Reduction of process taste in sugar solution with carbon filtration

Transcription

1 Reduction of process taste in sugar solution with carbon filtration Josefine Cragnell Department of Chemical Engineering, Lund University, P. O. Box 124, SE Lund, Sweden In this study different activated carbon products ability to adsorb impurities in beet sugar solution are examined. The impurities arise in the sugar juice during the refining and cause an unwanted additional flavour and hence the name process taste. Column experiments and isotherm tests are performed in laboratory scale with various activated carbon products. The used carbon product in the plant (Jacobi s AquaSorb H200) is examined more thoroughly in regard to empty-bed residence time and cycle time. All experiments are evaluated with sensory analyses since the identity of the impurities is unknown and consequently no concentrations can be measured. The sugar juice samples which were filtrated with Norit s GCN 830 Plus, Chemviron s Aquacarb 607C and Jacobi s AquaSorb CS appeared to have the lowest process taste. The highest empty-bed residence time tested, 38 minutes, resulted in the lowest process taste. Introduction Nordic Sugar in Arlöv is a sugar refinery which among others produces sugar solutions. One of the sugar solutions has a lower quality compared to the other sugar solutions. This sugar solution is called smakförbättrad lösning (SFL) and can be translated to taste enhanced solution in English. The SFL is produced in a different way and is sold to a lower price. Before SFL is transported to the customers a sensory analysis is performed. In the sensory analysis a panel tastes the SFL. Sometimes the SFL fails in the sensory analysis. It depends on an additional flavour which has been obtained from the refinery process in the plant. To overcome this problem and to raise the quality of the SFL the carbon filter in the plant is investigated. The activated carbon used in the plant is manufactured by the carbon company Jacobi and is called AquaSorb H200. The carbon filter does not remove the impurities satisfactory and it either depends on that the carbon filter is poorly designed or that the activated carbon in the filter has too low affinity and selectivity for the target impurities. Perhaps there are other carbon products that have higher affinity and selectivity for the target impurities. Therefore seven different activated carbons from three different suppliers are examined together with Jacobi s AquaSorb H200. These carbons are mentioned below together with their raw materials. Chemviron s CPG LF, Bituminous coal Chemviron s Aquacarb 607C, Coconut shell Jacobi s AquaSorb CS, Coconut shell Jacobi s AquaSorb H200, Bituminous coal Jacobi s ColorSorb 5000, Lignite Norit s GCN 830 Plus, Coconut shell Norit s ROX 0,8, Coal Norits s GAC 1240 Plus, Coal The empty-bed residence time in the carbon filter can be too short for the applied cycle. Too many impurities then get breakthroughs before the cycle is terminated. An empty-bed residence time of 38 minutes and 19 minutes are therefore compared with the same cycle time. The cycle time in the plant s carbon filter can also be too long for the applied empty-bed residence time. It also leads to too many impurities get breakthroughs. To investigate this, the carbon that has been in the plant for a whole cycle is examined. All experiments are evaluated with sensory analyses since the identity of the impurities is unknown and consequently no concentrations can be measured. Adsorption system A simple adsorption system consist of one adsorbate (one target impurity) a fluid (where the impurity is dissolved) and one adsorbent, for example activated carbon. The fluid enters the column which contains the activated carbon. The impurity adsorbs at first in the beginning of the column bed. The adsorption moves further into the column bed as more contaminated fluid flows into the bed (1). The concentration of the impurity in the fluid transfers ordinarily along the bed like a sigmoid curve, due to internal and external mass transfer resistance and axial dispersion. The concentration of the impurity in the fluid along the bed is often called concentration profile (2). When the adsorption takes place near the end of the bed, the bed is almost saturated and the impurity soon begins to appear in the outflow. The breakthrough starts to occur. Normally the cycle is terminated before or in the beginning of the breakthrough, depending on the purity requirement (1).

