You are on page 1of 18

A RESPONSE SURFACE METHODOLOGY APPROACH IN THE OPTIMIZATION OF A COFFEE ROASTING PROCESS Ruel M. Mojica11, E !a" o #. Ca$a$%&e$$ie C. Elau"ia' , , E(!

el)e"* +. Pe"al*a, a( Ma"il-( C. Elau"ia.

A/STRACT Optimization is a vital tool in agricultural and food engineering for the efficient operation of processing systems and unit processes yielding a highly acceptable product. The response surface methodology approach is a statistical technique that was used by authors in most studies involving optimization. The same methodology was used to optimize several unit operations in agricultural process engineering using fractional factorial experimental designs. The coffee roasting process operation was optimized with parameters affecting the process as independent variables while the responses, designated as dependent variables. An analysis of variance (A O!A" was performed to determine which of the parameters significantly affected the independent variables. #econd order polynomial (#O$" models as functions of the independent variables generated response surface and contour plots. Optimum regions were derived by superimposition of the contour plots based on the industry standards. Optimum operating conditions of %&'.(&) temperature, %& min time of roasting and *+, initial moisture content of green beans were established.
#

#eparate validation experiments conducted at optimum conditions verified predictions and adequacy of the #O$ models. -eywords. optimization 1. %. '. coffee roasting,

INTRODUCTION

Coffee, both as beverage and as a plant originates from northeastern Africa; the plant is a woody perennial evergreen and produce mainly economically in developing countries. Coffee beans are initially processed by removing the outer layer of the fleshy pulp that may be accomplished by a dry or a wet procedure. The wet (or washing) process is the more complex and time consuming procedure but leads generally to a higher uality of final product ( !art and "ursten, #$%&'() and *elit+, #$$$)',) . -reen coffee beans cannot be consumed as such but need to undergo a process of roasting that is essential in the formation of flavor and aroma. The different degrees of roasting (light, medium, medium. light, dar/, medium dar/, and very

Assistant $rofessor /// and )hair, 0epartment of Agricultural and 1ood 2ngineering, )avite #tate 3niversity, /ndang, )avite. 24mail. ruelmo5ica6yahoo.com7

dar/) produce different coffee aroma profiles and thus, a variety of coffee beverages (0odello et. al. ,11&)'%). 2tatistics is a science dealing with the analysis of experimental data or information and drawing conclusions from such obtained information. 3nformation or data are only beneficial when it can be examined, analy+ed and findings can be recommended for intended users. 2tatistics is then considered as a universally accepted essential tool for all types of research and resulted in diversified statistical procedures (-ome+ and -ome+, #$%() '&) for different applications. 4ne of the statistical procedures developed was the response surface methodology (520). 5apid food engineering operations such as roasting reduces the overall cost of operation or processing. "owever, adverse effects happen to biological products being roasted. These are development of off flavors, undesirable colors, volatili+ation of flavor compounds and loss of essential vitamins and amino acids. 4ptimi+ation of the roasting process is performed to ensure rapid processing conditions yielding an acceptable uality product and a high throughput capacity. 6or instance, the coffee uality aspects may include final moisture content as a measure of flavor, aroma, body and degree of roast which are important uality factors of roasted coffee. 4n the other hand, process parameters to be optimi+ed may include7 temperature, time, initial moisture content, speed of the auger and many other related criteria. 520 is a statistical procedure fre uently used for optimi+ation studies. 3t uses uantitative data from

an appropriate experimental design to determine and simultaneously solve multivariate problems. The e uations describe the effect of the test variables on the responses, determine interrelationships among test variables and represent the combined effect of all test variables in the response. This approach enables an experimenter to ma/e efficient exploration of a process or system. The coffee roasting process consists primarily of cleaning, roasting, cooling, grinding, and pac/aging operations. *ags of green coffee beans are hand or machine 8opened, dumped into a hopper, and screened to remove debris. The green beans are then weighed and transferred by belt or pneumatic conveyors to storage hoppers. 6rom the storage hoppers, the green beans arfe conveyed to the roaster operating at temperatures between 9:1o and &(1oC and the beans roasted for a period of time ranging from a few minutes to about 91 minutes. 5oasters are typically hori+ontal rotating drums that tumble the green coffee beans in a current of hot combustion gases; the roasters operate in either batch or continous modes and can be indirect.or direct. fired. 3ndirect.fied roasters are roasters in which the burner flame does not contact the coffee beans, although the combustion gases from the burner do contact the beans. !irect.fired roasters contact the beans with the burner flame and combustion gases. At the end of the roasting cycle, water sprays are used to ; uench< the beans. 6ollowing the roasting, the beans are cooled and run through a ;destoner<. !estoners are air classifiers that remove stones, metal fragments, and other waste not removed

during initial screening fomr the beans. The stoners pneumatically convey the beans to a hopper, where the beans are stabili+ed and dry (small amounts of water from uenching exist on the surfacethe surface of the beans. This stabili+ation process is called e uilibration. 6ollowing this, the roasted beans are broundground, usuall byusually by multi.stage grinders. 2ome roasted beans are pac/aged and shipped as hwolewhole beans. 6inally, the ground coffee is vacuum sealed and shipped. Additional operations associated with processing green coffee beans include decaffeination and instant (soluble) coffee production. !ecaffeination is the process of extracting caffeine from green coffee beans prior to roasting. The most common method ofmethod of decaffeination process used in the =2 is supercritical carbon dioxide (C4,) extraction. 3n this process, moistened green coffee beans are contacted with large uantities of supercritical C4, pressure maintained at about (111 >23 and temperatures between $1o and #11oC that removes about $:? of the caffeine from the beans. The caffeine is then recovered from C4, using activated carbon adsorption system. Another method is commonly by solvent extraction method, using oil (extracted from roasted coffee) or by ethyl acetate as a solvent. 3n this process, solvent is added to moistened green coffee beans to extract most of the caffeine from the beans. After the beans are removed from the solvent, they are steam.stripped to remove any residual solvent. The caffeine is then recovered from the solvent, and solvent is re.used. !ecaffeinated coffee beans

have a residual caffeine content of about 1.# percent on a dry basis. @ot all facilities have decaffeination operations and decaffeinated green coffee beans are purchased by many facilities that produce decaffeinated coffee. 6or example in the optimi+ation of the roasting process, the roasting temperature, roasting time and to some extent initial moisture content of green beans are considered important factors affecting the uality of roasted beans. The following responses can be investigated7 flavor, aroma, bodybody, and aftertaste. A three.level five. parameter experimental design can be used to evaluate the optimum roasting process conditions. The number of functions is dependent on the number of response variables. 3n the following case, three mathematical functions of / are assumed to exist for A/7 A/ B / (T, t, 0C) C n '#) where T for instance is the roasting temperature, t is the time of roasting and 0C is the initial moisture content of green beans. 520 was employed by several authors including !ilidili (,11#) and 0adamba, #$$:'D) (,11,)'D) to optimi+e unit operations resulting in acceptable responses. 4ther authors used 520 in optimi+ation studies for biotechnology, drug preparation, ultra.filtration, and surveying. The obEective of the paper was to determine the best operating conditions for automatic mechanical coffee roaster developed for small farmers application of the 520 approach as a statistical techni ue in and optimi+e the ing coffee roasting process.

,. MATERIALS AND METHODS %.1 T0e Co11ee Roa$*i(! Mac0i(e The coffee roasting machine used was the automatic mechanical roaster developed by !r. 5uel 0oEica, ,119'::) which is intended for small scale farmers. initial wor/ on the design of the machine started with gathering relevant outputs of, and other information related to, existing mechanical roasters. =pon incorporating all necessary information, conceptuali+ation was made. As conceptuali+ed, the prototype machine The roaster (6igure #) consisted of the following parts7 (a) hopper for loading of green beans; (b) roasting chamber where actual roasting too/ place; (c) casing assembly; (d) heater; (e) insulation; (f) frame support; (g) electric motor support; (h) power transmission assembly; and (i) discharge outlet (6igure #). The roaster is made of initial concept on the materials to be used was conceived to be made up of purely stainless steel . "owever, the price of the stainless steel suggested that other parts of the machine except for the roasting chamber and the auger which are should be made up of -.3. sheet (gauge F#D). 2uch thought on the design process, nevertheless, decided to be done away without hampering the desired operation of the machine.The roastedr is operatesd with the usea of microcontroller to maintain and control the temperature setting inside the rotating drum and the time of

roasting. The microcontroller served as the ;heart< in the operation of the machine. 3t controlled the time of operation and the re uired temperature inside the roasting chamber. 4ne of the significant factors that affect the roasting process was the temperature inside the roasting chamber (0oEica and >eralta, ,119). The actual roasting temperature must be carefully controlled to achieve a good uality of roasted product. To get into this, an eClectric heater was serves used foras as a stable source of heat energy duringfor roasting. The electric heater supplied the necessary heat re uired in the roasting process. 2ince it was important that the temperature inside the roasting chamber be maintained at a desired level, a microcontroller circuit was incorporated in the design. The microcontroller served as the ;heart< in the operation of the machine. 3t controlled the time of operation and the re uired temperature inside the roasting chamber. 4nce the roasting process was completed, the beans must be immediately moved from the chamber to avoid production of burnt beans. This was the very reason why most of the roasting machines used in large scalelarge.scale operation had its cooling e uipment attached to the machine. To facilitate the unloading operation, a circular opening at the bottom of the roasting chamber was provided. The cap was provided with threads to ensure safety while the machine was in operation. %.% T0e Re$2o($e Su"1ace So1*3a"e

&

>54C 525C(response surface regression) of 2tatistical Analysis 2ystem, 2A2 v.%.# (,111,11#) was used for optimi+ation studies in of coffee roasting. C uation ',) was used for the model of 525Cwhich525C- that also draws the contour as well as the surface plots produced. Contour plots of the responses can also be drawn using the more user.friendly 2tatistica for Gindows Her. :.1.

moisture content, ? dry basis). Corresponding levels were designated .#, 1, and J#, 1 being the center point. Table , presents the independent variables as they were coded in the optimi+ation study. The corresponding values of K#, K, and K9 were based on the results of the preliminary test runs.

Table #. A *ox and *ehn/en design matrix for an incomplete three.factor factorial experimental design.
/O6 AND /EHN+EN RUN NUM/ERS # , 9 ( & D : % $ #1 ## #, #9 #( #& 61 J# J# .# .# 1 J# J# .# .# 1 1 1 1 1 1 6% J# .# J# .# 1 1 1 1 1 1 J# J# .# .# 1 6' 1 1 1 1 1 J# .# J# .# 1 J# .# J# .# 1

6igure #. microprocessor controlled 6igure #. The microcontroller.based mechanical coffee coffee roaster roaster. %.' T0e E42e"i5e(*al A(al-$i$ 3n the analysis of the roasting process, a fractional three.level.three.factor factorial experiment was used to establish the optimum operating conditions of the machine. Table # shows the design matrix developed using the 0esign umber * of *ox and *ehn/en (#$D1) ',). The design had #& runs as contrasted with the ,: runs which a 9 x 9 x 9 complete factorial design would have. 3ndependent variables were coded as I# (roasting temperature, 1C), I, (roasting time, min), and I9 (initial

Table ,. 3ndependent variables used in the optimi+ation study.


