Determination of the Process Parameters Relative Influence on k L a Value using Taguchi Design Methodology

Transcription

1 J. Ind. Eng. Chem., Vol. 13, No. 7, (2007) Determination of the Process Parameters Relative Influence on k L a Value using Taguchi Design Methodology Marko TRAMŠEK and Andreja GORŠEK University of Maribor, Faculty of Chemistry and Chemical Engineering, SI-2000 Maribor, Slovenia Received January 3, 2007; Accepted November 22, 2007 Abstract: This article describes experimental determination of the relative impact of significant process parameters that influence volumetric oxygen mass transfer coefficient (k L a) using Taguchi design methodology. For this purpose an automated RC1 reaction calorimeter (Mettler-Toledo), which was originally developed for chemical processes, was modified for the bioprocesses. Simple fermentation using Baker s yeast was studied to illustrate the design procedure. Orthogonal array L 25 was selected for the proposed design and ANOVA method was used for recognizing the relative influence of the process parameters. Within the observed range of temperature (ϑ), fermentation media volume (V FM ), and yeast mass concentration (γ Y ), these process parameters were found to be unimportant compared to the volumetric air flow rate (q V,a ) and rotational frequency of the stirrer (f m ). The q V,a had a substantial effect on the k L a value (89.2 %) and the f m had just a small one (3.6 %), meanwhile the remain fraction to 100 % represents error. The results refer strictly to the selected case study. Anyhow, the proposed procedure shows that application of the Taguchi approach for analyzing the oxygen mass transfer based on the experimental data obtained from a highly-automated laboratory reactor appears to have potential usage in general biopharmaceutical process design. Keywords: volumetric oxygen mass transfer coefficient, Taguchi method, process parameters, bioprocess, analysis of variance Introduction 1) The pharmaceutical industry is the professional and business activity containing the most important scientific potential of the humans. The fast development of industrial biotechnology as a sustainable alternative to traditional chemical production has caused significant changes in the pharmaceutical industry [1,2]. For instance, highly molecular biopharmaceuticals (recombinant therapeutic proteins, monoclonal antibody-based products, nucleic acid-based medicinal products) represent as much as one-fourth of all newly developed pharmaceuticals [3]. Due to increasingly demanding regulatory requirements in the global market, their production needs to fully follow the principles of current Good Manufacturing Practice (cgmp) [4]. GMP envisages the use of optimally planned processes ensuring the highest To whom all correspondence should be addressed. ( andreja.gorsek@uni-mb.si) possible product quality, safety and effectiveness. The basic production process of biopharmaceuticals is fermentation [4]. The oxygen mass transfer (OMT) at the gas-liquid interphase is of decisive importance for the rate of the aerobic fermentation course, consequently, the volumetric oxygen mass transfer coefficient, k L a, is considered to be one of the most important factors when planning bioreactors [5,6]. Despite numerous well known theoretical equations for the assessment of its value [7,8] in fermentation processes it is usually determined experimentally [9]. It was established that OMT was influenced by different process parameters [10,11]. In order to provide the optimum performance of the fermentation process, complying with cgmp [12], should also be necessary to determine the relative influence of those different process parameters affecting the k L a value. Traditionally, the influence of parameters on process performance has been determined experimentally through planning and implementing experiments on large

2 Determination of the Process Parameters Relative Influence on k La Value using Taguchi Design Methodology 1055 Figure 1. Block scheme of adjusted RC1 (Bio-RC1). industrial devices. With the technological development, however, the process parameters analyzing has been transferred to laboratory scale, which resulted in increased effectiveness and reduced planning costs. The technique for the determination and investigation of the influential experiment parameters at different levels is called the design of the experiment [13]. The selection of relevant planning techniques depends on the number of parameters influencing the product quality, and the type of problem. However, conventional techniques involve altering of one parameter at a time keeping all other parameters constant. When we want to study any given system with a