2 When a simple adsorption system is evaluated the sample collection begins right before the breakthrough takes place. The fluid at Nordic Sugar consists of a sugar juice that contains many different impurities that together give rise to the process taste. The adsorption cycle is terminated when too many impurities received breakthrough. In this system some impurities get breakthroughs in the beginning of the cycle and some later and some in the end of the cycle. The breakthroughs occur in other words continuously during the cycle. For this reason the sample collection in this study starts from the beginning of the cycle. Activated carbon Activated carbon is a porous adsorbent and the pore size distribution is typically trimodal (3). The large micro pore volume in the activated carbon decides almost completely the capacity since it gives rise to the large surface area (4). The surface area in commercial activated carbon is often between m 2 /g (5). The huge surface area and its relative low price makes activated carbon the most commonly applied adsorbent (6). The surface of activated carbon is non-polar or slightly polar. The slightly polar nature depends on oxidized functional groups on the surface and inorganic minerals in the carbon (4). The interaction between the impurities and the surface of carbon occur mainly through London forces and close-range repulsion (6). The selectivity of the impurities on the adsorbent is due to their isotherms or their kinetics. The isotherm competition is favoured when there is a large difference between the impurities onecomponent-isotherms on the adsorbent (4). On activated carbon the isotherm mainly depends on the molar mass of the impurity, polarity of the impurity and solubility of impurity in the fluid. An impurity with larger molar mass has a superior affinity than an impurity with smaller molar mass, with the understanding that they have approximately the same solubility in the fluid (6). When a porous adsorbent is used, such as activated carbon, the kinetic selectivity can be significant (3). The kinetic selectivity occurs when an impurity adsorb faster than the other. Generally this happens as a result of a higher pore diffusion coefficient. This effect becomes significant when the micro pore size is comparable to the dimension of the adsobate. An extreme case of kinetic selectivity is when the adsorbate cannot enter the pores, which is called steric hindrance (7). Scaling down In most cases when adsorption is applied the impurities have favourable isotherms. That is when mass adsorbate per mass adsorbent is higher than mass adsorbate per volume fluid, during equilibrium. If the bed is furthermore sufficient deep, as it normally is in the industry, a constant pattern is reached (8) (9). At constant pattern the shape of the concentration profile is constant as it travels along the bed (2). In this situation the velocity in the carbon filter of the plant should only be considered and kept in the scaling down process since the shape of the concentration profile does not change with the bed length. This is with the understanding that the shorter bed (laboratory bed or pilot bed) is deep enough to develop the constant pattern (9) (10). In this study the velocity is not maintained under the scaling down process. Instead the empty-bed residence time is maintained, although it is reasonable to assume that the constant pattern appears in the laboratory column. The reason for this inconvenience is that the carbon suppliers have given Nordic Sugar incorrect instructions and since it was not possible to convince them, the scaling down process had to be performed in that way. Even though this is incorrect, it is possible to compare the various carbon products since the experiments are performed under the same conditions. However, it is difficult to determine if the result had been the same at the correct conditions. In the plant the empty-bed residence time of the carbon filter is 19,2 minutes and the cycle time is 200 hours. If the velocity in the plant was to be maintained the empty-bed residence time had been 2,26 minutes and cycle time had been 23,5 hours. Sensory analyses There are two different sensory methods, analytical test and consumer test (11). The aim of the sensory analysis is to analyse the samples and not to get the consumers opinions about the products. Therefore analytical test is used. Two different types of analytical tests are often mentioned in the literature, the difference test and the descriptive analysis (12). In this study several samples will be compared and since no scale has been elaborated the difference test is chosen. In the difference test, it is evaluated whether there are differences between the samples or not. The most common difference test is duo test and triangle test. The panel that is used in the difference test is called discrimination panel. The members of the discrimination panel must have a special sensitivity for the requested sensory property (11). The panel that is used in this study is Nordic Sugar s own. The persons in the panel have been selected after they have undergone extensive tests. The tests investigated among others the persons sensitivity for the process taste.