INDEPENDENT #ARIA/LES 5oasting temperature (1C) 5oasting time (min) 3nitial moisture content (?) CODED #ARIA/LES S-5)ol Co e I# I, I9 71 K# K# K# 8 K, K, K, 91 K9 K9 K9

%., O2*i5i:a*io( o1 *0e Roa$*i(! P"oce$$ 5esponse 2urface 0ethodology (520) is essentially a particular set of mathematical and statistical methods used by researchers to aid in the solution of certain types of problems which are pertinent to scientific and

engineering processes. The procedures are a collection of experimental strategy, mathematical methods, and statistical inference which when combined, enable the experimenter to ma/e an efficient empirical exploration of the system in which he is interested. =sing the 520 one can (#) find a suitable approximating function for the purpose of predicting future response; and (,) determine what values of the independent variables are optimum values as far as response is concerned (0yers, #$:#) ':). 3n optimi+ing the roasting process with 520 as a tool, several assumptions were considered7 #. That three input factors (I#, I,, I9) were closely controlled during the entire observation process and were measured with negligible error; ,. That the dependent variables (A#, A,, A9, A() defined the system and were experimentally measured; 9. That there were three mathematical functions, f/ (/B#,L,&) that described the relationship between responses, A/ and the factors Ii that were ta/en to be both uantitative and continuous, such as 89 : f9 (;*, ;%, <,;+"7 (. That the functions could be approximated by second.order polynomial e uations in the form of7
9 9 , 9 i =# i =# i =# 5 =i +#

Ii were coded independent variables; and A/ were the responses generated from the study. The effect of the different levels of temperature, time and initial moisture content was compared using Analysis of Hariance (A@4HA). The 525eg >rocedure of 2A2 (version %) was used in ma/ing approximations of the / functions. 525eg was found suitable to $ approximate the said functions using the second order polynomials. Ki/ewise, using the same procedure, optimum independent variable values were determined such as would give the best performance and output responses. Gith the 525eg >rocedure, parameters of a complete uadratic response surface were fitted and critical values were determined such that response optimi+ation with respect to factors in the model was effected. The uadratic model obtained from regression analysis was used to build a 9.dimensional graph in which the different variable Ai was represented by a curvature surface as a function of Ii. The relationship between the response and independent variables was directly visuali+ed from the response surface plot. The information that the 9.dimensional graph conveyed was the same as that from the mathematical e uation.

#1

89 = =o + =i ; i + =ii ; i, + =i5 ; i ; 5
i < 5i
5 = i +#

C n ',) '. RESULTS AND DISCUSSIONS '.1 Re$2o($e Su"1ace A(al-$i$

where7 *o, *i , *ii were regression constantsMcoefficients which were estimated from the data of the study;

The 5esponse 2urface Analysis using the procedure of 2tatistical Analysis 2ystem (2A2) 3nstitute was conducted to determine the levels of temperature, time and initial moisture content coded as I#, I, and I9, respectively, that would give optimum values of the dependent parameters namely7 final weight (A#); final moisture content (A,); overall cup uality (A9); general acceptability (A(); and willingness to buy (A&). The stationary point was identified for each independent parameter to determine its turning points; a maximum, a minimum or a saddle point. 5idge analysis was performed when the stationary point is a saddle point. The optimum operating conditions of the mechanical roaster were sought in order to come up with the performance levels that were deemed as close to the ideal conditions as possible. 3n the optimi+ation study, 5esponse 2urface Analysis using the procedure of 2tatistical Analysis 2ystem (2A2) 3nstitute was used. A *ox.*ehn/en experimental design with three independent variables at three levels was used to study the effects of factor variables on the response variables. A *ox.*ehn/en experimental design has the advantage of re uiring fewer experiments (#& runs) than would a full factorial design (,: runs). 6actor variables, also /nown as independent variables, were coded I#, I,, and I9 and represented the roasting temperature, roasting time, and the initial moisture content of the coffee beans, respectively. !ifferent factor variable levels were li/ewise given codes7 .#, 1, J#, as shown in the same

table, with their corresponding code (actual) values. Table 9 presents the experimental data on response variables using *ox.*ehn/en design for the response surface analysis obtained during the final testing of the mechanical roaster. The weight of roasted beans ranges from ,.% /gs to 9.% /gs from its original weight of & /gs. This is in consonance with the findings of Nnox and "uffa/er (#$$:) as cited byin7 0oEica and >eralta (,119) '##) indicating that #,.,&? of the total weight of coffee beans was lost due to roasting. 5un number :, together with run number (, li/ewise showed to have the highest moisture content as against run number #Os outcome having the lowest value. 3n terms of overall cup uality, the highest rating of %$.&? was obtained as against the :,.9? lowest rating. 0aximum and minimum general acceptability and willingness to buy values were at :.$ and &., as well as %., and (.%, respectively. Gith the experimental data as sole bases of reference, experimental run number & having an actual combination of ,11 1C temperature, ,1 minutes time of roasting and #9 ? initial moisture content outran/ed all the rest in terms of overall cup uality and general acceptability and willingness to buy the product. '.% E11ec* o1 Fac*o" #a"ia)le$ o( Re$2o($e$ To determine the over.all effect of factor variables such as roasting temperature, roasting time and initial moisture content on the response

##

variables namely7 final weight; final moisture content, overall cup uality general acceptability and willingness to buy, the 5esponse 2urface 5egression of the 2A2 3nstitute was conducted. To functionally relate the response variable to factor variables, *ox and *ehn/en',) have advanced the validity of a second order polynomial as a graduating function to ade uately approximate the functional relation between the true response and independent variables. The second degree e uation li/ewise provided information on first and second order effects as well as interaction effects of the variables.
Table 9. Cxperimental data on final weight using *ox.*ehn/en design for the response surface analysis.
SAM RUN NO. # , 9 ( & D : % $ #1 ## #, #9 #( #& ROASTNG PROCESS CONDITIONS 61 6% 6' ,,& ,,& #:& #:& ,11 ,,& ,,& #:& #:& ,11 ,11 ,11 ,11 ,11 ,11 ,& #& ,& #& ,1 ,1 ,1 ,1 ,1 ,1 ,& ,& #& #& ,1 #9 #9 #9 #9 #9 #D #1 #D #1 #9 #D #1 #D #1 #9 RESPONSE #ARIA/LES Y1 ,.% 9.# 9.( 9.$ 9.& 9.( 9., 9.% 9.D 9.& 9.& 9., 9., 9.# 9.( Y% 9.% &.1 :.D %., &.% &.D &.# %., %.# &.$ (.$ &.1 &.D &., &.% Y' %9.D %(.1 :D.# :,.9 %$.& %,.$ %&.$ :%.# :9., %:.& %1.D %#.# %,.: %9.% %%.& Y, D.( :.9 D.1 D.D :.$ &.( :.1 &., D.& :.: D.( D., (.D :.( :.D Y. D.9 D.: :., D.: %., D.1 :.# &., D.& :.& &.% D.& (.% D.: :.(

A& B Gillingness to buy, in percent rating (?) 0eanwhile, Table ( presents the Analysis of Hariance (A@4HA) showing the significance of response surface models; the effect of treatment variables as linear, uadratic or interaction (cross.product) terms. A second.order polynomial e uation was used to obtain the different models that would fit the mathematical function / for the five response variables A# to A&. The A@4HA showed meaningful response surface regression models for all the response variables, having a highly significant effect on final weight, final moisture content, overall cup uality, general acceptability and at &? significance level on willingness to buy. These significant values were largely generated by the highly conse uential contribution imparted by both the linear and uadratic components of all response regression models, and more so, by the total regressed model itself.
Table (. Analysis of Hariance (A@4HA) table showing the significance of surface regression model and factor variables as linear, uadratic, and interaction terms (cross product) on each of the response variables.
DEGREE OF FREEDOM SUM OF S;UARES Y1 Y% Y' Y, Y.

where7 Fac*o" #a"ia)le$ I# B 5oasting temperature, 1C A# B 6inal weight, /g I, B 5oasting time, min; and I9 B 3nitial moisture content, ? d.b. Re$2o($e #a"ia)le$ A, B 6inal moisture content, ? d.b. A9 B 4verall cup uality -eneral acceptability, in percent rating (?)

0odel Kinear Puadratic Cross product Total 5esidual Kac/. of.fit >ure error Total Crror
,

9 9 9 $ 9 , &

#.9&:&QQ Q 1.9%$%QQ Q 1.1#,&ns #.:&$%QQ Q 1.19:&QQ Q (.#:1$ C. #( 1.19:& 1.$:$# ,.&%,D

,1.%&:&QQ Q &.#(9,QQQ 1.#$,&ns ,D.#$9,QQ Q 1.9&:&Q 1.11DD: 1.9D(, 1.$%D9 (.&1%1

#D%.D#:&QQ Q #:%.9&$%QQ Q ,1.#1,&ns 9D:.1:$%QQ Q #1.,$:&ns ,.1111 #,.,$:& 1.$D:% #.$#,%

(.,%:&QQ &.$(&:QQ Q ,.,$&1QQ ,.&,%,QQ Q 1.&%,&ns 1.1(D: 1.D,$, 1.$&,, &.(#%&

5 Coefficient of Hariation (?) QQQ significant at #? QQ significant at &? Q significant at #1? ns not significant

3t can also be noted in Table ( the residual variance that is comprised of lac/.of.fit and pure experimental error. Kac/.of.fit test measures the failure of the model to represent data in the experimental domain at points which are not included in the regression. The coefficient of determination (5,) is the proportion of variability in the data explained or accounted for by the model while the coefficient of variance (CH) is the ratio of the standard error to the mean value of observed response expressed as percentage. 3t is a measure of reproducibility of the model. As a general rule, a model can be considered as highly as reproducible if its CH is not greater than #1? (Tan uilut, ,11: citing 0iranda #$$#). 3t is evident that the model developed for the three response variables A9, A( and A& appeared to be very ade uate having possessed no lac/ of fit and a satisfactory 5, value of 1.$D:%, 1.$&,, and 1.%$%$, respectively. The results indicated that more than $1? of the variation in the response was accounted for by the function. 0oreover, its significant lac/ of fit suggested fitted y values as having approached close to A

