AP Application of Weibull Hazard Analysis to the Determination of the Shelf Life of Roasted and Ground Coffee . Cardelli and T. P. Labaza* © Cardelli Pillsbury Technology Center, 330 University Ave, SE, Minneapolis, MN 55414 (US.A) TP. Labuza: Department of Food Science and Nutrition, University of Minnesota, 1334 Fekles Ave, St. Paul ‘MN 55108 (US.A) (Received January 19, 1999; accepted April 28, 2000) Roasted and ground coffer was sored at constant O. partial pressre (0.5-21.3 Pa), a, (0.105-0.408) and temperature (488°C) Product aceptabliy was monitored by ase of a modined Weibull Hazard sensory method where the end of shelpife was the te at ‘which 50% conrumers found the product unacceptable. The effect of O, a, and temperature was studied from a kinetice standpoint ‘Oxygen increae fom 0.5 to 215 ka accelerated deterioration 20fold d water aciiy merease of 0.1 led fo a 80% increase It ‘elerioration uggesting non-nsymatc browning acti, while a temperature increase af 10°C rose the rate of deterlortion about 15-28%. The activation for shelf ife was ~ 13 kJimole indicating diffusion within the glassy matric i controlling deteraration. ©) 2001 Academic Prose ‘Keywords: Weibull Hazard; shelf; coffe; staggered design Introd Shelfife can be defined as the time until a product becomes unacceptable to consumers under a given stor age condition, Consumer acceptability is usually deter- mined by means of ‘consumer tests using a large number of untrained tasters in which samples with various ages are tasted together (Meilgaard et al, 1991). This is ac~ ‘complished by using samples of product manufactured at different times or by freezing aged samples to stop de- terioration. Both techniques introduce undesired error, ice, batch to batch variability or the continuation of the Asterioration during the frozen storage. Another technique used in industry for shelP-life deter- ‘isation is team judgment. Its frequently applied in the carly stages of product development, where prototypes are evaluated during the storage bya team familia with the product. The main drawback is that results are lime ited to the tasting team making necessary the validation with consumers prior to launching ‘The use of corzelation between trained panel evaluation ‘and consumer response has been reported in the liter3- (0023-6438/01/080273 + 06 $35.00/0 @ 2001 Academic Prose ture (Fritsch et al, 1997), First, 9 team trained to recog- nize deteriorating attributes quantifies changes in their intensity dusing storage. Samples ate frozen and then, used in a second test where consumer acceptability is determined, Consumer response is correlated to trained panel data to determine the limite for the intensity of deteriorating altributes in product at the end of con- sumer shelf-life. The shelHlife of new samples is found: equal to the time when intensity limits for deteriorating aticibutes are detected by the trained panel. The main advantage of the technique is that one set of consumer ddata is used multiple times when correlating (o trained panel astessment, The main disadvantages are that results depend on the type of the scale used for measuring. the deterioration (Shepherd et al, 1988) and that sample storage introduces error which in turn is amplified when coupling two sets of sensory data. The technique has created intense debate in the literature, Fritsch et al (1997) and sensory paycho-physicists support the use of ‘rained tasters to predict consumer sensory responses (Moskowitz, 1996, 1997), where as statisticians do not (Dagle, 1996). ‘An ideal method would be to use samples of one batch of product and evaluate them at various times during the storage with a large number of consumers. This is too costly and time consuming. One way to simply this is to lo 10,1006 2000.0732 All articles availble online at htp/rwwdeabbrary com on IDE) 273use staggered sampling designs in which reduced testing is done in the beginning of the storage focusing on samples close to the end of shelf-ife. This approach has been extensively used in the mechanical and electrical industries Johnson, 1964) Gacula (1975) proposed the use of staggered sampling designs for shell-ife studies in foods, The technique com sists on evaluating an inereasing number of samples as storage time progresses. The main advantages of a stag- gered design over the methods described above are that all samples come from one batch, product is tasted throughout the whole storage period until unaccepta- ble and that untrained consumers can be used for the evaluation Staggered sampling methods have been