InvestigationofTetraceneDecompositionVariationUsing Taguchi MethodologyWilliamFox,KristenVogelsang,JohnHarris2,AndrewLathropIntroductionOver the past decade, it has become increasingly apparent that it is essential formilitarycustomerstomaintaincloseworkingrelationshipswithproductionfacilities and contractors toensure qualityperformance.TheArmamentResearch,Development,andEngineeringCenter(ARDEC)hashadalonghistoryofworkingcloselywiththeLakeCityArmyAmmunitionPlant.themainproducer of small caliberammunitionfortheU.S.Military.Recently,this plantexperienced difficulties in the production and testing of Tetracene,an initiatingexplosiveusedinammunitionprimers.Toresolvethisproblem,governmentandcontractorpersonnelworkedtogethertoinvestigateandmodel themanufacturingprocess of Tetracene.Asapartofthismodelingprocess,testmatrices weredevelopedusingtheTaguchi Method ofQuality Engineering.These modeling efforts resulted ina better understandingofthefactors effectingtheproduction,performance,and testing forTetracene, ultimately allowingtheplanttoavoid productiondelays.Tetracene is a compound usedinammunitionprimersto increase and providestabilityinthesensitivity of ignition.The original militaryspecifications (Mil-Spec)requireTetracenetobemanufacturedwitha"meltingpoint"ofbetween128-132degreesCelsius.Tetracenedoes not actually have a“melting point"butratheradecompositionpoint.(Decompositionpoint-isthepointinwhichthecompoundisseparated into its constituents throughchemicalreaction).Itisthisdecompositionpointthatwasofconcern.TheoriginofthisMil-Specrequirementisnotknownforcertain,exceptthat itwas instituted intheearly1960's.TheMil-SpecrequiresperiodictestingofTetracenesamplestoinsurethattheymeetthestandard.Duringoneof thesetests,thedecompositionpointresultssuddenlybeganadownwardtrendintotherangeof122-125degreesCelsius.Thisproblemthreatenedtoshutdownproductionofall5.56mmand7.62mmammunitionattheLakeCityArmyAmmunitionPlant (LCAAP).TheSmall CaliberAmmunitionBranchoftheU.S.ArmyArmamentResearch,Development&Engineering' U.S. Army ARDEC, Picatinny Arsenal, NJ2 Worked on article as a cadet math major.183
183 Investigation of Tetracene Decomposition Variation Using Taguchi Methodology William Fox, Kristen Vogelsang1 , John Harris2 , Andrew Lathrop2 Introduction Over the past decade, it has become increasingly apparent that it is essential for military customers to maintain close working relationships with production facilities and contractors to ensure quality performance. The Armament Research, Development, and Engineering Center (ARDEC) has had a long history of working closely with the Lake City Army Ammunition Plant, the main producer of small caliber ammunition for the U.S. Military. Recently, this plant experienced difficulties in the production and testing of Tetracene, an initiating explosive used in ammunition primers. To resolve this problem, government and contractor personnel worked together to investigate and model the manufacturing process of Tetracene. As a part of this modeling process, test matrices were developed using the Taguchi Method of Quality Engineering. These modeling efforts resulted in a better understanding of the factors effecting the production, performance, and testing for Tetracene, ultimately allowing the plant to avoid production delays. Tetracene is a compound used in ammunition primers to increase and provide stability in the sensitivity of ignition. The original military specifications (Mil-Spec) require Tetracene to be manufactured with a “melting point” of between 128-132 degrees Celsius. Tetracene does not actually have a “melting point” but rather a decomposition point. (Decomposition point- is the point in which the compound is separated into its constituents through chemical reaction). It is this decomposition point that was of concern. The origin of this Mil-Spec requirement is not known for certain, except that it was instituted in the early 1960’s. The Mil-Spec requires periodic testing of Tetracene samples to insure that they meet the standard. During one of these tests, the decomposition point results suddenly began a downward trend into the range of 122-125 degrees Celsius. This problem threatened to shut down production of all 5.56 mm and 7.62 mm ammunition at the Lake City Army Ammunition Plant (LCAAP). The Small Caliber Ammunition Branch of the U.S. Army Armament Research, Development & Engineering 1 U.S. Army ARDEC, Picatinny Arsenal, NJ 2 Worked on article as a cadet math major
