxidant Capacity (TAC) measurement by FRAP (Ferric Reducing/Antioxidant Power), as well as content of glutathione and total phenols analysis during the beer fermentation. Besides the health effects of beer consumption, the predicted TAC by software analyzer can serve as a criterion for selecting appropriate beer brewing strains and as a marker of future quality of the final market product (11). For the design of software analyzers for beer antioxidants by implementing artificial neural networks (ANN) as the Biotechnol. & Biotechnol. eq. 24/2010/3

predictors, data concerning yeast biomass, limiting substrate and α-amine nitrogen (α-N2) concentration are used. Contemporary food biotechnologies devote a special place to creation of waste less, ecologically clean production cycles. For this reason in beer and wine manufacturing fermentation processes wastes- biomass, substrates and separated carbon dioxide are utilized. Useful tools in this case are software analyzers- calibration curves for transformation of the wet biomass weight into dry one (14). Wide applications for biomass specific growth rate prediction have the process state evaluation approaches, one of the most popular among which is the extended Kalman filter. Analytically, on the basis of the transformation formula of sugar, limiting substrate of mostly fermentation processes, the expected quantity of carbon dioxide for industrial needs can be predicted with relatively high accuracy (14). It is well known that biochemical analysis of beer quality factors is time and labor consuming. Introducing knowledgebased software tools in the practice when the technology is established for a long time and will not change is an alternative method to measuring of different parameters and is a lowcost solution. Such software tools are not only profitable but equipped with the software system for fault detection and isolation is very efficient in the automatic control of beer brewing.

Conclusions

Nowadays, it is not a tradition to monitor and control in an on-line manner the key variables in beer brewing processes and the beer quality factors that are presented in our research. However, the increasing demands of the world market of foods and beverages forces the technology to improve its measurement methods through the expert knowledge that is gathered for many centuries from skilled brewers and foodmakers. Computational intelligence-based technology provides a mechanism by which human expertise and learning capacity can be embedded and implemented to solve problems; provide assessment of process from input data; and provide an intelligent control. The combination of both analyzing methods- conventional and modern AI-based software analyzers offer to the producer of high-quality beverages a double assurance for the quality of products.

REFERENCES

1. Aarts R.J. and Rousu J. (1996) Proc. Third EU Workshop, EWCBR-96, Lecture Notes in Artificial Intelligence, 1168, 16-27. 2. Austin G.D., Celotto G., D’Amore T., Syu M.-J., Tsao G.T. (1992) Second BCOA Meeting, St. Louis.

3. Becraft W., Lee P., Newell R. (1991) Proc. of the 12th International Joint Conference on Artificial Intelligence, 832-837. 4. Beil S., Dors M., Havlik I., Kreibaum U., Schneider F., Simutis R., Lubbert A. (1992) Monatschrift für Brauwissenschaft, 45(6), 196-204. 5. Cummins S., Plant N., Kelleher P., O’Connor J.B. (1998) Proc. Intern. Symp. on Automatic Control of Food and Biological Processes, SIK, G?teborg, Sweden, 459467. 6. Gopal C.V., Hammond J.R.M., Pye J.M. (1993) MBAA Technical Quarterly, 30(3), 71-79. 7. Johnson C.R., Burnham K.J., James D.J.G. (1998) Proc. of Intern. Conf. on Simulation, 259-262. 8. Kashihara T., Mawatari M., Inque T., Prior J.J., Cooney C. (1993) Journal of Fermentation and Bioengineering, 76(2), 157-159. 9. Kronlof J. (1994) PhD Thesis, VTT Publications 167, Technical Research Centre of Finland. 10. Linko S., Zhu Y.-H., Linko P. (1995) In: Computer Applications in Biotechnology: A post print volume from the 6th Intl. Conf. on Computer Application in Biotechnology, Garmisch-Partenkirchen, Germany. 11. Mileva S. and Vassileva S. (2008) Proc. Intern. Sc. Conference Computer Science, 2008, Kavala, Greece, Heron Press, 164-169. 12. Montague G. and Morris J. (1994) Trends in Biotechnology, 12(8), 312-324. 13. Narendra K.S. and Parthasarathy K. (1990) ieee Transactions on Neural Networks, 1(1), 4-27. 14. Petrov P., Vassileva S., Mileva S. (2005) Proc. of the Second Intern. Sc. Conf. “Computer science”, Chalkidiki, Greece, 242-248. 15. te Braake H.A.B. (1997) PhD Thesis, Delft University of Technology, Department of Electrical Engineering, Delft, The Netherlands. 16. Vassilleva S., Huong V.L., Votruba J. (1994) Folia Microbiol., 39(6), 1994, 489-492. 17. Vassileva S., S. Mileva (2006). Proc. ECOLOGY’2006, SCIENTIFIC ARTICLES of the Int. Conf. ECOLOGY’2006, Sunny Beach, Bulgaria, 1, 262-271. 18. Vassileva S. and Mileva S. (2008) In: Advanced topics on Neural Networks (L.A. Zadeh, J. Kacprzyk, Honorary Eds., D. Dimitrov, S. Jordanova, V. Mladenov, N. Mastorakis, Eds.), Artificial Intelligence Series Publ., WSEAS Press, ISBN: 978-960-6766-56-5, 48-53. 19. Vassileva S., Mileva S., Bachvarov V.Chr. (2009) Journal ecology & Safety, 3(1), 430-440.

Biotechnol. & Biotechnol. eq. 24/2010/3

1939


相关文章:
更多相关标签: