You are on page 1of 10

Journal of Food Engineering 95 (2009) 479–488

Contents lists available at ScienceDirect

Journal of Food Engineering


journal homepage: www.elsevier.com/locate/jfoodeng

A case study of investigating reliability and maintainability in a Greek juice bottling


medium size enterprise (MSE)
Panagiotis H. Tsarouhas a, Ioannis S. Arvanitoyannis b,*, Zafiris D. Ampatzis c
a
Alexander Technological Educational Institute of Thessaloniki, Department of Standardization and Transportation of Products – Logistics, Hellas, Greece
b
University of Thessaly, School of Agricultural Sciences, Department of Agriculture, Icthyology and Aquatic Environment, Fytokou Street, 38446 Nea Ionia Magnesias,
Volos, Hellas, Greece
c
Alexander Technological Educational Institute of Thessaloniki, Department of Electronics, Hellas, Greece

a r t i c l e i n f o a b s t r a c t

Article history: The reliability and maintainability analysis of an automated mechanical equipment of juice bottling
Received 21 April 2009 industry for a period of forty-five months at machine, workstation and entire line level was made. The
Received in revised form 29 May 2009 descriptive statistics of failure and repair data, the identification of the most important failure modes,
Accepted 5 June 2009
and the determination of the theoretical distributions parameters that best fit to failures data was carried
Available online 11 June 2009
out. Furthermore, the reliability, maintainability, failure rate, and repair rate models of the production
line for all workstations and the entire production line were developed. The models could prove to be
Keywords:
a useful tool both to assess the current conditions and to predict the reliability for upgrading the opera-
Juice production line
Reliability models
tions management policies of the production line. It was pointed out that (a) the availability of the juice
Maintainability models production line is 85.66% and the efficiency amounts to 82.10% because the equipment’s failures. (b) The
Performance evaluation WS1 and WS4 display the most frequent failures and the lowest availability with 94.54% and 96.13%,
Quality respectively. Moreover, they show the largest number of different failure modes of the production pro-
Field failure–repair data cess due to the complexity of the equipment with 28 and 30 different failure modes respectively. (c)
In the production line, a failure occurs every 12.5 h or equivalent to two-failures per day approximately,
whereas the mean TTR a failure is approximately two hours per failure. (d) The production line for the TTF
follows the Weibull distribution, while the TTR follows the lognormal distribution. This particular meth-
odology can also be utilized in the bottling industry sector by the machinery manufacturers and the man-
ufacturers of bottled products to improve the design and operation management of the juice bottling
production line.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction nient impact is to increase both reliability and maintainability of


the production system.
Reliability and maintainability are the most important parame- There is vast literature on reliability and maintainability with a
ters that determine the quality of the product. They aim at estimat- high number of books and articles. Blischke and Murthy (2000),
ing and predicting the probability of the failure, and optimizing the Ireson et al. (1996) and Usher et al. (1998) described concurrent
operation management related with the prevision of the failures, engineering in detail and presented application examples in differ-
i.e. maintenance policies. The negative results of failures are ent industries. Eliashberg et al. (1997) studied the models and
numerous and vary, having an economic impact. If the failure evaluated reliability as a function of time and usage. Blischke and
occurs frequently, then the manufacturer loses enormous capitals Murthy (2003) reported on 26 cases on reliability and maintain-
due to low production rate, product recall, lawsuits, cost of war- ability and statistical techniques illustrated included modelling,
ranty, loss of reputation etc. In some cases, an injury or even death reliability assessment and prediction, simulation, testing, failure
may be involved in extreme events. On the other hand, the occur- analysis, failure mode and effects analysis (FMEA), use of expert
rence of failures is inevitable, implying that equipment will fail judgment, preventive maintenance, statistical process control,
sooner or later. The only way to avoid or minimize the inconve- regression analysis, reliability growth modelling and analysis,
repair policy, availability analysis, and many others. Seifoddini
and Djassemi (2001) reported that the impact of delays, due to
machine breakdowns, is not limited exclusively to the production
* Corresponding author. Tel.: +30 24210 93104; fax: +30 24210 93157.
E-mail addresses: ptsarouh@logistics.teithe.gr (P.H. Tsarouhas), parmenion@ rate but affects the scheduling and productivity of the entire man-
uth.gr (I.S. Arvanitoyannis), zafiris@el.teithe.gr (Z.D. Ampatzis). ufacturing operations as well. Montgomery (1985) identified a link

0260-8774/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfoodeng.2009.06.011
480 P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488

between quality improvement and productivity. The relationship described, (b) in Section 3, the field failure–repair data were dis-
between machine reliability and system productivity was also played, (c) the most important descriptive statistics of the failure
investigated by Koren et al. (1998) who determined the productiv- and repair data for all machines, workstations and the entire line
ity throughput based on different states of the same system. were shown in Section 4, (d) in Section 5, the determination of
Ben-Daya and Duffuaa (1995) highlighted the relationship the important failure modes were presented, (e) in Section 6, the
between maintenance and quality, stressed its importance, and failure distributions were identified. (f) The reliability, failure rate
proposed a broad framework for modelling the maintenance–qual- and maintainability, repair rate models were developed in Section
ity relationship. 7, and finally, (g) the conclusions were drawn in Section 8.
In literature, there is a scarcity of field failure data of production
lines. Inman (1999) showed four weeks actual production data 2. Production process of juice bottling production line
from two automotive body-welding lines. His aim was to reveal
the nature of randomness in realistic problems and to assess the The automated juice bottling production line consists of seven
validity of exponential and independent assumptions for service workstations in series controlled with a common system. All the
times, inter-arrival times, cycles between failures, and times to re- processes are automated with mechanical means. The juice bot-
pair. Wang et al. (1999) described the failure analysis of computer- tling production line consists of seven workstations (Fig. 1):
ized numerical control (CNC) lathes; the field failure data was blower, labeler, mixer, rinsing–filling–closing, multipacker, pallet-
collected over a period of two years on approximately 80 CNC izer, and stretch wrapper. Every workstation consists of one or
lathes. more machines and each machine may be susceptible to different
In the case of food industry, the literature of field failure data is failure modes.
extremely limited. Liberopoulos and Tsarouhas (2002) reported a
case study of speeding up a croissant production line, based on ac- 2.1. Dlower, WS1
tual data collected over ten months, by inserting an in-process buf-
fer in the middle of the line to absorb some of the downtime, based This workstation is responsible for PET formation. It consists of
on the simplified assumption that the failure and repair times of four machines, compressor air at 40 bars (M.1.1), furnace pre-form
the workstations of the lines follow exponential distributions. Lib- blower (M.1.2), water chillers (M.1.3), and conveyor belts (M.1.4).
eropoulos and Tsarouhas (2005) published a statistical analysis of The bottles in their initial form are of 10 cm maximum length
failure data for an automated pizza production line. The analysis and only their diameter is equal to the bottle opening. The pre-
included identification of failures, computation of statistics of the forms heating occurs by means of lateral bulbs. Then air is intro-
failure data, and parameters of the theoretical distributions that duced at 40 bars pressure so that the pre-forms are shaped accord-
best fit the data, and investigation of the existence of autocorrela- ing to the required final shape. Cooling water circulates in the
tions and cross correlations in the failure data. Tsarouhas et al. serpentine (water blanket) to avoid the formation of hazy PET.
(2009a,b) developed the reliability and maintainability analysis The blower is assisted by a unit producing air at 40 bars and a cool-
of strudel and feta cheese production line at machine, workstation er (M.1.3) for pre-form chilling. The potential of the WS1 is
and entire line level; descriptive statistics of the failure and repair 22,400 bph and the mean treating time is 47 s.
data was carried out and the best fitness index parameters were
determined, and the reliability and hazard rate modes for all work- 2.2. Labeler, WS2
stations and both production lines (strudel and feta cheese) were
calculated. It consists of two machines; the labeler (M.2.1), and the con-
In this study, reliability and maintainability analysis was con- veyor belts (M.2.2). There is a coupe of labeling systems; the first
ducted for juice bottling industry by applying statistical techniques is providing a label per bottle and the other a label per pallet. This
on field failure data. Reliability, maintainability, failure rate, and workstation includes a furnace for heating the glue provided by
repair rate models of the production line for all workstations and one of the suppliers. The average processing time per bottle is 15 s.
the entire production line were developed. Data collection from
the line and their analysis were valid over a long time of forty-five 2.3. Mixer, WS3
months. The developed models could prove to be a useful tool both
to assess the current conditions and predict the reliability for The workstation includes three machines; mixer (M.3.1), pas-
upgrading the operations management policies of the production teurizer (M.3.2), and brix analyzer (M.3.3). In this workstation
line i.e. maintenance strategy, spare parts, inventory control etc. the syrup is produced with which the PET bottles will be filled.
The rest of this manuscript is organized as follows: (a) in The water used coming from a drill goes through a de-aerator to
Section 2, the production process of the juice bottling line was remove any residual air traces. Then, the water is mixed with the

