Beruflich Dokumente
Kultur Dokumente
Darja Noskieviov1,*
1 FMMI, VB-TU Ostrava, Ostrava - Poruba, Czech Republic
* Corresponding author E-mail: darja.noskievicova@vsb.cz
DOI: 10.5772/DOI
2013 Noskieviov et al.; licensee InTech. This is an open access article distributed under the terms of the Creative
Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly cited.
AbstractIdentificationoftheassignablecausesofprocess special(assignable)causes(theprocessisconsiderednot
variability and the restriction and elimination of their statisticallystable)usingcontrolcharts.Themaingoalof
influencearethemain goalsofstatisticalprocesscontrol SPC is the identification of abnormal variability caused
(SPC).Identificationofthesecausesisassociatedwithso by special (assignable) causes, with the aim to make the
calledtestsforspecialcausesorrunstests.Fromthetime process stable, to minimize process variability and to
of the formulation of the first set of such rules (Western improve the process performance. To meet these goals
Electric rules) several different sets have been created SPC must be built as a problemsolving instrument and
(Nelson rules, Boeing AQS rules, Trietsch rules). This the sequence of the subprocesses Outof control signal
paperdealswiththecomparisonanalysisofthesesetsof revelationRootcauseidentificationActionacceptance
rules, their basic statistical properties and the mistakes Verification of action must be the axis of the SPC
accompanyingtheirapplicationusingSWsupport.Atthe application. A control chart is the main SPC instrument
end of this paper some recommendations for the correct for the analysis of process variability over time. It is a
applicationoftherunstestsareformulated. graphical depiction of process variability and its natural
andunnaturalpatterns.Thecontrolchartdisplaysavalue
Keywords Statistical Process Control, Common Causes, of the quality characteristic that has been measured or
Special Causes, Runs Tests, Rules For Special Causes, somesamplestatisticsthathavebeencomputedfromthe
NaturalPatterns,UnnaturalPattern measured values in a sample versus the sample number
ortime.Thecentrallineinthecontrolchartrepresentsthe
averagevalueofthequalitycharacteristiccorresponding
1.Introduction tothestatisticallystableprocess(onlythenaturalpattern
is presented and only common causes influence the
SPC is an approach to process control that has been process). The control limits are set so that when the
widely used in both industrial and nonindustrial fields. process is statistically stable, nearly all the points in the
It is primarily a tool for understanding process control chart fall between them (for most applied 3
variability. SPC is based on Shewharts conception of control limits this is 99.73% of points). As long as the
process variability. This conception distinguishes the points are within the control limits and they show a
variability caused by obviously effected common causes natural random pattern, the process is considered
(the process is considered to be in a state of statistical statistically stable. But if some points are out of the
control [1]) from the variability caused by abnormal, controllimitsorthepointsinsidethelimitsshowanon
Table4.Analysedsetsofruntests
Therelationsbetweenunnaturalpatterns(seeTable1and2)
andanalysedruntestscanbefoundinTable5.
Unnaturalpattern Rule
Figure1.Definitionofthezones
Largeshift(strays,freaks) 1,2
Smallersustainedshift 3,4 AsitcanbeseeninTable4themostcomplex,intermsof
Trends 5,3 therateofcovering,ofalleightdefinedrulesareNelson
Stratification 6 rules [9]. In addition to the Western Electric set, Nelson
Mixture 8 definedaruntestfortheoscillationandtrends.InRule4
SystematicVariation 7 he changed the length of the run from eight to nine
Table5.Relationsbetweenunnaturalpatternsandrules points. As compared to the Western Electric rules the
sequenceoftheruleshadbeenchangedbyNelson,too.
Historically,thefirstruleforthedetectionoftheassignable The standard for the Shewhart control charts ISO 8258
causewasdevelopedShewhart[7](Rule1inTable4).As [10]haswhollycopiedtheNelsonsetofrules.Alsoin[2]
we can see, this basic rule (known as Criterion 1) is TrietschworkswiththeNelsonrulesbutinRules6,7and
incorporated into all sets covered by Table 4. The set of 8 he changes the length of the runs to reach better
Western Electric rules [8] added to the Shewhart criterion statisticalpropertiesofthesetests.
run tests, covering unnatural patterns such as smaller or
sustained shifts (called zone tests), mixtures, stratification
Boeing and GE [11, 12] have less complex sets of rules,
andsystematicvariation.Tobeabletoapplytheseteststhe
basedonthefirstfourWesternElectriczonerules.
regionbetweenthecontrollimitsinthecontrolchartmust
bedividedintosixzones,eachofonesigmawidthandthe 4.Determinationoftherootassignablecause
location of the predefined sequences of the points in
relation to these zones must be evaluated. As the normal Precise recognition of unnatural patterns will not be
distributionoftheappliedstatisticsissupposed,inzoneC effectivewhenwearenotabletoassigntherootcausesto
68.27%ofvalues,inzoneB27.18%ofvaluesandinzoneA the patterns. Determination of the root cause asks for a
4.28% of values are expected when the process is high level of experience and deep knowledge of the
statisticallystable(seeFigure1). analysed process. In spite of this crucial fact the generic
Rules 1 8 have a large advantage, the possible general
Rules2,3and4weredesignedforquickeridentification root causes linked to them are defined in many
of the shifts from the mean as compared toTest 1 (early publications(seeTable6).
warningindicator).