values, which would be indicative of a linear regression function (0yers':) and 0ontgomery, #$$&'D) as cited by 5afosala, ,111). 3nsignificant lac/ of fit indicated that variations in the models were due to random error. 5esponse variable final weight exhibited a highly significant lac/ of fit values while final moisture showed evidence of significant lac/ of fit values. A significant lac/ of fit would indicate that the regression function would not be linear. A There could occur a number of significant variations which the random error might account for can occur, variations which might be caused by an un/nown factor which the response surface model did not ta/e into account. As such, careful thought should be done for conclusions arising from this process that other forms of the model should be investigated. The coefficient of variation (CH) relates experimental error as a percentage of the mean indicating the degree of precision when comparing treatments and serving as a sufficient index of the reliability of the experiment being conducted (-ome+ and -ome+, #$%()'9). CH values listed in Table $ showed relatively low values of ,.&%, (.&#, #.$#, &.(, and :.#D for A#, A,, A9, A( and A& response variables, respectively. The same way all the response variables had sufficiently high 5, values which measured the degree of reduction in the variability of the responses obtained when subEected to regressor variables I#, I, and I9. A highly significant effect of temperature and time on final weight, final moisture content and overall cup uality is in consonance with the study

#:

t e mp 230

215

200

of 0oEica and >eralta (,119). The temperature (I#) and ta/ing the forward higher the roasting temperature and the direction of time (I,). Thus a minimum longer the time of roasting the lower the weight and moisture content can be weight of the roasted beans, and reali+ed at a maximum roasting conse uently, the lower the final temperature and longer roasting time. moisture is. The result also proves the 4n the other hand, response curve of I # 6igure ,b. Contour plot for final moisture content (? 1 d.b.) as a function of temperature ( C) and time (min). claim of 2ivet+ (#$%9)'%) that among and I, (6igures ,c to ,e) formed a the factors that affect the roasting maximax system for response variable process are temperature and time of A9, A( and A&. The center point of the roasting. curve represents the maximum point. 4n the other hand, initial moisture This means that a decrease in responses content had no significant effects on the will ta/e place in the bac/ward move in final weight, final moisture content and I# and I, directions. Thus, an increase willingness to buy. This explains the in overall cup uality, general fact that the ranges of values of initial acceptability and willingness to buy #% moisture content that have been used as could be reali+ed on a move toward the factor levels were so closed to each directions of temperature and time. other (#1?, #9? and #D?). *ut it should also be noted that the selected 6igure ,c. Contour plot for overall cup uality (?) as a values were based on the recommended function of temperature (1C) and time (min). moisture content of the green beans, that is, #, 8 #9 ? dry basis. '.' Co(*ou" Plo*$ o1 Re$2o($e #a"ia)le$ 3t is convenient to visuali+e the relation between response and factor levels geometrically. The three dimensional 6igure ,a. Contour plot for final weight (/g) as a function strategy gave a clear understanding of on temperature (1C) and time (min). the behavior of the response variables throughout the total operating region. A very useful representation of the surface was obtained by drawing lines of e ual 6igure ,d. Contour plot for general acceptability (scale of #.#1) as a function of temperature (1C) and time (min). response on a graph whose coordinates denoted the levels of the factors. The response curves (6igures ,a to ,e) show the influence of temperature and time on the response variables A#, A,, A9, A( and A&. 6igures ,a and ,b show a saddle point which implied the existence of two distinct regions of maximum response. This means that a decrease in final weight and final moisture content will occur by moving away from the axis of roasting 6igure ,e. Contour plot1 for willingness to buy (scale of #.#1) as a
185 170 10 15 20 25 30 t i me f i nal MC 4. 0 7. 0 4. 5 7. 5 5. 0 8. 0 5. 5 8. 5 6. 0 9. 0 6. 5 t e mp 230 215 200 185 170 10 15 20 25 30 t i me over al l c up ual i t ! 70 82 72 84 74 86 76 88 78 90 80 92 t e mp 2 30 2 15 2 00 t e mp 230 1 85 215 1 70 10 15 20 25 30 200 t i me f i nal we i g h t 2. 50 3. 40 2. 65 3. 55 2. 80 3. 70 2. 95 3. 85 3. 10 9. 00 3. 25 185 170 10 15 20 25 30 t i me gener al ac c ept a"i l i t ! 4. 3 6. 4 4. 6 6. 7 4. 9 7. 0 5. 2 7. 3 5. 5 7. 6 5. 8 7. 9 6. 1 8. 2
t e mp 230 215 200 185 170 10 15 20 25 30 t i me 5. 2 7. 3 wi l l i n g n e # # t o "u! 4. 6 6. 7 4. 9 7. 0 5. 5 7. 6 5. 8 7. 9 6. 1 8. 2 6. 4 8. 5

function of temperature ( C) and time (min).

The eventual goal of canonical analysis is to be able to determine the nature of the stationary point and the entire response system. The analysis begins with a translation of the response function from the origin 4 (I# B 1, I, B 1, I9 B 1) to a new origin at the center of the system at 2, i.e. at the stationary point. The response function then is expressed in terms of new variables G#, G,, G9, the axes of which corresponds to the principal axes of the contour system. The calculation procedures are given by !ilidili (#$%9) citing !avies, et. al. (#$D9) and 0yers (#$:#). The general representation of the canonical form is given by7 8 : 8# > ?*@*% > ?%@%% > <<<. > ?9@9% C n '9) Ghere A2 denotes the response at the stationary point 2, and the center of the system is at I#2, I,2, and I92. The coordinates of the stationary point can be found by ta/ing the partial derivatives of the fitted response surface with respect to I#, I,, L.. IN, e uating to +ero, and solving for the IN2. 2ubstituting the values of the coordinates to the estimating e uation gave A2. The coefficient (R#, R,, LLLR/) of the canonical e uation, sometimes called as eigenvalues, aids in determining the nature of the stationary point and the response system by observing the sign and magnitude of ROs. 6or instance, if the eigenvalues are all negative, then the solution is ;maximum<; if eigenvalues are all positive, then solution is ;minimum<; and in case where ROs differ in sign, the stationary point is a ;saddle point<. 3f one of the R values turns out to be very small in comparison with the others, then the