Fig. 1. Schematic presentation of juice bottling production line.


P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488 481

appropriate sugar quantify in a tank. This process lasts for 45 min. The packing is carried out by means of plastic film which is
The mixture is filtered, sterilized and mixed in another tank with shrinked when it goes through a chamber with high temperature.
standardized syrup (blown color) and additives (i.e. preservatives). The shrinking process takes place twice and the average packing
The syrup is pasteurized by means of a heat exchanger at temper- time in this station is 202 s.
ature from 71 °C up to 105 °C and for 30 up to 60 s. The pasteurized
syrup (orange color) is cooled down to 7 °C in a second heat ex-
2.6. Palletizer, WS6
changer in order to get mixed with CO2 at 6 bar (pressure). The
pasteurized and carbonated syrup is temporarily stored in tanks
This workstation includes three machines; divider (M.6.1.), pal-
of varying capacity between 9600 and 25,000 l and then moves
letizer (M.6.2.), and pallet storage (M.6.3.). The six-item packs are
to the filling machine. The product can remain in tanks for 48 h
properly stratified by means of a robotic system to ensure a contin-
without being spoiled (Kourtis and Arvanitoyannis, 2001).
uous flow (influx). The average processing time per bottle amounts
to 1.25 s.
2.4. Rinsing–filling–closing, WS4

This workstation is composed of five machines; rinser (M.4.1), 2.7. Stretch wrapper, WS7
filler (M.4.2), closing (M.4.3), coder (M.4.4), and conveyor belt
(M.4.5). The plastic bottles PET are first rinsed by being turned up- The workstation includes stretch wrapper (M.7.1), small cylin-
side down so that any residual dust or other remnants are re- ders-conveyor (M.7.2), and the bar coder (M.7.3). The pallet is
moved. Rinsing can be conducted either with ionized air (water wrapped with plastic film for ensuring stability of the packed bot-
economy) or by using two nozzles; (i) sprinkling gasified deter- tles. Then, the pallet moves to an adjacent position where the label
gent/sterilizer and (ii) sprinkling sterilized water. The syrup filling is applied. The average processing time this station is 1.25 s per
of PET bottles is the next stage where two valves (opening and clo- bottle.
sure) ensure the passage of juice and CO2, interchangeably. The fi- Throughout the paper we use the following notation to distin-
nal two stages in this workstation are bottle stoppage and coding guish between the different levels of failures in the production
(consumption date with indelible ink). The total processing time line:
in this station amounts to 180 s. Then, the bottles move to the con-
veyor belts. WS.i = Workstation i,
M.i.j = Machine j of workstation i,
2.5. Multipacker, WS5 F.i.j.k = Failure mode k of machine j of workstation i.

This workstation consists of six machines; conveyor belt Based on the above notation, the workstations and machines of
(M.6.1), labeling machine per two bottles (twin) (M.6.2), machine the juice bottling production line are given in Table 1. The number
for handle placing (M.6.3), conveyor belt for twin pack (M.6.4), of recorded failure modes per machine is indicated as well. The
multilacker furnace (M.6.5), and conveyor belts (M.6.6). This work- processing time per juice bottle at the machine level is indicated.
station is practically composed of two similar ‘lines’ aiming at This time is equal to the capacity of the particular machine or
packing twice the PET bottles: first two bottles together with han- workstation divided by the nominal production rate of the line,
dle and secondly three bottles twice (that is six individual bottles). according to Little’s formulas (1961) in Queueing Theory.

Table 1
The workstations and machines of juice bottling production line.

Workstation Machine Failure modes Failures Cycle time Description of machines


WS 1 M.1.1 8 49 1a Compressor air at 40 bars
Blower M.1.2 14 384 30 Furnace pre-form blower
M.1.3 3 8 1 Water chiller
M.1.4 3 6 15 Conveyor belts
WS 2 M.2.1 10 118 5 Labeler
Labeler M.2.2 3 7 10 Conveyor belts
WS 3 M.3.1 6 45 14,340 Mixer
Mixer M.3.2 3 9 60 Pasteurizer
M.3.3 1 1 0 Brix analyzer
WS 4 M.4.1 6 21 7 Rinser
R–F–C M.4.2 10 220 35 Filler
M.4.3 6 75 7 Closing
M.4.4 3 11 1 Coder
M.4.5 5 11 130 Conveyor belt
WS 5 M.5.1 2 4 60 Conveyor belt
Multipacker M.5.2 5 68 40 Labeling machine per two bottles(twin)
M.5.3 5 27 2 Machine for handle placing
M.5.4 3 8 30 Conveyor belt for twin pack
M.5.5 5 17 30 Multilacker furnace
M.5.6 4 12 40 Conveyor belts
WS 6 M.6.1 1 1 0.25 Divider
Paletizer M.6.2 9 104 1 Palletizer
M.6.3 1 6 0 Pallet storage
WS 7 M.7.1 4 31 0 Stretch wrapper
Wrapper M.7.2 4 13 0 Small cylinders-conveyor
M.7.3 2 5 0 Bar coder
a
Processing time per bottle in seconds.
482 P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488

3. Field failure–repair data to undertake ‘preventive actions’ in advance thereby ensuring the
production line reliability.
Failure and repair data of the juice bottling production line were Based on the data reported the registrations-failures amounted
collected from the files of the technical department by the end of to 1261 and for the entire line were categorized in 126 different
each shift the recorded in prints by the technicians in charge failure modes. Furthermore, 29 failures with null TTF were counted
(mechanical and electrical). These files covered a period of that regarded failures which even if the technicians repaired it, the
1362 days that is three years and nine months. Over this period, machine did not operate due to a second failure occurring simulta-
the line operated in total 928 days excluding weekends, holidays, neously with the first. For these failures it was considered that TTF
and the strike days. Out of the 928 continuous working days the equals to a minute instead of zero that is actually, thus influencing
501 days line operated in two shifts and the rest 427 in three shifts. the collected data that at the minimal possible degree. This is a
The records included the failures that had occurred per shift, the necessity because the software package MINITAB cannot make
action taken to repair the failure, the down time, and the exact parametrical analysis if there are null times in the data. Fig. 2 dis-
time of failure. Therefore, there is the exact time both for the ma- plays the histograms of TTF and TTR over the entire production line
chine failure and between failures. This means that the precision in level. The histograms of TTF and TTR exhibit the typical skewed
computing the time to failure (TTF) of a failure for machine, work- shape to the right, and as a result the Weibull, lognormal, or expo-
station and the entire line itself is in the order of minutes. The time nential distribution will be studied in order to find out which pro-
to repair (TTR) a failure was also recorded in minutes. vides the best fit.
TTF of equipment is defined as the time that elapses from the
moment the equipment goes up and starts operation after a failure, 4. Descriptive statistics of the failure data
until the moment it stops operation due to a new failure. The TTF
can be considerably influenced from the maintenance policy, The application of descriptive statistics to the failure data is
which is undesirable but inevitable, because of the manufacturers very effective for drawing conclusions with regard to the identifi-
who do not wish to operate the production lines for long time cation of most important failures, and the determination of distri-
intervals without act no preventive or proactive maintenance, in butions described by the TTFs and the TTRs. The sample mean, the
order to collect unbiased data for TTF. However, in our case the im- standard deviation (SD) and the coefficient of variation (CV) at
pact of TTF is minimum because the maintenance policy seldom al- machine, workstation and line levels were computed based on
lows replacements of spare parts before they failed. The failure the records. The availability is defined as the percentage of time
data are operation dependent failures, meaning that a machine that the equipment is up (Eq. (1)), and is calculated accordingly:
may fail while being in operation.
Availability ¼ mean TTF=ðmean TTF þ mean TTRÞ ð1Þ
TTR of failed equipment is defined as the time that elapses from
the moment the equipment goes down and stops until the moment Table 2 shows the descriptive statistics of the juice production
it starts operating again. The TTR includes the access time from the line where N is the sample size of the number of failures. The CV
technicians who realize that a fault exists, the diagnosis time, spare of a random variable is a dimensionless measure of the variability
parts procurement, replacement time, checkout time which con- of the random variable and is defined as the ratio of the standard
firms that the machine is operational, and adjustment time. When deviation over the mean of the random variable.
the damage is diagnosed, one of the technicians goes to the ware- A close observation of Table 2 leads to the following results: (a)
house in order to bring the spare parts while the others begin the the WS1, WS4, and WS5 have the most frequent failures and the
repairing process. The equipment operator also helps the techni- lowest availability with 94.54%, 96.13%, and 98.63%, respectively.
cians during the machine repairing. The WS1 and WS4 presented the largest number of different failure
The maintenance policy of the bottling production line is pre- modes of the production process due to the complexity of the
ventive and corrective maintenance. The preventive maintenance equipment with 28 and 30 different failure modes, respectively.
of the line takes place manly over the weekends when there is a (b) The machines M1.2, M4.2, M2.1, and M6.2 have the lowest avail-
drop in the product demand in order the production line to remain ability with 95.30%, 97.34%, 98.74%, and 98.82%, respectively. (c)
functional and the future failures to be minimized. The preventive The availability of the line is 85.66%. Furthermore, on the produc-
maintenance is based on an application of specific preventive tion process there is a break of twenty minutes per shift for the
actions such as cleaning (water/air filters, nozzles, cameras, and workers thereby providing an additional loss on the production
bulbs), checking and replacement of bulbs/pre-form triggers, rate, which equals to 1/24 = 4.16%. Consequently, the effective
furnace catena, and machinery oil. The aim of preventive mainte- production rate becomes (85.66%)(100%  4.16%}) = 82.1%. This
nance is to detect the contamination failure probability in order finding is in agreement with the 82% output efficiency reported

Histogram of TTF Line Histogram of TTR Line


900
300
800
700 250
600
Frequency

200
Frequency

500
400 150

300 100
200
50
100
0 0
0 5000 10000 15000 20000 25000 30000 0 100 200 300 400 500 600
TTF Line TTR Line

Fig. 2. Histograms of TTF and TTR for the juice bottling production line.
P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488 483

Table 2
Descriptive statistics of the failure data at the machine, workstation and line levels.

TTF TTR
N Mean SD CV Mean SD CfV Availability
M.1.1 49 21,641 41,079 1.8982 141.8 79.5 0.5606 0.99349
M.1.2 384 2723 4757 1.747 133.39 79.95 0.5994 0.9533
M.l.3 8 80,289 94,105 1.1721 116.3 37.1 0.319 0.99855
M.l.4 6 152,568 82,880 0.5432 120 48.5 0.4042 0.99921
M.2.1 118 8985 12,937 1.4398 114.45 59.11 0.5165 0.98742
M.2.2 7 136,553 220,410 1.6141 88.6 40.9 0.4616 0.99935
M.3.1 45 22,596 41,240 1.8251 141.4 71.6 0.5064 0.99378
M.3.2 9 106,538 105,578 0.991 131.7 58.2 0.4419 0.99877
M.3.3 1 – – – 50 – – –
M.4.1 21 51,253 50,142 0.9783 111.7 62.9 0.5631 0.99783
M.4.2 220 4792 9161 1.9117 130.7 68.48 0.5239 0.97345
M.4.3 75 14,201 29,218 2.0575 111.13 65.57 0.59 0.99224
M.4.4 11 70,261 91,293 1.2993 114.5 55.8 0.4873 0.99837
M.4.5 11 63,156 73,705 1.167 113.2 84.6 0.7473 0.99821
M.5.1 4 201,405 230,290 1.1434 83.8 37.1 0.4427 0.99958
M.5.2 68 15,871 24,941 1.5715 115.44 48.34 0.4187 0.99278
M.5.3 27 39,281 71,140 1.8111 84.07 32.64 0.3882 0.99786
M.5.4 8 131,891 141,120 1.07 100.6 39.2 0.3897 0.99924
M.5.5 17 64,366 124,209 1.9297 129.7 41.5 0.32 0.99799
M.5.6 12 90,776 157,852 1.7389 121.7 61.3 0.5037 0.99866
M.6.1 1 – – – 50 – – –
M.6.2 104 10,292 16,883 1.6404 122.4 69.89 0.571 0.98825
M.6.3 6 111,712 184,003 1.6471 111.7 37.6 0.3366 0.999
M.7.1 31 29,289 53,679 1.8327 100.81 42.84 0.425 0.99657
M.7.2 13 74,572 117,437 1.5748 168.1 120.3 0.7156 0.99775
M.7.3 5 208,984 213,680 1.0225 153 52.4 0.3424 0.99999
W.S.1 447 2321 4180 1.8009 133.83 78.94 0.5899 0.94548
W.S.2 125 8477 12,647 1.4919 113 58.43 0.5171 0.98685
W.S.3 55 18,468 27,583 1.4936 138.18 69.59 0.5036 0.99257
W.S.4 338 3082 5681 1.8433 124.08 67.92 0.5474 0.9613
W.S.5 136 7956 14,825 1.8634 109.74 47.13 0.4295 0.98639
W.S.6 111 9636 16,154 1.6764 121.17 68.48 0.5652 0.98758
W.S.7 49 21,217 44,234 2.0848 124 77.2 0.6226 0.99419
Line 1261 748.3 1754.3 2.3444 125.25 70.24 0.5608 0.85662