set of independent variables (process parameter) over a specific region of interest (levels region) and intend to improve the process planning strategy and quality optimization of the process parameters at the same time, we use the so-called Taguchi methodology [13]. The use of its algorithm is observed in various optimization problems, starting with optimization of diesel engine parameters [14], the leaching of non-sulphide zinc ore in the ammonium-sulphate solution [15], to the production of laccase by Pleurotus ostreatus 1804 [16], etc. However, considering available literature, it can be stated that this method was not used for experimental determination of relative impact of the significant process parameters that influence the k L a value. In contrast to the traditional design of experiment (DOE), the standardized Taguchi s methodology for two independent problem solution plans usually brings the same results, which enables determination of individual process parameters relative influence on the final result. This methodology envisages implementation of a minimum number of experiments, which are defined by specific standard orthogonal arrays (OA). Selection of relevant OA is conditioned by the number of parameters and levels. We exclude unnecessary additional experiments by implementing experiments using the technically perfect and computer supported laboratory equipment, thus enabling full repeatability. Among such type of equipment, there is also a highly automated RC1 reaction calorimeter (Mettler Toledo) intended for investigation of temperature changes in basic chemical and physical processes. By using the specific modifications in hardware and software, first introduced by Marison and others [17,18], it could also be used for investigations in bioprocesses [19-21]. This research is aimed at studying the possibility of experimental determining the relative influence of various process parameters on the k L a value in biotechnological processes, by using the Taguchi s process planning methodology. Simple fermentation of glucose with the Baker s yeast (Saccharomyces cerevisiae) in the adjusted RC1 reactor (Bio-RC1) was used as a case study. We investigated the effects of stirrer rotational frequency, f m, volumetric air flow rate, q V,a, fermentation media volume, V FM, temperature, ϑ, and the Baker s yeast mass concentration, γ Y. Materials and Methods Equipment The RC1 reaction calorimeter is a computer-controlled batch laboratory reactor (V = 2 L) for the performance of isothermal and adiabatic reactions, basic physical operations, and the determination of thermal data and constants. Its detailed description can be found in previous work [22]. For the purpose of bioprocess investigations, a hardware modification to the original RC1 was introduced, which was different from the one introduced by Marison and others [17,18]. Thus, RC1 was modified to Bio- RC1. Modifications (air compressor, oxygen electrode) are represented by the two shaded blocks in Figure 1. Bottom valve of the glass reactor was replaced by a separately constructed aeration module, which is connected to the air compressor (Senco; model: PC1010) through the volumetric air flowmeter. This modification enables aeration of the fermentation medium. A dissolved oxygen sensor (Mettler Toledo; lnpro6800/ 12/320) was installed on the glass cover. Through the transmitter (Mettler Toledo; model: O2 4100e) it was connected to an RD10 controller. After this connection, dynamic measurements of the dissolved oxygen concentration were possible using the PC. The optimal mixing regime at the aeration of the fermentation mixture was achieved by replacement of the existing stirrer (Mettler Toledo; model: anchor) with the intermig one (Ekato; model: intermig).

3 1056 Marko TRAMŠEK and Andreja GORŠEK Table 1. ANOVA Equations and Their Relationships Eq. No. Expression Eq. No. Expression (4) (5) (10) (11) (6) (12) (7) (13) (8) (14) (9) Determination of Volumetric Oxygen Mass Transfer Coeficient The k L a in the fermentation system was determined using the dynamic method proposed by Mignone and Ertola [23], which is based on a step change in the f m during cultivation. The method considers the liquid film effect on the diffusion along the membrane of the oxygen electrode, and the effect of oxygen electrode response time resulting from electrode diffusion resistance. Both effects can be neglected by the application of a top-level dissolved oxygen sensor (Mettler Toledo) with quick response, and by assuring effective fermentation media mixing. During aerobic fermentation with constant air flow rate and at a rotational frequency of the stirrer, f m,1, the first steady state is reached: OTR = (1) where: OTR - oxygen transfer rate (g/(l s)) - dissolved oxygen mass concentration in the first steady state (g/l) k L a 1 - volumetric oxygen mass transfer coefficient at f m, 1 (s -1 ) - saturated oxygen mass concentration (g/l) The step change of the stirrer s rotational frequency to f m,2, causes approaching to the second steady state. When defining the dimensionless concentration variable, D γ : (2) then transition from the first to the second steady state can be described as follows [23]: where: - dissolved oxygen mass concentration (g/l) - dissolved oxygen mass concentration in second steady state (g/l) k L a 2 - volumetric oxygen mass transfer coefficient at f m,2 (s -1 ) t - time (s) (3) The oxygen-dissolving kinetics in liquid phase at a rotational frequency of the stirrer, f m,2, is described by Eq. (3). Therefore, the experimental measurements ( f(t)) during the approach to the steady state after the step change of stirrer s rotational frequency from f m,1 to f m,2, enable determination of k L a value at f m,2. Taguchi s Experiment Planning Methodology Taguchi has defined the optimization criterion quality (in our case, namely, k L a) as a consistency in achieving the desired value through minimization of the deviation [13]. This goal is connected with the performance of a series of experiments with different process parameters at different levels. The process parameter is a factor affecting the optimization criterion quality, and its value is called the level. The number of experiments and their sequence are determined by standard OA. When planning the experiments using six process parameters at five levels, we use the OA L 25. Such a plan envisages the performance of 25 experiments, which is significantly less when compared to the traditional factorial methodology with 5 6 = experiments. Due to performing only a part of the envisaged experiments by using the traditional factorial methodology we have to include an analysis of the results confidence. The standard statistical technique is used for this purpose, the so-called analysis of variance (ANOVA), which

4 Determination of the Process Parameters Relative Influence on k La Value using Taguchi Design Methodology 1057 Table 2. Process Parameters and Their Levels Process Parameter Level A: f m/min B: q V,a/(L/min) C: γ Y/(g/L) D: V FM/L E: ϑ/ C recognizes the relative influence of the process parameters for the optimization criterion value. The mathematical algorithm of the ANOVA statistical technique is based on calculation of the variance, which is an indicator of the optimization criterion quality. The ratio between the variance of the process parameter, and the error variance shows, whether the parameter has affected the product s quality. The equations required for calculating the relative influence of the process parameters affecting the optimization criterion are presented in Table 1. The meanings of symbols indicated in Table 1, are described in the chapter Nomenclature. We compare F j to the statistical value F m,n for certain confidence rate, which is obtained from the standard F Tables [13], whereby m stands for the f j and n means the f e, and thus determine the process parameter influence accordingly. In the case where the F j falls below F m,n, the process parameter has no effect on the optimization criterion, therefore, it is pooled and ignored in the calculations. Consequently, the V e changes, as the S j and f j of the pooled process parameter are added to the S e and f e respectively. By using the adjusted V e, we determine new F j and compare them again by the F m,n. The process of pooling is sequential, which means that the parameter having the smallest effect on the optimization criterion should be pooled first, then we recalculate the F j and continue pooling until each process parameter meets the condition F j > F m,n. When the pooling procedure is completed, the X j and X e can be calculated using Eqs. (13) and (14). Chemicals Simple fermented medium water-glucose was used for the research purposes with a mass concentration of 3D-(+) Glucose anhydrous (Fluka), γ G = 5 g/l, purchased from Sigma-Aldrich Co (St. Lous, MO). Fermentation was performed with the Baker s yeast, in the fresh compressed form, obtained from the local grocery store. It was composed by natural yeast (Saccharomyces cerevisiae) and water. The dry matter was more then 30 %. As the shelf life of the fresh compressed yeast is short (t = 45 d), the proper storage conditions, clean and dry place between ϑ = (0-4) o C, were taken into consideration. Experimental Work Determining the relative influence of various process parameters on the k L a in bioprocesses based on the Taguchi s approach, requires the performance