3 The difference tests used in this study are duo test and rank sum test. In the rank sum test the panel is asked to rank the samples in respect to which sample that taste most like a sugar solution. In the duo test two samples are compared in regard to the same sensory property. During the performance of the rank sum test it is important to make sure that not too many samples are compared at the same time. This depends on that the taste buds become saturated and the risk that the panel members lose motivation increase. This also concerns duo-test where a maximum of three duo tests are applied during the same test session (12). Statistics Rank sum test In this study Friedman s test is used to examine if the samples in the rank sum test come from the same population. Friedman s test is a non-parametric statistical evaluation and is widely used in rank sum tests within the sensory analysis. In the test the parameter T is calculated according to the equation below. The samples are assumed to belong to a chisquare distribution. The T-value is compared to the critical tabulated chi-square value which is read from the number of degrees of freedom, k-1, and significance level. When the T-value is larger than or equal to the critical tabulated chi-square value the null hypothesis can be rejected at that significance level. In Friedman s test the null hypothesis is formulated as below. H 0 =The samples come from the same population i.e. theirs medians represent the same population. H 1 = At least two samples represent populations with different medians. In order to get the null hypothesis rejected it is enough that two of the samples do not come from the same population (13). When the null hypothesis in Friedman s test is rejected it is desirable to investigate which of samples that differ significant. To answer this question the Fishers LSD is calculated according to the equation below. If the difference between the sums of rank for two samples is more than the LSD-value the samples differ significant from each other (14). Duo test In this study one-sided duo tests are performed. In one-sided duo tests the number of panellists who thinks that A (A is the sample with most ) tastes most like a sugar solution is calculated and then the total number of judges is also calculated. To determine if A tastes significantly more like a sugar solution than B a table which is calculated from the binomial theorem is used. In the table the critical value at a specific level of significance and total number of judges are read. If the number of judges that choose A correspond to the critical value or if the number of judges that choose A is more, A is significantly more like a sugar solution than B at that level of significance (11). Serving The sensory samples are served to the panel after they have been ph adjusted with 5 M NaOH solution to ph 6,8 and diluted with tap water to Rt The temperature of the samples is 20 C and the sample volume is the same for all the samples. The samples are served in random order to erase the results which depend on the serving order. Material The input material, the beet sugar juice, is the same for all subsequent experiments. The sugar juice is collected after an ion exchanger and before the carbon filters in the operating line. In order to get a representative material the collection is performed by extracting different fractions from the stream. Every fraction is extracted in regard to how many tons that have passed through the ion exchanger. The extracted sugar juice is stored in a large container with a capacity of one cubic meter. In order to not let the sucrose undergo hydrolysis during the storage it is necessary to raise the ph from 3,2 to 4,85 since acid catalyses the reaction. Below the irreversible hydrolysis reaction is specified. As can be seen the sucrose breaks down into fructose and glucose. The experiments The ideal approach would be to analyse each activated carbon for a full column cycle. The column tests would then have to be performed in parallel since a cycle is 200 hours (when the residence time is maintained). Unfortunately there were not so many columns available which made it impossible. Instead a column test was made with one carbon product at a time and the tests were terminated after an hour of trials. The reason for the short test time 1 Refractometric Dry Substance

4 (cycle time) was that the taste of the samples was likely to change if they were stored too long in the refrigerator. In order to evaluate the different carbon products further isotherm tests were performed on all activated carbons. All experiments, column tests and isotherm tests, were carried out at 60 C. Granulated activated carbon was used in the column tests and the ph of the input material was 4,85. In the isotherm tests powdered carbon was used. The powdered carbon products had been grinded by respectively carbon supplier so that 95 % of the carbon mass went through a 325 mesh sieve. In the first isotherm test below the input material was ph adjusted to 3,2 with 5 M HCl. In the other isotherm tests the input materials were ph adjusted to 3,5 with 5 M HCl. Column tests In all column tests the