by the industry records’ for the same period on the production line It is evident from Tables 3–6 that the failure mode deserving the
and it confirms the effectiveness of the applied method we used. greatest attention, because it considerably restricts the equipment
(d) The CVs for the TTF at all levels are greater than one, implying availability (98.69%), is the F.1.2.12. The latter refers to failures of
that the TTF have high variability except for the M1.4, M3.2, and M4.1 blower. The second in order lowest availability (99.17%) is due to
the CVs of which are lower than one. On the other hand, the CVs for general type failures of blower (F.1.2.1). The term ‘general type
the TTR at all levels are less than one, meaning that the TTR have failures’ refers to machine calibration and cleaning. The very low
low variability, therefore the maintenance staff spends approxi- availability is due to the rather high average TTR (269.3 min). This
mately the same time to repair a failure of the same mode. (e) machine heats the pre-forms by means of bulbs in the oven. Occa-
The sample size of certain machines is very small. In the cases of sionally, some pre-forms melt due to excessive heating and as a re-
machines M3.3 and M6.1 there was only one failure, so the sample sult the oven gets contaminated.
size is still too small to provide any reliable information about Machine cleaning is a painful and time consuming process. If
the data, especially for TTF. (f) Although the production line pre- one looks carefully at Table 3 he may realize that F.1.2.1 takes place
sents a failure every 748.3 min or 12.5 h or equivalent to two-fail- 20-times and the entire working period under study (928 working
ures per day approximately, the mean TTR a failure is 125.25 min days). Should one divide this entire working period (501 day-
or almost two hours per failure. s16 h/day + 427 days24 h/day = 18,264 h) with number of fail-
ures (20) the TTF = 18,264/20 = 913.2 h (or 54,792 min) instead of
32,542 min as given in Table 3. A closer look at the data reveals
5. Determination of the important failure modes that 20 failures of F.1.2.1 do occur in less than two first years in-
stead of the four-years in study. This rational conclusion explains
To optimize the operation of the production line, the aim is to the low average TTF of 32,542 min of Table 3. Following a discus-
proceed in improvements. The best way is to begin with the fail- sion with the technicians of the production line it was concluded
ures that have the most important negative impact on the smooth that F.1.2.1 was a failure which was finally corrected. To be more
operation of the line. To determine the most important failures specific there was a problem in oven bulb circuit. The latter were
than all the failures in level of failure modes, we isolate those that overheated and pre-forms occasionally melted. After this problem
occur more than ten times. Table 3 displays the descriptive statis- was addressed (by a specialized technician) it never occurred
tics of the most failure modes at machine level with frequency again. The third lowest availability (99.47%) which will have to
higher than ten times. In Table 4 one can identify the top ten-most be taken into account is attributed to F.1.2.13 and regards the
important failure modes, among those failures that follow certain failures in valve pre-forms of blower. The rest of the failures are
criteria: (a) smallest mean TTF, largest CV of TTF, largest of mean classified in a similar way.
TTR, and smallest availability. The description of those failure Table 4 shows that some failures, though they often occur, are
modes appears in Table 5. not included among the ten-most important failures with the
484 P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488

Table 3
Descriptive statistics of the most important failure modes at machine level.

TTF TTR
N Mean SD CV Mean SD CV Availability
F.1.1.4 14 59,158 73,339 1.2397 161.8 119 0.7367 0.99727
F.1.2.1 20 32,542 47,685 1.4653 269.3 108 0.4014 0.99179
F.1.2.2 32 31,544 32,858 1.0417 166.6 104 0.6261 0.99475
F.1.2.3 25 42,283 49,991 1.1823 97.8 33.7 0.3443 0.99769
F.1.2.4 10 72,912 82,985 1.1382 154 75.4 0.4896 0.99789
F.1.2.5 36 25,962 29,154 1.1229 143 73.2 0.5122 0.99453
F.1.2.6 39 27,974 31,775 1.1359 123 58 0.473 0.99563
F.1.2.9 29 26,594 46,344 1.7426 126 60.5 0.4821 0.9953
F.1.2.12 124 8410 18,739 2.2282 112 60.7 0.5422 0.98686
F.1.2.13 51 19,892 31,081 1.5625 107 57.3 0.536 0.99466
F.2.1.1 11 88,122 155,384 1.7633 116 45.9 0.3974 0.99869
F.2.1.2 16 62,338 90,369 1.4497 133 73.8 0.557 0.99788
F.2.1.4 15 37,471 68,740 1.8345 101 33.9 0.3362 0.99732
F.2.1.6 20 53,553 73,741 1.377 126 77.9 0.6207 0.99766
F.2.1.9 12 73,886 113,540 1.5367 140 67.3 0.4793 0.9981
F.2.1.10 22 42,485 88,526 2.0837 88.4 39.4 0.4458 0.99792
F.3.1.4 17 39,281 56,590 1.4406 152.1 87 0.572 0.99614
F.3.1.5 11 89,808 95,020 1.058 130 29.3 0.2256 0.99855
F.4.2.1 16 35,515 45,565 1.283 173.8 94.6 0.5443 0.99513
F.4.2.2 34 30,551 39,050 1.2782 140 83.5 0.596 0.99544
F.4.2.3 67 16,030 54,378 3.3923 123 28.8 0.2349 0.9924
F.4.2.6 22 46,610 55,350 1.1875 156.4 101 0.6464 0.99666
F.4.2.8 14 51,549 64,837 1.2578 168.2 120 0.7105 0.99675
F.4.2.9 49 21,949 46,077 2.0993 107 42.6 0.3983 0.99515
F.4.3.1 10 1E + 05 78,876 0.7362 160.5 77.6 0.4835 0.9985
F.4.3.4 18 58,090 100,142 1.7239 99.4 39.6 0.3985 0.99829
F.4.3.5 39 27,026 57,912 2.1428 108 71.3 0.659 0.99601
F.5.2.3 41 26,398 38,655 1.4643 114 48.5 0.4261 0.9957
F.5.2.4 18 59,840 101,874 1.7024 118 50.4 0.426 0.99803
F.6.2.1 14 67,401 110,040 1.6326 146 90.9 0.6239 0.99784
F.6.2.3 26 41,522 88,516 2.1318 113 57.1 0.5058 0.99729
F.6.2.5 11 92,387 111,519 1.2071 115 72.9 0.6367 0.99876
F.6.2.8 23 43,067 38,403 0.8917 138 92.8 0.6725 0.99681
F.7.1.2 14 64,988 100,616 1.5482 88.9 29.8 0.3345 0.99863

Table 4 Table 5
The ten-most important failure modes under several criteria. Description of the most important failure modes.