'., Ca(o(ical A(al-$i$ o1 *0e Re$2o($e Su"1ace

,# system approximates that of a ;ridge<. weight, that is, the valley orientation of the Thus, the eigenvalues (R) indicate the response curve is more aligned with roasting degree of sensitivity of the systemtemperature if and the hill with time. The largest one has to depart from the stationary eigenvalue of 1.$&,,&D for temperature indicates point. that temperature effect is more pronounced and The relationship between the canonical axes GOs the curvature of the response surface was in its and the original axes IOs serve also as a valuable direction. 3t is evident that the surface for final piece of information. 3t can be determined moisture by content is more sensitive also to roasting using the transformation given by G B 0O (I temperature. 8 I2), with the columns of matrix 0 being the 6igure 9 represents the response surface plots of normali+ed eigenvectors associated with the overall cup uality. 2ince all the eigenvalues eigenvalues R#, R,, LLLR/. The eigenvector for possessed negative signs, it is indicative that the the largest eigenvalue gives the direction stationary of point of all the response variables is a steepest ascent if positive, or steepest descentmaximum. if 3n this case, the maximum values are negative. The eigenvectors corresponding the to optimum values. small or +ero eigenvalues point in directions of relative flatness. The canonical properties of the fitted surfaces are shown in Table &. The eigenvalues (R) or the coefficient of canonical variables (G) indicates the nature of surface occurring in the system Table &. Canonical property of fitted under consideration. The eigenvalues of the response surfaces. eigenvectors for final weight are R# B 1.,,91,,; R, B .1.1%9#$&; R9 B .1.,#:9,%, which indicates that PREDICT RESUL ED RESPON EIGE T OF the stationary point is a saddle point, since the FACTOR CRITIC #ALUE SE N RESPO #ARIA/ AL AT #ARIA/ #ALU NSE eigenvalues had different signs (J and .) and the LES #ALUE STATION LES ES SURFA ARY CE predicted final weight value at stationary point is POINT 6inal Temp 1.,,#9 ,,#.$9, 9.,:#19% 2tationar 9.,:. 0oreover, the final weight eigenvalue of Geight Time 1,, 1($ y point is 3nitial 0C . #%.#%(& a saddle 1.,,91,, shows that the orientation of the valley 1.1%9# %( point $& #(.:&%D is less curve than the hill orientation of the . ,# 1.,#:9 response curve for final weight is more aligned ,% 6inal Temp 1.$&,, ,#9.91& &.:$$D#& 2tationar with roasting temperature and the hill with time. 0oisture Time &D :D% y point is Content 3nitial 0C . #D.#11& a saddle The largest eigenvalue (absolute value) of 1.1,D1 %& point $# ,D.&9D( 1.,,91,, for temperature indicates that . 9( 1.D9%D temperature effect is more pronounced and the D& 4verall Temp . ,#1.%1( %$.($D1%# 2tationar curvature of the response surface was in its Cup Time ,.9%,$ &($ y point is Puality 3nitial 0C :1 #$.&D(1 a direction. The surface for final weight is more . D$ maximu 9.D:%9 #,.&(#( m sensitive to roasting temperature. &% $# . A saddle point also occurred for final moisture D.#&## :, content with eigenvalues R# B 1.$&,,&D; R, B -eneral Temp . ,1&.%9& :.$D1#9% 2tationar Acceptabi Time 1.9#%: (1: y point is .1.,%1$#; R9 B .1.D9%DD& and a predicted value at lity 3nitial 0C DD #:.&(&, a . %& maximu its stationary point of &.:$$D#&. As the orientation 1.1D($ ##.($1: m 1$D $9 of the valley and the hill, the final moisture . #.,&:# content has similar scenario with that of the final 9%

Gillingne ss To *uy

Temp Time 3nitial 0C

. 1.9#%# 19 . 1.D#9& &# . #.#%1% (D

,1#.&%1 &1% ,1.1D%D (( #,.#%&( ,(

:.:%:D11

2tationar y point is a maximu m

ov er al l

cup

ual i t !

,(
Temperature (1C)
30. 00

89. 94

82. 54

75. 14

23. 33 67. 74 230 210 t e mp 190 170 10. 00 t i me

16. 67

6igure 9. 5esponse surface plot of overall cup uality


Time (min)

6igure (. 2uperimposed contour plots of the response variables showing optimum region.

'.. O2*i5al P"oce$$i(! Co( i*io( 3t is the main obEect of this experiment to obtain the optimum operating conditions of the mechanical roaster that would give maximum response. To reali+e this, contour plots of statistically significant responses were superimposed to obtain the optimum region of the independent parameters. 2uperimposition yielded the optimum experimental region shown in (6igure (). Considering the nature at which roasting temperature values were chosen with their accompanying limitations and the ranges of the different levels of time and initial moisture content, the optimum region point was derived by determining the centroid by midpoint analysis. As such, the optimum values are ,1(.& 1C roasting temperature; #$.:& min roasting time; and #,.,& ? moisture content dry basis. To validate the mathematical model for the optimum process region, another roasting operation was performed using the above optimum values. A comparative analysis was done on the values of the predictive model for each parameter and the validated experiment.