of a series of experiments. A Bio-RC1 reactor was involved in the research for this purpose. As a bioprocess test example, we used a simple fermentation of glucose into ethanol by the Baker s yeast. It is well known that the optimization criterion (k L a) would be the most significantly affected by f m, q V,a, γ Y, V FM, and ϑ. We examined the relative influence of the selected process parameters at five different levels, as shown in Table 2. Usually the selection of levels and their ranges depends on how optimization criterion is affected due to different levels settings. However, we defined levels ranges according to optimal process conditions (γ Y and ϑ) and equipment range-capacity (f m, q V,a, V FM ). The boundary levels (1 and 5) were selected first followed by selection of others (2, 3, and 4). During the first stage of the experimental work, it is necessary to prepare the experiment performance plan. The plan envisages determining the number of experiments, their performance conditions, and their sequence. Based on the assumption that the k L a would be affected by five process parameters being considered at five levels, we chose the L 25 array as the most adequate OA requiring the performance of 25 experiments [13]. The OA L 25 is usually intended for the investigation of six process parameters at five levels; however, it may also be used in our case (five parameters at five levels) by ignoring the process parameter F. The experiment performance plan is presented in Table 3. The first column of Table 3 presents the experimental serial number. Each experiment was defined by the process parameters (A, B, C, D, E, and F) marked at specific levels by numbers from 1 to 5. During the second stage of the experimental work, we implemented the proposed plan by performing the experiments in the Bio-RC1 reactor. The performance procedure was the same for all experiments. Individual experiments were implemented by means of first charging the reactor by the fermentation medium quantity determined by the plan and heating it up to the working temperature under the experimental conditions (q V,a and f m ). After establishing the steady state (temperature and saturation by oxygen), we charged the reactor by the Baker s yeast mass defined by the plan and reduced the rotational frequency of the stirrer to f m = 50 min -1. The addition of the yeast caused a launch of the aerobic fermentation and consumption of oxygen. The concentration of the dissolved oxygen decreased until the first steady state was reached; then we increased f m back to the original value. Consequently, there was a second

5 1058 Marko TRAMŠEK and Andreja GORŠEK Table 3. Experiment Performance Plan OA L 25 Experiment Process parameter b A B C D E F a a In our case process parameter F is not considered. b Process parameters are indicated in Table 2. Table 4. Experimental Values of k La OA L 25 Experiment k La s Experiment k La s Experiment k La s steady state reached. The was monitored. Based on the experimental data during the transition from the first to the second steady state, we used the Eq. (3) to calculate the k L a value at the conditions determined by the experiment performance plan. Results and Disscussion The k L a values experimentally determined under different conditions proposed by the experimental performance plan (Table 3), are presented in Table 4. All k L a values are of a similar size rank, as was reported by Badino and coworkers [24]. Thus, we have basically confirmed the relevance of the adjusted RC1 (Bio-RC1) reactor for further investigations the OMT from the gas to the liquid phase in bioprocesses. Table 4 shows the lowest value of k L a = s -1, and the highest one, k L a = s -1. Both values were achieved at the lowest and highest q V,a, respectively, which indicates that the q V,a might be the most important parameter. Moreover, the average effects of the process parameters along with interactions at the assigned levels on the k L a value are shown on Figure 2. The difference between levels of each process parameters indicates their relative influence [16]. The larger the difference, the stronger is the influence. It can be observed from Figure 2 that among process parameters studied, q V,a showed the stronger influence. However, we defined the actual relative influence of the process parameters effects by the ANOVA statistical method. The results of the variance analysis by the ANOVA statistical method are shown in Table 5. The f j and f e equaled (f j = f e = 4) in all cases. At 95 % confidence (level of importance 0.05), the value F 4,4 = was determined through standardized Tables of F-statistics.