Smallest Largest Largest Smallest Failure modes Description


Mean TTF CV of TTF Mean TTR Availability
F.1.2.1 General failures of blower
F.1.2.12 F.4.2.3 F.1.2.1 F.1.2.12 F.1.2.2 Hydraulic failures of blower
F.4.2.3 F.1.2.12 F.4.2.1 F.1.2.1 F.1.2.5 Failures in pistons of blower
F.1.2.13 F.4.3.5 F.4.2.8 F.4.2.3 F.1.2.6 Mechanical failures of blower
F.4.2.9 F.6.2.3 F.1.2.2 F.1.2.5 F.1.2.9 Failures in furnace of blower
F.1.2.5 F.4.2.9 F.1.1.4 F.1.2.13 F.1.2.12 Failures in matrices of blower
F.5.2.3 F.2.1.10 F.4.3.1 F.1.2.2 F.1.2.13 Failures in valves of blower
F.1.2.9 F.2.1.4 F.4.2.6 F.4.2.1 F.2.1.1 General failures of labeler
F.4.3.5 F.2.1.1 F.1.2.4 F.4.2.9 F.2.1.4 Labeler glue cylinder
F.1.2.6 F.1.2.9 F.3.1.4 F.1.2.9 F.2.1.10 Failures in photo cell and inductive coil
F.4.2.2 F.4.3.4 F.6.2.1 F.4.2.2 F.4.2.1 General failures of filler
F.4.2.2 Hydraulic failures of filler
F.4.2.3 Pneumatic failures of filler
F.4.2.9 Failures in filler nozzles
lowest availability because they have very short repair time. Rep- F.4.3.4 Failures in ‘‘guides” of stopper/closure
F.4.3.5 Failures in upper part of stopper/closure
resentative examples are the mechanical nature failures of blower
F.5.2.3 Failures in labeler guides and conveyor belts
(F.1.2.6), failures in caping machine (F.4.3.5), and the failure in con- F.6.2.3 Electrical failures of palletizer
veyor belts of labeling machine (F.5.2.3).
Another observation is that there are failures with long repair-
ing times which are not included in Table 4 with the lowest avail- the empirical and the theoretical distributions. The first approach
ability because their occurrence is low. Typical examples are is a method that derives the empirical distributions directly from
mechanical failures in the filling machine (F.4.2.8), and electrical the failure data, and does not require the estimation of the distri-
failures occurring in compressor air machine (F.1.1.4). bution’s parameters. The second approach focuses on identifying
the candidate theoretical distribution, estimating the parameters
and performing a goodness-of-fit test. Usually, a theoretical distri-
6. Identification of failure distribution data bution is preferred to empirical distribution because: (a) the
empiric model does not provide any further information beyond
The determination of the failure data distributions per machine, a small segment of data of the sample under examination, (b) the
workstation, and entire line level is essential. There are two ap- determination of the basic probabilistic nature of the failure pro-
proaches to identify the distributions for the failure–repair data; cess is of great interest. The sample is very small (random), subset
P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488 485

Table 6
The Anderson–Darling statistic for TTF and TTR at all levels.

TTF TTR
N Weibull Lognormal Exponential Normal Weibull Lognormal Exponential Normal
F.1.2.1 20 1.074 1.549 2.731 2.814 1.373 1.108 4.548 1.417
F.1.2.2 32 0.821 0.910 0.829 2.091 1.172 0.831 3.240 1.883
F.1.2.5 36 0.713 1.078 1.956 2.299 1.355 0.890 4.935 1.981
F.1.2.6 39 0.520 0.797 1.022 2.921 1.132 0.922 5.955 1.491
F.1.2.9 29 1.204 0.831 6.445 5.394 1.133 0.952 4.639 1.427
F.1.2.12 124 2.570 0.527 28.99 23.059 2.586 0.952 15.433 4.606
F.1.2.13 51 0.835 0.623 7.027 6.145 1.806 1.049 7.346 2.582
F.2.1.1 11 1.444 1.553 2.215 3.051 1.619 1.649 2.971 1.658
F.2.1.4 15 1.195 1.204 5.157 2.999 1.276 1.372 3.843 1.290
F.2.1.10 22 1.194 0.810 7.910 5.065 0.926 1.061 3.776 0.995
F.4.2.1 16 1.098 1.072 1.648 2.277 1.486 1.248 2.872 1.835
F.4.2.2 34 0.803 1.411 4.166 2.879 1.383 0.754 4.519 2.113
F.4.2.3 67 1.987 1.531 32.763 17.799 8.041 5.134 21.486 6.463
F.4.2.9 49 1.426 0.830 8.558 9.304 2.878 1.758 10.865 2.86
F.4.3.4 18 0.980 1.068 2.641 3.099 1.058 1.183 3.933 1.074
F.4.3.5 39 0.957 0.904 5.114 7.165 2.522 1.017 6.318 3.560
F.5.2.3 41 0.718 0.811 4.270 4.671 0.983 0.602 7.056 1.275
F.6.2.3 26 0.772 0.857 8.912 5.201 1.230 0.884 4.378 1.547
M.1.1 49 1.072 0.603 8.603 7.570 3.260 1.538 8.371 4.353
M.1.2 384 1.538 1.216 39.386 51.008 8.730 3.022 40.747 17.086
M.l.3 8 1.756 1.875 2.168 2.101 2.225 2.027 3.403 2.279
M.l.4 6 2.176 2.244 2.654 2.184 2.480 2.580 2.978 2.423
M.2.1 118 0.640 1.186 13.047 11.883 3.466 1.326 17.204 5.215
M.2.2 7 2.064 2.00 5.314 2.765 1.856 1.920 2.680 1.865
M.3.1 45 0.735 1.570 5.923 5.872 2.238 1.334 6.914 3.155
M.3.2 9 1.712 1.834 1.809 1.745 1.609 1.577 2.808 1.671
M.3.3 1 – – – – – – – –
M.4.1 21 1.166 1.308 1.169 1.554 1.156 1.074 3.068 1.459
M.4.2 220 0.899 2.232 29.658 35.233 14.49 5.227 41.079 18.044
M.4.3 75 1.524 0.508 8.659 12.469 2.433 0.808 10.498 3.706
M.4.4 11 1.626 1.537 2.257 2.369 1.617 1.566 2.863 1.741
M.4.5 11 1.462 1.768 1.882 1.975 2.409 1.986 2.936 2.947
M.5.1 4 3.030 3.036 3.436 3.032 3.037 3.105 3.394 3.006
M.5.2 68 0.724 0.728 1.707 8.688 1.186 0.655 11.333 1.651
M.5.3 27 0.757 1.093 4.105 4.084 1.525 1.181 5.583 1.750
M.5.4 8 1.686 1.880 1.681 2.018 1.992 2.053 3.025 2.025
M.5.5 17 1.187 0.984 5.861 3.516 1.381 1.678 4.348 1.391
M.5.6 12 1.677 1.508 4.195 3.051 1.470 1.396 2.513 1.617
M.6.1 1        
M.6.2 104 0.487 1.380 9.233 13.476 3.460 1.273 14.406 5.421
M.6.3 6 2.155 2.125 2.656 2.880 2.295 2.371 3.171 2.265
M.7.1 31 0.698 0.725 5.284 4.563 0.683 0.890 5.189 0.765
M.7.2 13 1.621 1.507 3.674 2.641 2.431 2.168 2.909 2.976
M.7.3 5 2.487 2.569 2.485 2.635 2.426 2.437 3.187 2.443
W.S.1 447 1.617 1.619 41.232 60.286 10.613 3.203 50.492 20.163
W.S.2 125 0.782 1.078 16.050 13.289 3.447 1.341 18.032 5.232
W.S.3 55 0.641 1.641 4.696 6.151 2.171 1.309 8.129 3.136
W.S.4 338 0.894 3.245 29.262 46.908 15.858 4.907 55.19 21.61
W.S.5 136 0.832 0.667 19.552 20.343 1.933 1.161 21.301 2.900
W.S.6 111 0.534 1.607 9.932 14.905 3.576 1.367 15.511 5.530
W.S.7 49 1.110 0.718 7.312 8.574 2.326 1.016 7.076 3.517
Line 1261 8.344 11.178 107.121 214.221 35.478 9.111 169.559 58.596