'.< #e"i1ica*io( o1 *0e O2*i5u5 P"oce$$ Co( i*io($ After having established the optimum point, verification test was performed using the optimum values (,1(.& 1C roasting temperature, #$.:& min roasting time, and ,& the #,.,& ? moisture content) to verify ade uacy of the response. A comparative analysis was done on the values of the predictive model for each parameter and the validated experiment. Table D shows the experimental results where corresponding standard deviation value and coefficient of variation are li/ewise shown. 3t also presents a comparison between predicted and actual experimental values. Among the five second.

order polynomial models, the model for A# (final weight) and A, (final moisture content) had come up with a predictive value that was lower (&.((? and #:.1,?, respectively) than the actual experimental value. This is in spite of the fact that the model along with those of A# and A, did exhibit a lac/ of fit which was notably significant. The model on A9 (overall cup uality) which was previously adEudged to be having a non.significant lac/ of fit, produced an experimental output that was almost e ual to the predictive value having an error of 1.%1? only. The model for A( (general and acceptability) and A& (willingness to buy) which were also adEudged to be highly ade uate models came up with a far #1.9, ? and #9.$1 ? over and above what was predicted by its second.order model, respectively. 3t was believed that one factor affecting the result was the fact that sensory evaluation, in many instances, was highly subEective. 2uch subEectiveness might have ta/en into account in the outcome of the verification tests, particularly on general acceptability and on willingness to buy the product. Table D. >redicted and actual experimental data for values at optimum point.
RESPONSE #ARIA/LE PREDICTED #ALUE E6PERIMENTAL #ALUE RANGE = ERROR

,. CONCLUSION AND RECOMMENDATION Conclusions can be made from the foregoing studies that coffee roasting process is one of the food engineering operations that can be optimi+ed. 4ptimi+ing the agricultural processing operations should be performed in order to recommend best conditions for roasting operation resulting in a superior uality product as well as maximi+ing throughput capacity and reducing processing costs. 520 was found to be a useful approach and it should be recommended that this methodology be adapted to all optimi+ation studies.

..

REFERENCES

A# A, A9 A( A&

9.9#$9($ &.&(D#%& %$.1D1&1 :.%D#$(& :.:%#::(

9.& 1.(, D.( 1.,% %%.9& (.#: :.1& #.,1 D.: #.(#

9.1 8 9.$, D.#, 8 D.D% %(.#% 8 $,.&, %.%& 8 %.,& &.,$ 8 %.##

&.(( #:.1 , 1.%1 #1.9 , #9.$ 1

'#) *ATAK4@, S=A@3T4 T. #$$%. 4ptimi+ation of Coconut Coir !ust Compaction. A 0asteral Thesis. =niversity of the >hilippines Kos *aTos, College, Kaguna. ',) *CK3TU, ".!. in7 6ood Chemistry. Springer.Herlag, *erlin #$$$, pp. %:(8%&. '9) *4I, -.C.>. A@! !.G. *C"@NC@. #$D1. 2ome @ew Three Kevel !esigns for the 2tudy of Puantitative Hariables. Technometrics, vol. ,, no. (. pp. (&&.(:&. '() !A5T, 2. N. and ". C. "=52TC@ in7 Coffee, Holume #, Chemistry, 5.S. Clar/e, 5. 0acrae (Cds.). Clsevier, Kondon #$%&, pp. ,,98 ,D&. '&) !3K3!3K3, S

'&D) -40CU, N.A. and -40CU, A. A. 2tatistical >rocedures for Agricultural 5esearch, ,nd edition. @ew Aor/7 Sohn Giley and 2ons, 3nc. 'D:) 0A!A0*A, >.2. #$$:. 4ptimi+ation of the !rying >rocess7 An Application to the !rying of -arlic. !rying Technology. #&(#). pp. ##:.#9D. ':%) 04S3CA, 5. 0. and >C5AKTA, C. N. ,119. !esign, Construction and Cvaluation of A *atch.Type Coffee 5oaster for 2mall.2cale 5oasting. >ublished 02 Thesis. =niversity of the >hilippines at Kos *aTos, College, Kaguna, >hilippines. '%$) 04@!CKK4, K. C42TA, 5. and TA@C"3!A, >. P. ,11&. 5eliable Characteri+ation of coffee bean aroma profiles by automated headspace solid phase microextraction mass spectrometry. www.ss.Eournal.de '%#1) 04@T-40C5A, !.C. #$$&. 5esponse 2urface 0ethodology. Sohn Giley and 2ons, @ew Ao/, =.2.A. '$##) 0AC52, 5.". #$:#. 5esponse 2urface 0ethodology. Allyn and *acon, 3nc. *oston, =.2.A. '#1#,) 23HCTU, 03C"ACK. #$%9. Coffee >rocessing Technology. Gestport, Connecticut7 AH3 >ublishing Co., 3nc. '###9) The >hilippines 5ecommends for Coffee. #$::. Kos *aTos, Kaguna, >hilippine Council for Agriculture and 5esources 5esearch and !evelopment. D". Ruel M. Mojica presently chairs the !epartment of Agricultural and 6ood Cngineering Cavite 2tate =niversity, 3ndang, Cavite, >hilippines

D". Ruel M. Mojica is an Assistant >rofessor 333 of agricultural and food engineering and presently chairs of the
!epartment of Agricultural and 6ood Cngineering, Cavite 2tate =niversity, 3ndang, Cavite. C.mail7 ruelmoEicaVyahoo.com