6 Determination of the Process Parameters Relative Influence on k La Value using Taguchi Design Methodology 1059 Figure 2. Individual process parameters performance at different levels on k La value. Table 5. Analysis of Variance OA L 25 Process Parameter S j 10 5 f j V j 10 5 F j A: f m B: q V, a C: γ Y D: V FM E: ϑ Error Total Table 6. Analysis of Variance with Pooled Temperature OA L 25 Process Parameter S j 10 5 f j V j 10 5 F j A: f m B: q V,a C: γ Y D: V FM E: ϑ Pooled Error Total Table 5 shows that the F j of the f m, γ Y, V FM, and the ϑ fell below F 4,4. In accordance with the Taguchi s method algorithm, we pooled ϑ from further statistical consideration as the least important process parameter, i.e., with the lowest F j value compared to F 4,4. Pooling of the ϑ as an insignificant process parameter requires a repeated variance analysis, whereby the S j and the f j of the pooled process parameter is added to the S e and the f e, respectively. The results in Table 6 show that, consequently, the F j of the remaining process parameters increase. In spite of this, a repeated comparison of F j indicated in Table 6 with the F-statistics value, F 4,8 = , shows that only q V,a meets the F j > F 4,8 condition. In a similar way, we further sequentially pooled the γ Y and the V FM from the statistical consideration. The final result of the variance analysis is shown in Table 7. By sequential pooling of insignificant process parameters, the f e increased to the value 16. The F j of both remaining process parameters (f m and q V,a ) met the conditionf j > F 4,16 = , therefore, we can use Eq. (13) to calculate the X j. Table 7 shows that the k L a value in a simple fermentation of glucose with the Baker s yeast was the most significantly affected by the q V,a (X = 89.2 %). A substantially smaller share of the effect may be attributed to the f m (X = 3.6 %) while the share of the error effect was accounted for X e = 7.2 %. Results obtained by this study are not directly comparable with some previously known results from the literature [25,26], where f m was established as the process parameter with the highest effect on k L a. Many other equipment setup factors exist (i.e., impeller and aeration configuration, distribution and size of air bubble, mixing delay time, bioreactor characteristics), which are highly specific for each aeration and agitation system, and also have an influence on k L a value. However, these factors

7 1060 Marko TRAMŠEK and Andreja GORŠEK Table 7. Final Results of Variance Analysis OA L 25 Process Parameter S j 10 5 f j V j 10 5 F j X j A: f m B: q V, a C: γ Y Pooled D: V FM Pooled E: ϑ Pooled Error Total were not considered in our case. Nevertheless, a relatively small impact of the f m regarding to q V,a may be explained by the replacement of the existing anchor stirrer by the intermig one providing equally-intensive distribution of small air bubbles within the entire reactor volume at a low rotational frequency of the stirrer (f m = 80 min -1 ). The further increase in the f m within levels range therefore, has not caused considerable increase in the active gasliquid surface; consequently, its share of effect on the k L a value was small. Furthermore, it is wellknown that the impact of ϑ compared to the q V,a on the k L a is small [27]. In our study, within the range of (24-32) o C it has also been found that the effect of the ϑ compared to the f m and the q V,a was irrelevant. A similar statement can be applied for other insignificant process parameters. Conclusion The k L a value, as one of the most important factor implied on the design, control and optimization of bioprocess, is affected by many process parameters. Their influence on k L a value is usually determined by conventional one-parameter-at-a-time techniques, which do not provide any insight into the behavior of the system. Moreover, these procedures are time consuming cumbersome and requires more experimental data. The Taguchi method of experimental design is an immediate replacement of the conventional techniques. After all, with a standard OA as a main part of this method the number of experiments is significantly reduced when compared to the traditional factorial experimental design methodology. Using experimental measurements in the adjusted RC1 (Bio-RC1), we examined the possibility of determining the relative influence of various process parameters on the k L a in bioprocesses by the Taguchi s design methodology. A simple case of glucose fermentation with the Baker s yeast was used as a case study. The effect of the f m, q V,a, V FM, ϑ, and the γ Y at five different levels were considered. The experimentally determined k L a values basically confirm the relevance of the RC1 reactor modification into the Bio-RC1 reactor, and the