of the population of failure times, and it is of interest to find the mon theoretical distributions i.e. Weibull, lognormal, exponential
distribution wherefrom the sample originated, (c) often the failure and normal distributions are given in Table 6. Table 6 shows that:
process is the result of some phenomenon described by certain (a) the production line for the TTF follows the Weibull distribution.
concrete distribution, and (d) small samples provide very little Moreover, all the workstations follow the Weibull distribution ex-
information with regard to the failure process. However, if the cept the WS5 and WS7 that comply with the lognormal distribu-
sample coincides with a theoretical distribution then one can ac- tion. The TTFs at machine and failure levels exhibit mixed
quire much more powerful results based on the attributes of the tendencies and follow the Weibull and lognormal distributions.
theoretical distribution (Ebeling, 1997). (b) On the other hand, the TTR at line and workstation levels follow
The identification of the candidate distributions from the theo- the lognormal distribution. The TTRs at machine and failure levels
retical distributions is a process that requires knowledge of theo- had an inclination to a lognormal distribution rather than to Wei-
retical distributions and their parameters as well as statistical bull distribution. The machines M.1.4, M.5.1 and M.6.3 that follow
analysis of data. To estimate the candidate theoretical distributions the normal distribution, have statistics AD (2.423, 3.006 and 2.265,
the least-square fit was applied per candidate distribution and the respectively), being very close to the Weibull (2.48, 3.037 and
Anderson–Darling statistics (AD) were determined by using the 2.295), and consequently they will follow the Weibull distribution.
software package MINITAB. A smaller is the AD statistic indicates The parameters of the Weibull and lognormal distribution for
that the distribution fits the data better. The Anderson–Darling sta- the TTF and TTR at machine, workstation, entire line, and for sev-
tistics of the failure and repair data at all levels, for the most com- eral important failure modes are summarized in Table 7.
486 P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488

Table 7
Weibull and lognormal parameters for TTF and TTR for several important failure modes, all the machines, workstations, and the entire line.

TTF TTR
Weibull Lognormal Weibull Lognormal
Scale parameter Shape parameter Location parameter Scale parameter Scale parameter Shape parameter Location parameter 1 Scale parameter
h b l s h b l s
F.1.2.1 22879.4 0.59783 0.380472 5.52425
F.1.2.2 31332.4 0.984844 0.584857 4.94172
F.1.2.5 22351.3 0.760345 0.533853 4.83006
F.1.2.6 25765.8 0.853625 0.452826 4.70673
F.1.2.9 1.78503 8.86627 0.459296 4.72705
F.1.2.12 1.67793 7.66994 0.509859 4.58771
F.1.2.13 1.70121 8.72753 0.464771 4.55647
F.2.1.1 62975.7 0.647208 129.995 2.94571
F.2.1.4 20601.6 0.540096 112.286 3.45194
F.2.1.10 1.86381 9.04738 99.8974 2.47709
F.4.2.1 1.5566 9.58765 0.48218 5.03547
F.4.2.2 21856.9 0.601342 0.506412 4.80598
F.4.2.3 2.21461 7.6496 0.206906 4.78698
F.4.2.9 1.66169 8.75191 0.35439 4.60847
F.4.3.4 42171.8 0.663128 111.658 2.75549
F.4.3.5 1.54563 9.06042 0.454861 4.56178
F.5.2.3 20069.7 0.6809 0.430678 4.64641
F.6.2.3 19910.0 0.50125 0.478237 4.61449
M.1.1 1.71724 8.72085 0.441436 4.84544
M.1.2 1.73879 6.76705 0.545477 4.74044
M.l.3 62406.4 0.65185 0.262839 4.71845
M.l.4 172,541 2.14428 134.702 3.18478
M.2.1 6566.03 0.6466 0.475967 4.62562
M.2.2 2.34729 9.75781 99.9942 2.55386
M.3.1 15152.3 0.599425 0.473296 4.8387
M.3.2 100,855 0.887005 0.434132 4.7908
M.3.3 – – – – – – – –
M.4.1 51199.2 0.997546 0.509697 4.58184
M.4.2 3159.16 0.613806 0.407055 4.77928
M.4.3 1.43846 8.50582 0.494705 4.58317
M.4.4 1.33537 10.3054 0.458659 4.63838
M.4.5 51,153 0.686837 0.476331 4.58308
M.5.1 139,751 0.56712 94.0482 2.97196
M.5.2 11,608 0.662164 0.425782 4.66138
M.5.3 25732.3 0.592647 0.357421 4.36544
M.5.4 123,042 0.858454 112.895 2.9487
M.5.5 1.85726 9.51868 144.108 3.74882
M.5.6 1.5786 10.1254 0.501413 4.67938
M.6.1 – – – – – – – –
M.6.2 7546.1 0.658999 0.503462 4.67822
M.6.3 1.6592 10.483 123.797 3.87657
M.7.1 18976.2 0.601818 113.75 2.57787
M.7.2 1.5514 10.0507 0.500145 4.96924
M.7.3 205,391 0.958762 170.148 3.63387
W.S.1 1704.01 0.673005 0.53094 4.75104
W.S.2 6013.6 0.632564 0.480816 4.61149
W.S.3 13,842 0.659212 0.479062 4.81402
W.S.4 2238.11 0.662838 0.448244 4.71253
W.S.5 7.70289 1.84368 0.435676 4.60590
W.S.6 7019.7 0.655774 0.500195 4.67007
W.S.7 1.68709 8.74699 0.507188 4.68422
Line 542.663 0.678844 0.492478 4.70425