experimental setup revealed below was used. The operating conditions were according to test 1 in table 1 when used and unused Jacobi s AquaSorb H200 were compared. The samples were evaluated in sensory duo tests. Isotherm tests In order to ensure that the equilibrium was reached between the impurities on the adsorbent and the impurities in the sugar juice an evaluation of the residence time was done with AquaSorb H200. Three different residence times were compared, four hours, six hours and eight hours. The samples were evaluated in a sensory rank sum test. Below the operating conditions for the three samples are given Table 2. The test conditions are given. Sampl e Residence time (h) Carbon dose (g/l) ph Sample volume (L) , , ,2 1 After the establishment of the residence time the carbon dosage was analysed. Three different carbon dosages were tested, 20 g/l, 40 g/l and 60 g/l. In the table below the operating conditions are given. The samples were compared in a sensory rank sum test. Table 3. The test conditions are given. Sampl e Residence time (h) Carbon dose (g/l) ph Sample volume (L) , , ,5 1 Figure 1. A schematic picture of the experimental setup is shown above. The various carbon products that were compared in the column tests were Chemviron s Aquacarb 607 C, Chemviron s CPG LF, Jacobi s AquaSorb CS, Jacobi s ColorSorb 5000, Norit s GCN 830 Plus and Norit s ROX 0,8. Test one in the table below specifies the operating conditions of the column tests when the carbon products were tested. The samples were compared in sensory duo tests. In the evaluation of AquaSorb H200 the residence time was varied by varying the flow. The test conditions is revealed in the table below. The obtained samples were compared in sensory duo tests. Table 1. The test conditions are given. Test Residence time (min) Flow (ml/min) Bed height (cm) Bed area m 2 Velocity (m/s) 1 19,2 31, , , ,4 15, , , The various carbon products were then examined with the established residence time and carbon dosage. The carbon products which were tested were Chemviron s Aquacarb 607 C, Chemviron s CPG LF, Jacobi s AquaSorb H200, Jacobi s AquaSorb CS, Jacobi s ColorSorb 5000, Norit s GAC 1240 Plus, Norit s GCN 830 Plus and Norit s ROX 0,8. The operating conditions in the isotherm tests were the same for all carbon products and they are given below. The samples were compared in rank sum tests. Table 4. The test conditions are given. Residence Carbon dose ph Sample volume time (h) (g/l) (L) ,5 1 Result Column tests The samples collected from the carbon filtration with different activated carbons were evaluated by comparing the samples in pairs (duo test). The result showed that the sample which had been filtrated with Chemviron s Aquacarb 607C tasted significant

5 more like a sugar solution than the sample filtered with Jacobi s ColorSorb The sample filtered with Jacobi s AquaSorb CS tasted significant more like a sugar solution than the sample filtered with Norit s ROX 0,8. Even the sample filtered with Norit s GCN 830 Plus tasted significant more like a sugar solution than the sample filtered with Chemviron s CPG LF. The results were obtained by calculating the number of for the sample with the most. The critical value was then read in a table. Below a summation is shown of the sensory analysis. Table 5. The compilation of the sensory analysis and colour analysis are shown. The number of refers to the number of judges who thought the sample tasted more like a sugar solution. Activated carbon Test occasio n The number of Mean value Colour (IE) Chemviron s ,5 52,8 Aquacarb 607 C Chemviron s CPG LF ,8 16,7 Jacobi s AquaSorb CS ,8 57,6 Jacobi s ColorSorb 3 8 2,7 19, Norit s GCN 830 Plus ,7 53,7 Norit s ROX 0, ,5 28,4 Below the sensory results are revealed from the column tests then new and used AquaSorb H200 were compared and when AquaSorb H200 with two different residence times were compared. Table 6. The compilation of the sensory analysis and colour analysis are shown. Duo test, Which sample has the highest process taste? AquaSorb H200 New V.S AquaSorb H200 used, T=19,2 Number of Colour (IE) AquaSorb H200 New 7 38,9 AquaSorb H200 used 12 68,6 Total 19 No significance at 5 % significance level. It requires 14 identical answers for 19 judgments to obtain a significant result at 5 % significance level. AquaSorb H200 T=19,2 V.S Number of Colour AquaSorb H200 T=38,4 AquaSorb H200 T=19, ,9 AquaSorb H200 T=38,4 6 22,9 Total 19 No significance at 5 % significance level. It requires 14 identical answers for 19 judgments to obtain a significant result at 5 % significance level. Isotherm tests In the evaluation of the residence time a sensory rank sum test was performed. The Friedman s parameter T was calculated to 3,25 and the critical tabulated chi-square value for 2 degrees of freedom and 5 % significant level was read to 5,99. The null hypothesis could not be rejected since 3,25 are less than 5,99. Therefore, there was no significant difference between the three residence times. Table 7. The compilation of the sensory analysis is given below. Number of people in the sensory analysis that think Invert quota the sample tastes most like a sugar