used to deter ‘mine the shelf le of luncheon meats (Gacula, 1975), cassava flour (Shitose et al, 1978), breakfast cereal (Pickering, 1984), ice ream (Wittinger & Smith, 1986), refrigerated meats (Andujar & Herrera, 1987), frozen foods (Tomasicchio, 1989), cottage cheese (Schmidt & Bouma, 1992), pasteurized milk (Duyveste, 1997) and. sausages (Thiemig, et al, 1998). Data analysis involves plotting hazard values versus time and using the Weibull distribution to determine shelf life equal to the time at which 50% of consumers find product unacceptable, Labuza and Schimdl (1988) and Fu and Labuza (1993) called the technique the ‘Weibull Hazasd Method’. More details of this method can be found in the literature (Nelson, 1969, 1972; Gacula & Kubala, 1975; Gacula & Singh, 1984). The objective of this paper is to present shelf life results for roasted and ground (R/G) coffee obtained with the Weibull Hazard method and to determine the effet of (0, partial pressure, ay and temperature on shelf-life, Material and Methods Coffee samples Freshly roasted Colombian Arabica coffee beans were ground in a McGarveys Commercial Grinder to “drip brewing size’ (McGarveys, Minneapolis, MN). Table 1 shows storage conditions used for the study. Levels of ©, partial pressure, TC and ay were selected to be representative of real world situations frequently found, in the storage of R/G coffee. Storage conditions were combined in a full factorial design Storage and sampling Samples of 120g of R/G coffee were weighed to 0.1 g in a Mettler PC 4400 balance (Mettler Toledo Ine, High- {stown, NJ) and placed into 11 amber bottles. The Doitles were hermetically sealed with rubber stoppers. ‘The head space of bottles was replaced once a week with 20L (NPT) of an O,/N, primary standard mixture at 4 specific concentration (Genex, Minneapolis, MN}. The gas mixture was humidified by bubbling through. saturated solutions of NaOH, MgCl, and KCO,, which have a, similar to that of coffee with an a, of 0.106, 0.248, and 0408, respectively. After Dushing, the internal pres- ‘Table 1 Storage conditions for sensory shelf life deter- mination in R/G colfee Factor Levels (©, partial pressure (KPs)———05, 10, 30,54, 103, 213 a (0.106, 0.248, dos reo) ans ll faetoral design: 633 = 54 treatments sure of botles was equilibrated for 1 min to atmospheric ressure. Bottles were kept horizontally in incubators at controlled temperatures of 4 0.5" 05°C and 35405°C. The depth of the R/G coffee layer was approximately 20 mm. Samples were initially taken once every 2/6 of the shelf life predicted from data published by Clinton (1980) using algorithms developed by Cardelli (1997). The accel erated part of the study started after 50% of samples ‘wore found unacceptable, Then, samples were taken once every 1/6 of predicted shelf life. The experiments were conducted unt six sample times were evaluated or the storage time reached 7 months, whichever came first, Sensory evaluation Samples of 20 of R/G coffee were removed at each sampling time, weighed to the 0.1 gon a Mettler PC 4400 balance (Mettler Toledo Inc., Hightstown, NJ) and brewed with 11 of bottled drinking water (Kandhyoi, Minneapolis, MN) using standard coflee Gtr paper (Mr. Coffee Inc, Bedford Hts, OH) ina tabletop coffee maker (Mr. Coffee Inc. Bedford Hts, OH). The brewed coffee ‘was kept in 1 L Thermos jars and then poured into pre- warmed ceramic cups (200mL capacity} and covered with aluminum foil. Samples were then identified with random numbers and served to untrained ‘Tasters were recruited among students (50! female, 20-31 years of age) and stall (5 female, 23-58 years of ag2) of the University of Min- nesota and allowed to participate in the study only ifthey regularly drank one or more cups of black coflee per day, without sugar or cream. Three tasters were used at the beginning of the test. The number of tasters was increased by C = 1 at each sampling time until half ofthe tasters found the samples unacceptable. After that the number of tasters for each period was increased by C+ U, with U = number of unacceptable responses for the previous test time, No more than eight samples per session were presented to tasters who were asked to grade the brew as being acceptable or unacceptable. Tasters were also asked to rinse their mouth with water and wait for 2 min between samples. Data analysis The results were transferred to a master spreadshect for havard calculation and then to a Weibull Hazard plot as described by Gacula and Singh (1984). The eriteria for 274the end of shelf life was set as the time for 50% probabil