7. Reliability, failure rate and maintainability, repair rate Reliability is the probability that the entire line (workstation or
models machine) will perform a required function, under stated operating
conditions, for a given period of time t. T indicates the TTF of the
The juice bottling production line, as analyzed above, exhibits line (workstation or machine); T P 0, then the reliability can be
availability and efficiency at 85.66% and 82.10%, respectively. It is expressed as, RðtÞ ¼ PðT P tÞ. Furthermore, the unreliability func-
well known that a production line is profitable when the produc- tion, is indicated as Q(t), and the probability of failure in t,
tion rate is higher than 85%. Therefore, the juice production line Q ðtÞ ¼ 1  RðtÞ ¼ PðT 6 tÞ. In reliability theory, the hazard or fail-
under investigation operates negatively. To avoid the negative ure rate function is indicated as, kðtÞ ¼ f ðtÞ=RðtÞ where f(t) is the
operation of the line, one may increase the actual production rate probability density function of the failure distribution. Maintain-
of the line, by applying good operation management (GOM). There- ability is the probability that a failed workstation or machine or
fore, one can reduce the downtimes of the line by prolonging the entire line will be restored to operational effectiveness within a
TTF and minimizing the repair time (see Eq. (1)). Thus, the reliabil- period of time when the repair is performed in accordance to the
ity and maintainability of the juice bottle production line were prescribed procedures. In other words, it is the probability of repair
examined. in a given time. The repair time includes access time, diagnosis
P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488 487

" #
time, spare part supply, replacement time, checkout time, and 1 ðln t  lÞ2
alignment time. rðtÞ ¼ pffiffiffiffiffiffiffi exp  ; t0 ð10Þ
st 2p 2s2
For the TTF it was found that the best fit at workstation and line
levels from the failure data is provided by the Weibull distribution The probability of a repair being completed in time t for the
except for WS5 and WS7 that follow the lognormal distribution (see juice production line is the cumulative function for repair, as
Table 6). The parameters of the Weibull distribution are shape (b) follows:
and scale (h). The scale parameter (SP) of the Weibull distribution  
(WD) influences both the mean and spread of the distribution. 1 t
Repair ðtÞ ¼ PrðT 6 tÞ ¼ U ln ð11Þ
Moreover, SP is called the characteristic life and has units identical s expðlÞ
to those of TTF. The parameter h equals to the duration of useful
time t that is the 63.2% of failures will have occurred at machine, where U () is the distribution function of the standard normal
workstation or entire line. The shape parameter of the Weibull dis- distribution.
tribution provides an insight into the behavior of the failure pro- The mean TTR, variance, mode and median of the line are shown
cess (or repair process). Furthermore, it affects the distribution below:
for several different values: (a) for b < 1; a decreasing failure rate
mean TTR ¼ expðl þ s2 =2Þ ð12Þ
function, (b) for b > 1; an increasing failure rate function, and (c)
for b = 1; a constant failure rate function and the distribution is 2 2 2
r ¼ expð2l þ s Þ½expðs Þ  1 ð13Þ
identical to the exponential, with k = 1/h. When 1 < b < 2, then 2
the increasing failure rate is concave. When b = 2, then the increas- tmode ¼ expðl  s Þ ð14Þ
ing failure rate is linear. When b > 2, then the increasing failure rate tmedian ¼ expðlÞ ð15Þ
is convex, and when 3 6 b 6 4, then the increasing failure rate ap-
proaches the normal distribution (Ebeling, 1997). The repair rate function of the production line is given by Eq.
Having identified the failure distribution from data and esti- (16):
mated the Weibull–lognormal parameters (see Table 7), the reli- h i
tlÞ2
ability and failure rate models for juice production line for all rðtÞ p1ffiffiffiffi
st 2p
exp  ðln 2s 2
kðtÞ ¼ ¼   ; t0 ð16Þ
workstations and the entire line were calculated, as follows: Repair ðtÞ U 1s ln expðt lÞ
"  0:6730 #   0:6731
t 0:673 t
RWS1 ðtÞ ¼ exp  ; kWS1 ðtÞ ¼ where l is the mean of the lognormal distribution of the repair
1704:01 1704:01 1704:01
times, and s is the standard deviation with which the lognormal dis-
ð2Þ tribution of the repair times is spread around the mean. The l and s
"  0:6325 #   0:63251
t 0:6325 t for TTR at workstation, and entire line are summarized in Table 7.
RWS2 ðtÞ ¼ exp  ; kWS2 ðtÞ ¼
6013:6 6013:6 6013:6 Then, one can calculate the maintainability and repair rate models
for the production line at all workstations and the entire line as
ð3Þ
"  0:6592 #   0:65921
follows:
t 0:6592 t
RWS3 ðtÞ ¼ exp  ; kWS3 ðtÞ ¼  
13; 842 13; 842 13; 842 1 t
Repair;WS1 ðtÞ ¼ U ln ;
ð4Þ 4:751 expð0:531Þ
"  0:6628 #   0:66281 h i
t0:531Þ2
RWS4 ðtÞ ¼ exp 
t
; kWS4 ðtÞ ¼
0:6628 t 1 pffiffiffiffi
4:751t 2p
exp  ðln2ð4:751Þ 2
2238:11 2238:11 2238:11 kWS1 ðtÞ ¼   ð17Þ
1 t
U 4:751 ln expð0:531Þ
ð5Þ
   
7:702  ln t 1 t
RWS5 ðtÞ ¼ U ; Repair;WS2 ðtÞ ¼ U ln ;
1:8436 4:611 expð0:481Þ
h i h i
t7:702Þ2
1 pffiffiffiffi
exp  ðln2ð1:843Þ t0:481Þ2
kWS5 ðtÞ ¼
1:843t 2p
7:702ln t
2
ð6Þ
1 pffiffiffiffi
4:611t 2p
exp  ðln2ð4:611Þ 2
U 1:843
kWS2 ðtÞ ¼   ð18Þ
"  # 1 t
0:6557   0:65571 U 4:611 ln expð0:481Þ
t 0:6557 t
RWS6 ðtÞ ¼ exp  ; kWS6 ðtÞ ¼  
7019:7 7019:7 7019:7 1 t
Repair;WS3 ðtÞ ¼ U ln ;
ð7Þ 4:814 expð0:479Þ
  h i
1:687  ln t t0:479Þ2
RWS7 ðtÞ ¼ U ;
1 pffiffiffiffi
4:814t 2p
exp  ðln2ð4:814Þ 2
8:7469 kWS3 ðtÞ ¼   ð19Þ
h i 1 t
1 pffiffiffiffi t1:687Þ2
exp  ðln2ð8:7469Þ 2
U 4:814 ln expð0:479Þ
8:7469t 2p
kWS7 ðtÞ ¼ 1:687ln t ð8Þ  
U 8:7469 1 t
"  0:6788 #   0:67881 Repair;WS4 ðtÞ ¼ U ln ;
t 0:6788 t 4:712 expð0:448Þ
RLine ðtÞ ¼ exp  ; kLine ðtÞ ¼ h i
542:663 542:663 542:663 t0:448Þ2
1 pffiffiffiffi
4:712t 2p
exp  ðln2ð4:712Þ 2
ð9Þ kWS4 ðtÞ ¼   ð20Þ
1 t
U 4:712 ln expð0:448Þ
TTR stands for the entire production line following the lognor-
 