solution. Number 1gives most like and number 3 gives less h ,614 6h ,274 8h ,737 A rank sum test was performed in the carbon dosage evaluation. The result showed no significance since Friedman s parameter T was 0,33 which was lower than the critical tabulated chisquare value for 2 degrees of freedom and 5 % significant level, 5,99. Below a summary of the sensory analysis is shown. Table 8. A summary of the sensory analysis and the invert quota are given. Number of people in the sensory analysis that think the sample tastes most like a sugar solution. Number 1gives most like and number 3 gives less. Invert quota g/l ,75 40 g/l ,25 60 g/l ,90 Not filtered 2,55 The isotherm tests with different carbon products were analysed with rank sum tests and one duo test. All carbon products could not be compared in one rank sum test and it was therefore necessary to make several rank sum tests. Below a compilation of those can be seen. Table 9. A compilation of the sensory analysis is given. The question asked was which sample tasted most like a sugar solution. Carbon Test occasion The numbe r of Mean value Chemviron s X Aquacarb 607 C Chemviron s CPG LF X Jacobi s AquaSorb H200, 60 g/l Jacobi s AquaSorb H200, 40 g/l Jacobi s AquaSorb CS Norit s GCN 830 Plus ,5 X Norit s GAC 1240 Plus Norit s ROX 0, The result showed that the sample filtered with Norit s GCN 830 Plus tasted significant more like a sugar solution than the samples filtered with Jacobi s AquaSorb CS, Norit s GAC 1240 Plus and Norit s ROX 0,8. AquaSorb H200, 60 g/l, tasted significant more like a sugar solution than the sample filtered with Norit s ROX 0,8. Best

6 Norit s ROX 0,8 tasted significant lesser like a sugar solution than the samples filtered with Chemviron s 607C, Norit s GCN 830 Plus and Chemviron s CPG LF. Jacobi s AquaSorb H200, 40 g/l, tasted significant lesser like a sugar solution than Norit s GCN 830 Plus, Chemviron s 607C and Chemviron s CPG LF. Table 10. The compilation of the sensory analysis is given. Which sample taste most like a sugar solution? Number of Jacobi s ColorSorb 5000, 60 g/l 5 Norit s GAC 1240 Plus, 60 g/l 0 No significance The sample that had been filtered with Jacobi s ColorSorb 5000 did not taste significant more like a sugar solution than the sample filtered with Norit s GAC 1240 Plus, see table above. Discussion Column test with AquaSorb H200 When the empty-bed residence time was varied in the column tests with AquaSorb H200 the higher residence time resulted in a lower process taste. The colour analysis also showed that the carbon filtration with higher residence time resulted in a higher colour reduction. This result is consistent with the theory, that a higher residence time results in a later breakthrough and hence fewer substances get breakthrough during the cycle (with the same cycle time). Therefore fewer impurities are present in the sample and a lower process taste is obtained. The result was not significantly, it fell on one vote. However, it is possible to draw a conclusion since the result was relatively unequivocal since 13 out of 19 were of the same opinion. It is possible with great certainty to assume that the process taste can be reduced, with the carbon being used today, by increasing the residence time when the same cycle time is used. In the comparison between new carbon and used carbon the result was that the sample which had been filtrated with new carbon had almost a significantly lower process taste than the sample that had been filtrated with used carbon. The result is reasonable since the new carbon is free from impurities in the beginning of the experiment and has therefore a higher capacity left. The result showed that the used carbon had lower capacity compared to the new carbon. The result also showed that the used carbon had little capacity left since seven out of nineteen said that the juice that had been filtrated with used carbon had a lower process taste than the new carbon. Isotherm test In the evaluation of the residence time in the isotherm test most thought that the sample that had been filtrated with six hours residence time tasted more like a sugar solution. Six hours residence time will therefore be used. The result was not significant which means that the result may be due to chance. It was therefore not certain that six hours residence time was enough to reach equilibrium. The invert quota of the input material was approximately 2,5 and during the experiment the invert quota increased. In order to avoid a high invert quota in the future the ph of the juice was raised to 3,5. Three different carbon dosages were then examined with six hours residence time. The result from the sensory analysis was obscurely because the samples were perceived as very similar. The aim of the isotherm tests was not to find the carbon which adsorbs the most impurities with the lowest carbon dosage. Instead it was to find the carbon that adsorbs most impurities regardless to carbon dosage. The highest carbon dosage was therefore selected, 60 g/l. In addition, the highest carbon dosage resulted in a lower invert which was probably due to the higher adsorption of glucose and fructose. Comparison between