k 6.Recommendationsforapplicationoftherules
1 (1 a j ) (1)
j1 In the professional literature several strategies for
application of the rules for assignable causes can be
Whengraduallyapplying2,4,6,7and8additionalrules found.Theyhavemanydifferencesbutthemainideaisa
(supposethatiforeveryparticulartestwillbe0.0027), commonone:runtestsmustnotbeappliedroutinely;itis
graduallyriseswiththenumberofrules(=0,005fork=2, veryimpracticaltoapplyallknownrulessimultaneously
=0,011fork=4,=0,016fork=6,=0,019fork=7 because of the danger of an unacceptable rise in false
and =0,02 for k =8)ascomparedto = 0,0027for the alarms.
basic rule (one point out of the control limit in the
Shewhartcontrolchart).Inaddition,thisformulasupposes As can be found in the mentioned literature, some
independent rules. In practice this assumption is often authors differ in their ideas of what rules to apply
observed. For that reason supplementary rules must be routinely, but predominantly they recommend Rule 1 in
simultaneouslyappliedwithahighcautionbecauseofthe conjunctionwithRule4.Foradditionalsensitizingofthe
dangerofmakingtheproblemsolvingprocesscomplicated control charts some of the authors recommend Rules 2
(formoredetailedinformationsee[3]).
Every analysed SW was searched using many different approximation is valid and control limits are
standpoints such as: the term used for the unusual reasonablysymmetrical
pattern,thetermusedfortheruntests,explanationofthe thesameforcontrolchartpanduwithconstant
statistical base of the run tests, the different default run control
definitions compared to the default number of rules, the tests 5 and 7 work sufficiently well for
possibility of choosing the rules, the possibility of continuous distribution because of their
changing the length of the runs, the possibility of nonparametricnature.
changingthedefinitionofthezones,theindicationofthe In a situation when the control charts have
patterns in the control chart, the possibility of recording reasonable symmetrical control limits, select the
the assignable causes in the control chart, the possibility suitabletestaccordingtothefollowingschema:
of recording the actions in the control chart, warning Rules6and8applyatthebeginningoftheSPC
about the simultaneous use of the rules, warning about implementationtoverifyrationalsubgrouping
the necessity to have an extensive knowledge of the in the phase of statistical stability, make sure to
process, the types of control chart covered, the same startwithRules1and4
default set of rules for different control charts, if additional sensitizing of the control chart is
information on what control chart rules are applicable, necessary add Rules 2 and 3 (or only one of
the existence of the definition of the potential general them)
assignable causes, description of the rules and when your knowledge of the process has
interpretation of rules. The former analysis of the increased during the previous SPC
propertiesoftheanalysedSWshowedthatallSW,more implementationaddsomeoftheotherstandard
or less, suffer from some disadvantages that can lead a testsoryourownfittedtest;
littleexperiencedusertotheincorrectconclusionsandto test5isappliedsingularly(itsmarginalincrease
incorrect information for the process control. This can in sensitivity to real signals is more than offset
unfortunately result in ignoring these run tests in the bythegreaterincreaseinfalsealarms(see[13,p.
analysis of the process stability or the whole SPC 137])
applicationfailing.Forthatreasoninthenextcharterthe Rules 1 4 apply to both halves of the control chart
simple methodology for the correct control chart separately.
interpretationisformulated. When using two control charts (for instance xbar
andR)startwithananalysisofthedispersioncontrol
9.Recommendationsfortheapplicationofrules chart.
forunnaturalpatternsrecognition Rememberthatwheretheruntestisviolateditdoes
not always indicate where the process change had
Based on the analysis of the professional literature
occurred.
(Chapters 2 7) and the analysis of the selected SW
Work out the list of possible root causes of the
products (Chapter 8), the author of the paper proposed
identified unnatural patterns (see Table 6) with
themethodologyfortheapplicationofruntests,effective
adequateactions.
control chart interpretation and successful SPC
Set the root causes of the identified unnatural
application. This methodology can be described as
patterns.
follows:
Realize adequate actions to remove or reduce the
detectedassignable(uncommon)causesinfluence.
Beforeapplyingtherunstestsorthetestsforspecial
causes check the default set of tests for the selected 10.Conclusions
controlchartsintheusedSW.
Never routinely apply all runs tests or tests for Based on the theoretical background in the introductory
assignable causes even if they are defaulted in your chapters, this paper deals with the analysis and
SW. comparison of several known sets of rules for special
Check if the chosen control charts have reasonably causesidentification(Shewhartrule,WesternElectricrules,
symmetricalcontrollimitsbecauseinsuchasituation Nelson tests, ISO 2859 tests, special sets formulated by
thenextconclusionsarevalid: large companies [Boeing, GE] and the most new set by
alltests(Rule1Rule8)canbeappliedtothex Trietsch). In light of the conclusions of these analyses the
bar chart and the individuals chart supposing next multicriteria analysis of the selected SW products,
thenormaldistributionofthevariable withafocusontestsforspecialcausesidentification,were
Rules 1 4 can be applied to the dispersion carried out with the aim to reveal the advantages and
charts(R,s)withoutanymodificationwhenthe disadvantages of every discussed SW from the point of
samplesizeisfiveormore view of support for meeting the main goal of SPC. Based
the rules can be applied without any problems on these analyses the methodology for the correct and
tocontrolchartsnpandcsupposingthatnormal effectiveapplicationofthesetestswasthenformulated.