relevance of its further use in the examinations of OMT from the gas to the liquid phase in bioprocesses. By analysis of the experimental data variance (ANOVA), we established that the relative impacts of the, ϑ, V FM, and the γ Y on the k L a, compared to the q V,a and the f m, were negligibly small or insignificant. The q V,a share of influence on the k L a was 89.2 %, the f m contributed by 3.6 %. However, the results are specific for Bio-RC1 and for described fermentation system. Therefore, they cannot be presented as a general for all bioprocesses and for all equipment configurations. Nevertheless, our study confirms that the Taguchi s planning methodology is appropriate for the quantitative identification of various process parameters impacts on the k L a in bioprocesses using different equipment setup. We have successfully linked the state-of-the-art laboratory equipment, the Taguchi s planning methodology and the bioprocess, thus contributing to an economically and dynamically more effective planning of biopharmaceutical applications. Nomenclature cgmp current Good Manufacturing Practice DOE Design of Experiments D γ dimensionless concentration variable (1) f e degree of freedom of error variance (1) F j variance ratio of process parameter j (1) f j degree of freedom of process parameter j (1) F m,n standardized value from the F Tables at defined level of significance (1) f T total degree of freedom of the result (1) f m rotational frequency of the stirrer (min -1 ) k L a volumetric oxygen mass transfer coefficient (s -1 ) L number of levels (1) M number of process parameters (1) N total number of experiments (1) N k number of experiments on k level (1) OA orthogonal array

8 Determination of the Process Parameters Relative Influence on k La Value using Taguchi Design Methodology 1061 OMT Oxygen Mass Transfer OTR Oxygen Transfer Rate (g/(l s)) q V,a volumetric air flow rate (L/min) S e error sum of squares ( / ) S j sum of squares of process parameter j (/) S T total sum of squares ( / ) t time (s) V e variance error (1) V j mean square (variance) of process parameter j (/) V FM fermentation media volume (L) X e relative influence of error on optimization criterion (%) X j relative influence of process parameter j on optimization criterion (%) Y i i value of optimization criterion ( / ) γ G glucose mass concentration (g/l) γ Y Baker s yeast mass concentration (g/l) dissolved oxygen mass concentration (g/l) saturated oxygen mass concentration (g/l) ϑ temperature ( o C) References 1. M. Gavrilescu and Y. Chisti, Biotechnol. Adv., 23, 47 (2005). 2. A. Rajapakse, N. J. Titchener-Hooker, and S. S. Farid, Comput. Chem. Eng., 29, 1357 (2005). 3. G. Walsh, 2005, Trends Biotechnol., 23, 553 (2005). 4. M. Narhi and K. Nordstrom, Euro. J. Pharm. Biopharm., 59, 397 (2005). 5. S. Hiraoka, Y. Kato, Y. Tada, S. Kai, N. Inoue, and Y. Ukai, J. Chem. Eng. Jpn., 34, 600 (2001). 6. J. A. Williams, Chem. Eng. Prog., 98, 31 (2002). 7. F. Garcia-Ochoa and E. Gomez, Chem. Eng. Sci., 59, 2489 (2004). 8. K. Van t Riet and J. Tramper, Basic bioreactor design, p. 236, Marcel Dekker, New York (1991). 9. M. Tobajas and E. Garcia-Calvo, Heat Mass Transfer, 36, 201 (2000). 10. M. Gavrilescu, R. V. Roman, and V. Efimov, Acta Biotechnol., 13, 59 (1993). 11. R. Lemoine and B. I. Morsi, Int. J. Chem. R. Eng., 3, 35 (2005). 12. M. Thiry and D. Cingolani, Trends Biotechnol, 20, 103 (2002). 13. K. R. Ranjit, A primer on the Taguchi method, Van Nostrand Reinhold, New York (1990). 14. M. Nataraj, V. P. Arunachalam, and N. Dhandapani, Indian. J. Eng. Mater S., 12, 505 (2005). 15. J. Moghaddam, R. Sarraf-Mamoory, Y. Yamini, and M. Abdollahy, Ind. Eng. Chem. Res., 44, 8952 (2005). 16. K. K. Prasad, S. V. Mohan, R. S. Rao, B. R. Pati, and P. N. Sarma, Biochem. Eng. J., 24, 17 (2005). 17. I. Marison, M. Linder, and B. Schenker, Thermochim. Acta, 310, 43 (1998). 18. I. Marison, J. S. Liu, S. Ampuero, U. Von Stockar, and B. Schenker, Thermochim. Acta, 309, 157 (1998). 19. M. Janssen, R. Patino, and U. Von Stockar, Thermochim. Acta, 435, 18 (2005). 20. F. Aulenta, C. Bassani, J. Ligthart, M. Majone, and A. Tilche, Water Res., 36, 1297 (2002). 21. P. Vellanki, J. Guhan, I. W. Marison, J. S. Liu, and K. Jayaraman, Thermochim. Acta, 309, 105 (1998). 22. M. Hvalec, A. Gorsek, and P. Glavic, Acta. Chim. Slov., 51, 245 (2004). 23. C. F. Mignone and R. J. Ertola, J. Chem. Technol. Biotechnol. B., 34, 121 (1984). 24. A. C. Badino, P. I. F. Almeida, and A. J. G. Cruz, Chem. Eng. Educ., 38, 100 (2004). 25. M. S. Puthli, V. K. Rathod, and A. B. Pandit, Biochem. Eng. J., 23, 25 (2005). 26. C. Bandaiphet and P. Prasertsan, Carbohydrate Polymers, 66, 216 (2006). 27. T. Kaskiala, Miner. Eng., 15, 853 (2002).