the various carbon products In the column tests the juice samples that had been filtrated with Chemviron s Aquacarb 607 C, Jacobi s AquaSorb CS and Norit s GCN 830 Plus got a lower process taste. All of these three carbons are made of coconut shell. The result depends on that these carbon products have a greater proportion of micro pores than the other carbons. The smaller molecules diffuse faster into these than the larger molecules and some cannot enter. In this way, molecules with small molar mass can compete with molecules that have larger molar mass. The result is very distinct since the carbons which reduce the process taste the most reduce the colour worse. Unfortunately Jacobi s AquaSorb H200 was not tested in this experiment since it was out of stock. It is therefore impossible to determine how AquaSorb H200 is in comparison to the other carbons in the column tests. In the isotherm tests the residence time is assumed to be sufficient in order to reach the equilibrium. Since the time needed to reach equilibrium is applied the kinetics is not important. In general, the equilibrium should benefit the molecules with large molar mass because they have a higher affinity and therefore a more favourable isotherm. However, there are times when molecules with larger molar mass cannot enter the micro pores. This is to the benefit of small molecules. The results of this test showed that the samples which had been filtrated with Norit s GCN 830 Plus, Chemviron s Aquacarb 607 C and Chemviron s CPG LF tasted

7 most like a sugar solution. Jacobi s AquaSorb H200 was included in this test and resulted in a higher process taste with same carbon dosage. Many members of the panel said after the sensory analysis that the filtrated sugar juice tasted very similar to a sugar solution. They were referring to the three best samples in the isotherm tests. This comment has never been heard before. The result indicates that the equilibrium capacity provides the best separation of the undesirable taste molecules. Another important comment is that an increasing in the ph of a fluid that is filtrated with activated carbon normally reduces the adsorption of the impurities. This probably also affects the result since all the experiments were not performed at the same ph. It is probably possible to increase the adsorption of the impurities in column tests by decrease the ph in the sugar juice. Conclusion The following conclusions can be drawn from the results: The carbons Norit s GCN 830 Plus and Chemviron s Aquacarb 607 C have a high selectivity of the target impurities in the column tests because of their equilibrium isotherms. Jacobi s AquaSorb CS has probably a high selectivity of the target impurities in the column tests due to the kinetics. Chemviron s CPG LF has an equilibrium capacity which benefits the flavour molecules while the kinetics benefits the colour molecules. The process taste can be reduced by increasing the empty-bed residence time when the same cycle time is used. Future work It would have been desirable to evaluate the carbons which resulted in the lowest process taste more profound, and then compare them with the currently used. The column tests had then been performed over an entire cycle which is 23,5 hours when the scaling down process is done properly. The ph of the sugar juice had also been adjusted to 3,5. Nomenclature = Number of treatments, number of samples and number of columns. = Least significant difference. = Number of blocks, number of persons and number of rows. = The sum of rank numbers for each treatment = Friedman s parameter. = The quantile of t- distribution. Literature cited 1. Rodrigues, Alírio E, Levan, M. Douglas and Tondeur, Daniel. Adsorption:Science and Technology. Netherlands : Kluwer Academic, Ruthven, Douglas M. Principles of adsorption and adsorption processes. USA : John Wiley & Sons, Inc, Ruthven, Douglas M. Adsorption, fundamentals. Kirk Othmer Encyclopedia of chemical technology. s.l. : John Wiley & Sons, Inc., Mersmann, Alfons. Adsorption. Ullmann s Encyclopedia of Industrial Chemistry. s.l. : John Wiley & Sons Inc, Norit. Norit. [Online] [Cited: 12 02, 2009.] enucat=introduction. 6. Yang, Ralph T. Fundamental Factors for Designing Adsorbent,. Adsorbents: Fundamentals and Applications. s.l. : John Wiley & Sons, Inc, Thomas, John and Crittenden, Barry. Adsorption Technology & Design. s.l. : Elsevier Ltd, Coppola, Albert P and M., Levan Douglas. Adsorption with axial diffusion in deep beds. Chemical engineering Science. 1981, Vol O.Cooney, David. Adsorption design for wastewater treatment. Florida : CRC Press LLC, Svedberg, U.G. Simulation and scale-up of adsorption columns.computers & Chemical Engineering.1979, Vol Lundgren, Birgit. Handbok i sensorisk analys. Sverige : Svenska Livsmedelsinstitutet, Lawless, Harry T and Heymann, Hildegarde. Sensory evaluation of food, principles and practices. New York : Chapman and Hall, Sheskin, David J. Handbook of parametric and nonparametric statistical procedures. s.l. : Chapman & Hall, Civille, Gail Vance, Carr, B. Thomas and Meilgaard, Morten. Sensory evaluation techniques. s.l. : CRC Press, Received for review August 27, 2010