ity of untrained tasters to grade the samples as being, unacceptable. This probability corresponds to # cumu= lative hazard of 69.3, The shape factor (f) was determined. as Ljslope. The Weibull distribution is unskewed for 2< p< 4, which leads to better shelf ie estimates, When the testis extended beyond shelf life, most of samples are judged unacceptable shifting § outside the optimum range. Tn thee situations, data were re-plotted up to 2 cumulative hazard of 100, Results and Discussion ‘Typical results and hazard values are shown in Tables 2 and 3, respectively. Figure 1 shows a Weibull Hazard plot for time versus cumulative hazard, A first sttempt resulted in > 4, therefore only cumulative hazard values range, i¢. an approximately 60% decrease in shell life per 0.1 ay increase as seen in Table 4, The strong a, effect on, ‘the Kinetics of shelf life deterioration may be related to the contribution of non-enzymatic browning. ‘The effect of temperature on shelf life was studied by dofermining Quo values and the energy of activation ( for kinetics of deterioration, Table § shove that the results for Qyo ranged from 1.15 to 123, is.a 15 10 23%: acceleration per 10 °C increase in temperature. The mag- nitude of E, was 13 kl/mole, indicating that sensory shelf life deterioration is controlled by a diffusion process Which has litle temperature sensitivity (Labwza & Schmidl, 1985), Even with this low temperature sens: tivity, refrigerated storage in air at 4°C versus room, temperature would give about a 44% increase in shelf ‘Table 3 Weibull hazard ranking table for R/G coffee at 100 Were plated with = 34 Shellfereitedin 225 L0KPa 05, 35°C and sy ~ 0248 weeks wth 83% confidence limits of 206-245 weeks Shelf life results were analysed from a kinetics standpoint Rank Werks Hyalue* with a fear on the eect of ©, para pressre ay and 3p D lemperatre onthe acseration ofdcteroraton Figures 3 a & 2tod show al the shelf data collced mn oueexper= 33 m 8 Iments Av expected the sensory self fe decresser wth 22 "3 & the incense in O; patil promure a, and tempertars. 2 03 a ‘Among these factor oxygen asthe most ciel olen 32 a x the deterioration with an approximately 204018 seser= Ig at Ea sion when increasing prvte fom 05 (0 213KPa 17 >I 35 Most oR G colle in canaisvacuim packaged at abou 16 at 83 ISPs af Os. Onjeen ie rapidly comrumed, which pos 2 6 vides a shelf life of approximately 2 years at room tem= if at Mu : 8 a iu erature (CHnton, 1980; Radke & Pinger, 1981) Bul 3 Bi 8 nee opened, the shed fess than a month Clarks, Ba a 1995) Theefee ofonygen war noticed bythetaterrwho 10 Bi oo ‘were invited to add comments in the score sheets. Terms : coed nb like ‘hes “ondized’ and stale’ were equenty wied to 8 iB describe the evolution ofthe vor of RG cote during 3 é the sty 5 Bi ne Onetechnigueto cvaluate heeft ofwateraciiyisto 4 Bi 30 catelate Q, which i the rato of abl hfe two water 2 a fctiviter a, unite apart (Loncin tal, 1968) ForR/G a8 ae coffee, a, had the socond most important fst with 2°Q, value of about 16 over the whole espedinestal “H— Huaed vale = 100m Table 2 Typical Weibull sensory data for R/G cole (10 KPa O,, 35 °Cand a, = 248) Wes Ter ey mo opti. bro et m2 2 ee 8 na * hb m FY 2 TK — a iB 2s t +2 et et bit ab ' acorplable ample a assored by am watrained (aster Ao unacceptable sample ae aaseered by an untrained taster. 275oss UL =246 Sie wscK= Fe prabty ig. 1 Weibull huzatd plot for R/G coffee (LOKP2 0s, 35°C and ay woo 0.268), Commlative hazard < 100 Shale (wes) pati reste KPO) ig 2 Sensory shelf fe of B/G cafe a 4°C, aye 0.106, life. Storage of an open can in the freezer (=— 15°C) ‘would increase shelf life over that at room temperature by about 70%. ‘The low E,s determined in our experiments could be ‘ consequence of the fact that coffee isin the glassy state for all storage conditions (Cardelli, 1997). In this state, the limiting factor is that oxygen needs to diffuse through, the matrix before it can react withthe coffee lipids. Even, though lipid oxidation has a high temperature depend- fusion controls the rate at which O, can diffuse 024s, a 0.408 to the lipid. Thus, when temperature is increased. the amount of oxygen available for reaction is what linsts the reaction rate Conclusions ‘The impact ofthe storage factors on shellife decreased in the following order ©, partial pressure, ay and temper- ature, Shelf life was reduced in about 20 times when. 276os Stele (wees) 20 She (veek) s 0 6 20 pail presse APs) 0106, C0248, a 0.408 ° 5 0 s 20 Fig. 4 Sensory shell ie of R/G coffee st 35°C, ay: MO.106, 0.248, & 0408 ‘Table 4 Water activity dependence, Q,, for the shelflife Table § Temperature sensitivity, Qso, and energy of of R/G Coffee stored at 10.1 kPa O, and 22°C activation for sensory shelf life of R/G coffee stored at tor era 04 and a = 0.248 te E 106 1s TEC) Qe krimole os 16 aoe oe ia iw 2 1s a 5 15 2 (0, level increased from 0.5 to 21.3 kPa, only 60% ay rose in 01 and approximately 20% Hazard analysis and Weibull probability distibution pper each 10°C provided a simple approach to shel life determination as increment. The use of staggered tasting combined to perceived by consumé 277References ANDUTAR,G. AND HnaeRA, I. The distribution offalure data Tor meat products. Proceedings of the European Meeting of ‘Meat Research Workers, eeinki, gust 2-7, No. 38 Vol {814 pp. 396-198 (1987), CARDELLE, C. Kinetics of the shelf Ne of roasted and (ground cofee a5 a fnction of oxygen concentration, water Gubity and temperature. PhD. thesis. University of Minnesota (1997) CLARKE RI. The thle of cote, In: Charalassbous (Bd), ‘Shall life studies of food and beverages. London: Eleeviet ‘Science Publishers 3V. (1998) CUNTON, Wo P. Constimer and expert evaluations of stored cofle products In: ASIC (Bd), Proceedings of the ih ASIC Meeting. London. ASIC, pp. 273-285, June 16-20, 980) DUGLE, J. Note on “Experts Veesus Consumers a Compari- son". Journal of Sensory Studies, 12, 147-153 (1996) DUYVESTEYN, W. Integration of the time-temperature history ecto the shelf ie of fla milk, MSe tes, Department of Food Saence and Nutntion, Univernty of Minnesota, St. Paul (1997) urtscr, C. W., HoFLano, C. N. ap VICKERS, Z. M, Shelf Tie af Sunflower Ketnels, Jounal of Foad Science, 62, 425-428 (1897) FU, BLAND LaaUZA, T. P. Shelf fe prediction: theory and application, Food-Coniro, 4, 128-133 (1993) GACULA, M,C, AND KUALA, JJ, Statistical models for shelf Ii failures. Journal of Food Science, 40, 04-409 (1975) GACULA, M. CAND SINGH, J. Stariaticl methods m food and ‘consumer research, New York: Academie Press Ine (1984) GACULA, M_C. The design of experiments [or shell ile study. “oural of Feed Sctenee, 40, 400-403 (1975). JOHNSON, L. G. The Statistical Treatment of Fatigue Experi. ‘ments. New York: Elsevier Pub, Co” (1963) Lasvza, TP. an Scrsaipi, M. K_ Accelerated shelf lie Testing Food Technology, 39, 97-62. 64, 134 (1985) LAaUZA,T.P. AND SCHMIDL, MLK. Use of Sensory Data in the ‘Shel Life Testing of Foods: Principles and Graphical Method for Evaltation Cereal Foods Word, 3, 195-206 (4988) LLONCIN, M, BIMENEE, JJ, AND LENGES, I, Influence of the tivity of water on the spoilage of foods. Journal of Food Technology (Br), 8.131 (1968) Menard, M. Civiaut, G, V. AND Car, BT. Sensory Braluation Techniques, 2nd BA, London: CRC Press 416 pp. 991) Moskownrz, 11 R. Expert versus consumers: A comparison Journal of Sensory Studies, H, 19-37 1995) Moskowrtz, H. RA zeply nole on experts versus consumers ‘A coniparzon. Journal of Sensory Statics, 12, 159-168 (1997), [NeLsow, W. Hazard plotting for incomplete failure dats. Jour- ral of Quality Technology. 1. 27-30 (1969) NELSON, W. Theory and applications of hazard ploting for ensored failure dala. Techmometrics, 14, 945-866 (1972) PICKIRING, S.C Prediction and analysis of shelf fe stability of ‘an oat bran cereal with antioxidants. M.Se. tess, Universi ‘of Minnesota, St.Paul (1984) RADTKE, Ro AND PIRINGHR, O. G. Zur problematik der ‘qualitasbeurteung von rostkaflee dusch quaniitetive Spurenanalyse Mushtiger__aromakomponenten. Deustche Eebonamittel Rudshow 77, 203-210 (1981) Sciiit, K- AND BOUMA, 1. Estimating shelf ie of cottage theese using hszatd analysis, Journal of Dairy Science, 7, 2922-2827 (1992) SHerneRD, R, Griperis, N. M. AND SweTH, K, The relation ‘shup between consumer preferences and trained panel sponses. Journal of Sensory Studies, 3, 19-35 (1988) ‘sutkose, [, AGUIRRE, J M.AND FERREIRA, V. L. P. Applica tion of a staggered experimental design to study of the shelbile of eatsava Nout enriched with foybean protein ex traction resitue Coleinea do Instituto de Tecnologia de Alimentos, 9, 313-334 (1978) ‘Tuna, F Buu, H. np Wotr, G. Characterization of self life and spoilage of resh foods. Firs ests withthe Weibull hazard analyses, Fleschwirtechal, 78, 152-135 (1998) ‘Tosusiceulo, M. ANDKHOTTL R. PikaZz0Us, P. AND Pez ZASI.G. Application of the Weibull statitical model to the Shellie of some frozen foods Industria Conserve, 64 102-109 1989) Warnivcrx, 8. A AND Soatu, D. E. Bet of sveeteness and stabilizers on selected sensory attibutes and shel lie of ice ream Journal of Pood Science, $1, 1463-1466 (985) 278