A$$i$*a(* P"o1. E !a" o #. Ca$a$ was born in Tiaong, Pue+on, >hilippines in #$&#; received his *. 2.C. in Agricultural Cngineering in #$:& from the =niversity of the >hilippines Kos *aTos in #$:&. Gent to =niversity of @ew 2outh Gales, Australia in #$%D for the 0aster in Applied 2cience in 6ood Cngineering, and then enrolled in >h! in 6ood 2cience in the same school in #$$,. "e has wor/ed in the =niversity of the >hilippines Kos *aTos immediately after graduation in #$:&as research assistant and training assistant and started teaching crop processing courses in #$:: until to date at the Agricultural and *ioprocess !ivision, 3nstitute of Agricultural Cngineering, College of Cngineering and Agro. 3ndustrial Technology, =niversity of the >hilippines Kos *aTos. Cmail7 evcasas1(Vgmail.com D". &e$$ie C. Elau"ia obtained his >h!. in Cnergy Cngineering and 0asterOs !egree in 0echanical Cngineering from the =niversity of the >hilippines !iliman, Pue+on City in #$$9 and #$%D respectively. "e graduated from =.>. Kos *aTos in #$:% with a degree of *.2. Agricultural Cngineering. "e is a >rofessor at the College of Cngineering and Agro.3ndustrial Technology at =.> Kos *aTos. "e also serves as Cconomic 5esearch 3nstitute of A2CA@ and Cast Asia (C53A) Gor/ing -roup 0ember on *iomass 2ustainability in Cast Asia. "e was appointed by former >resident 6idel H. 5amos as !irector of Cnergy =tili+ation 0anagement *ureau

of the !epartment of Cnergy (!4C) from #$$& to #$$%. "e has published several papers in refereed Eournals, developed nationally policy papers, monograph, D". Ma"il-( M. Elau"ia is an Associate >rofessor in Agricultural Cconomics and obtained the following degrees7 *.2. Agriculture, 0.2. Agricultural Cconomics, and >h.!. in 0anagement. 2he has served =.>. Kos *aTos since #$:% as instructor and has been involved in research and extension activities of the =niversity. "er fields of expertise include 6arm 0anagement, 6arm Accounting, Agricultural 0ar/eting, >roEect Analysis, 0onitoring and Cvaluation, Cnergy Cconomics, Agricultural 6inance and 3nvestment 0anagement. !r. 0. 0. Clauria is a recipient of the ,1#1 =>K* Centennial >rofessorial Chair Award, =>K* =nited Coconut >lanters *an/ >rofessorial Chair Award, => !iamond Subilee 6aculty -rant and the Asian 5egional 5esearch >rogramme in Cnergy, Cnvironment, and Climate 5esearch 6ellowship at the Asian 3nstitute of Technology, Thailand. 2he has served as proEect leader of several foreign and locally funded proEects, has five international refereed publications and co. author of 6arm 0anagement *oo/ in the >hilippines.
ENGEL/ERT +. PERALTA is an Associate >rofessor of Agricultural Cngineering at the A*>54!, 3AC, CCAT, =niversity of the >hilippines Kos *aTos; obtained *2 in Agricultural Cngineering from the =niversity of the >hilippines at Kos *aTos in #$:%; registered as a >rofessional Agricultural Cngineer at the >rofessional 5egulation Commission in the following year. 3mmediately after graduation, wor/ed as a process engineer and

handboo/, with several papers published in proceedings and more than &1 papers presented in conferences and wor/shops here and abroad.
later as an instructor in =>K*; resigned in #$:$ to pursue 0. Cngineering from the Asian 3nstitute of Technology in 2eptember #$%# as scholar of the Australian -overnment. "e reEoined the =niversity of the >hilippines Kos *aTos in #$%#, became study leader of *34TCC" from #$%, to #$%&. C.N. >eralta pursued >h! in Agricultural Cngineering at Texas A W 0 =niversity, #$%D.Together with his adviser 4.5, Nun+e, co.authored ;2imple 5elative "umidity 2ystems that 6issure 5ice and 4ther -rains< at the 3nternational 2ummer 0eeting of the American 2ociety of Agricultural Cngineers, 5apid City, 2outh !a/ota and at the 5ice Technical Gor/ing -roup 0eeting, =niversity of California at !avis on Sune #$%% later published in the 323 Sournal Applied Cngineering in Agriculture. The American 2ociety of Agricultural Cngineers (A2AC) later recogni+ed The Nun+e research group for their wor/ on rice fissuring, considered as one of the ;top outstanding achievements in agricultural engineering of the ,1th century in the field of 6ood and >rocess Cngineering; was a consultant for the Appropriate Technology 3nternational >roEect on ;2tate of the Art 2tudy of Coconut *y.>roduct =tili+ation.< was a consultant for the Appropriate Technology 3nternational >roEect on ;>ilot >lant for Coconut >rocessing< #$$#.#$$9. 3n ,11,, C.N. >eralta with his doctoral student 5itchilda Halerio and Clda *. Csguerra were awarded the 4utstanding 2cientific >aper award by the @ational Academy of 2cience and Technology for their paper ;5heological >roperties of 0ango (Aangifera indica) 6ruits in relation to "andling and Transport.< >A2 %((9)7,9,.,(1. Currently, he is the editor of the >hilippine Sournal of Agricultural and *iosystems Cngineering (>SA*C). As Chairman of the 3AC Academic Affairs Committee in ,1## led the transformation of the *2 Agricultural Cngineering curriculum to Agricultural and *iosystems Cngineering. "is present research interests include applications of nanotechnology on agricultural and *iosystems

engineering.

You might also like