mal failure low, and T is assumed as the continuous random 1 t
Repair;WS5 ðtÞ ¼ U ln ;
variable representing the time to repair. The time to repair (T) of 4:605 expð0:435Þ
the production line will be lognormally distributed if Q = ln (T) is h i
t0:435Þ2
normally distributed. If l and s2 are the mean and variance of Q,
1 pffiffiffiffi
4:605t 2p
exp  ðln2ð4:605Þ 2

respectively, the corresponding probability density function with kWS5 ðtÞ ¼   ð21Þ
1 t
U 4:605 ln expð0:435Þ
respect to repair time is expressed by:
488 P.H. Tsarouhas et al. / Journal of Food Engineering 95 (2009) 479–488

 
1 t failure modes of the production process due to the complex-
Repair;WS6 ðtÞ ¼ U ln ;
4:670 expð0:50Þ ity of equipment with 28 and 30 different failure modes,
h i
t0:50Þ2 respectively.
1 pffiffiffiffi
4:670t 2p
exp  ðln2ð4:670Þ 2
kWS6 ðtÞ ¼   ð22Þ (g) The machines with the lowest availability are M1.2, M4.2,
1 t
U 4:670 ln expð0:50Þ M2.1, and M6.2 that have 95.30%, 97.34%, 98.74%, and
  98.82%, respectively. Consequently, the statistical analysis
1 t
Repair;WS7 ðtÞ ¼ U ln ; in conjunction with the reliability and maintainability
4:684 expð0:507Þ models that were calculated may estimate the cons of the
h i
t0:507Þ2
1 pffiffiffiffi
4:684t 2p
exp  ðln2ð4:684Þ 2
production line thereby intervening to improve the mainte-
kWS7 ðtÞ ¼   ð23Þ nance strategy and increase the performance and the
1 t
U 4:684 ln expð0:507Þ production rate of the line.
 
1 t
RLine ðtÞ ¼ U ln ;
4:704 expð0:492Þ
h i References
t0:492Þ2
1 pffiffiffiffi
4:704t 2p
exp  ðln2ð4:704Þ 2
kLine ðtÞ ¼   ð24Þ Ben-Daya, M., Duffuaa, S.O., 1995. Maintenance and quality: the missing link.
1 t
U 4:704 ln expð0:492Þ Journal of Quality in Maintenance Engineering 1 (1), 20–26.
Blischke, W.R., Murthy, D.N.P., 2000. Reliability: Modelling, Prediction and
Optimization. Wiley, New York.
Blischke, W.R., Murthy, D.N.P., 2003. Case Studies in Reliability and Maintenance.
8. Conclusions John Wiley and Sons Inc., Hoboken, New Jersey. pp. 351–445.
Ebeling, C.E., 1997. An Introduction to Reliability and Maintainability Engineering.
McGraw-Hill, New York.
It was pointed out that
Eliashberg, J., Singpurwalla, N.D., Wilson, S.P., 1997. Calculating the reserve for a
time and usage indexed warranty. Management Science 43 (7), 966–975.
(a) the availability of the juice production line is 85.66% and the Koren, Y., Hu, S.J., Weber, T., 1998. Impact of manufacturing system configuration on
performance. Annals of the CIRP 47, 369–372.
efficiency amounts to 82.10% because of equipment’s
Little, J.D.C., 1961. A proof of the queueing formula L = kW. Operations Research 9,
failures. 383–387.
(b) The production line presents a failure every 12.5 h or equiv- Montgomery, D.C., 1985. Introduction to Statistical Control. John Wiley and Sons
alent two-failures per day approximately, while the mean Inc., New York, NY. pp. 69–85.
Inman, R.R., 1999. Empirical evaluation of exponential and independence
TTR is approximately two hours per failure. assumptions in queuing models of manufacturing systems. Production and
(c) The TTF of the production, at line level, follows the Weibull Operations Management 8 (4), 409–432.
distribution. All the workstations follow the Weibull distri- Ireson, W.G., Coombs, C.F., Moss, R.Y., 1996. Handbook of Reliability Engineering
and Management. McGraw-Hill, New York.
bution except for the WS5 and WS7 that follow the lognor- Kourtis, L.K., Arvanitoyannis, I.S., 2001. Implementation of HACCP system to the
mal distribution. The TTFs at machine and failure levels drink industry: II. Non-alcoholic beverages. Food Reviews International 17 (4),
display mixed tendencies and follow the Weibull and log- 451–486.
Usher, J.M., Roy, U., Parsaei, H.R., 1998. Integrated Product and Process
normal distributions. Development. Wiley, Hoboken, NJ.
(d) The TTR at line and workstation levels follow the lognormal Liberopoulos, G., Tsarouhas, P., 2002. Systems analysis speeds up Chipita’s food
distribution. The TTRs at machine and failure levels have an processing line. Interfaces 32 (3), 62–76.
Liberopoulos, G., Tsarouhas, P., 2005. Reliability analysis of an automated pizza
inclination to a lognormal distribution rather than to
processing line. Journal of Food Engineering 69 (1), 79–96.
Weibull distribution. Seifoddini, S., Djassemi, M., 2001. The effect of reliability consideration on the
(e) The CVs for the TTF at workstation and entire line levels are application of quality index. Computers and Industrial Engineering 40 (1–2),
65–77.
greater than one, meaning that the TTF have high variability.
Tsarouhas, P., Varzakas, T., Arvanitoyannis, I., 2009a. Reliability and maintainability
On the other hand the CVs for the TTR at all levels are less analysis of strudel production line with experimental data; a case study. Journal
than one, implying that the TTR have low variability, there- of Food Engineering 91, 250–259.
fore the maintenance staff spends approximately the same Tsarouhas, P., Arvanitoyannis, I., Varzakas, T., 2009b. Reliability and maintainability
analysis of cheese (feta) production line in a Greek medium-size company: a
time to repair a failure of the same mode. case study. Journal of Food Engineering 94, 233–240.
(f) The WS1 and WS4 have the highest frequency of failures and Wang, Y., Jia, Y., Yu, J., Yi, S., 1999. Failure probabilistic model of CNC lathes.
the lowest availability with 94.54% and 96.13%, respectively. Reliability Engineering and System Safety 65 (1), 307–314.
Furthermore, they display the largest number of different

You might also like