Center(ARDEC)atPicatinnyArsenal wastaskedwiththeinvestigationandmodelingofthisTetracenedecompositionproblem.Wewilldiscussourmodel,adesignof experiments usedtotesttheTetraceneproduction,the analysis of Taguchimethods appliedtothisdesign, and theconclusionsfoundbythemodelers.Theseeffortsresultedinabettelunderstanding of thefactors effecting the production, performance, and testmethodologyforTetracene,ultimatelyallowingtheplantto avoidproductiondelays inthemanufacturingof small caliberammunition.ChangesweremadetotheMilitary-Specificationasaresult ofthisdesigntest.Production and TestingProceduresforTetraceneProductionTetraceneis created inamannerthatgreatly resembles alargecookingtypeproduction.Most of thework isaccomplished byhand byemployeesthathavebeenproducingTetraceneforyears.Themethodhaschangedlittleoverthepastseveraldecades,andduetotheexplosivenatureofthecompound,completeautomationwithelectronicdevicesisdifficult.Tetraceneisformedinathree step method:1.Mixasodiumnitritesolutionwithwater.2.Mix anamino guanidine bicarbonate (AGB)solution withwater.Slowlyaddsulfuricacidattherateof5Occperminutetoproduceaminoguanidinesulfate(AGS)3. Mix the sodium nitrite solution (from Step 1) with the AGS (from Step 2). Themixingisdonebya machine inalarge closed room.Alarge spatula-likearmgentlystirs (oragitates)themix.The reaction isheldtoatemperaturerangeof95-101degreesCelsiusforfivehours,atwhichtimethereactionisassumedtobecomplete.Duringthefivehourtimeperiod,Tetraceneisprecipitatedfromthemixture.This precipitate is filtered out and washed,then collected in conductivecontainers.TestingThepurposeof taking a decompositionpointreadingis that it serves as anindicator of the purity of the compound.An important part of determining thefactors affecting the decomposition point is the actual testing process itself. AMettlertestingdeviceisusedtocapturethetemperatureinwhichthecompounddecomposes.Tetraceneisadecomposingcompoundthatundergoesanexothermicreactionfromsolidstraighttoliquid.Becauseofthischaracteristic,ithasbeenverydifficulttomeasuretheexactdecompositionpoint.TheMettlerdecompositiontesting deviceelectronicallymeasuresthedecompositionpoint using a computer controlledrateof change.The operator184
184 Center (ARDEC) at Picatinny Arsenal was tasked with the investigation and modeling of this Tetracene decomposition problem. We will discuss our model, a design of experiments used to test the Tetracene production, the analysis of Taguchi methods applied to this design, and the conclusions found by the modelers. These efforts resulted in a better understanding of the factors effecting the production, performance, and test methodology for Tetracene, ultimately allowing the plant to avoid production delays in the manufacturing of small caliber ammunition. Changes were made to the Military-Specification as a result of this design test. Production and Testing Procedures for Tetracene Production Tetracene is created in a manner that greatly resembles a large cooking type production. Most of the work is accomplished by hand by employees that have been producing Tetracene for years. The method has changed little over the past several decades, and due to the explosive nature of the compound, complete automation with electronic devices is difficult. Tetracene is formed in a three step method: 1. Mix a sodium nitrite solution with water. 2. Mix an amino guanidine bicarbonate (AGB) solution with water. Slowly add sulfuric acid at the rate of 50cc per minute to produce amino guanidine sulfate (AGS). 3. Mix the sodium nitrite solution (from Step 1) with the AGS (from Step 2). The mixing is done by a machine in a large closed room. A large spatula-like arm gently stirs (or agitates) the mix. The reaction is held to a temperature range of 95-101 degrees Celsius for five hours, at which time the reaction is assumed to be complete. During the five hour time period, Tetracene is precipitated from the mixture. This precipitate is filtered out and washed, then collected in conductive containers. Testing The purpose of taking a decomposition point reading is that it serves as an indicator of the purity of the compound. An important part of determining the factors affecting the decomposition point is the actual testing process itself. A Mettler testing device is used to capture the temperature in which the compound decomposes. Tetracene is a decomposing compound that undergoes an exothermic reaction from solid straight to liquid. Because of this characteristic, it has been very difficult to measure the exact decomposition point. The Mettler decomposition testing device electronically measures the decomposition point using a computer controlled rate of change. The operator
simply inserts a capillarytube witha sample of Tetracene into a the machine andsets the rate of change and the initial temperature.The computer senses whentheTetracenebeginstodecompose.TheMettlerdevicereplacedthepreviouslyusedVanderkampdevice.TheVanderkampdevicewasfoundtobelessreliableduring thedesignanalysis.It was the Vanderkamp device that originally was used in the finding that thedecompositionpoint had changedtothe 122-125degreerange.TheMettlerdevice confirmedthisrangeoftemperatureforthedecompositionpointforthesamples of Tetracene.We developed a model,an experimental design test on themanufacturingprocess of the Tetracene. Many"applied statistics"options can be used.Due toa combination ofbudgetaryand time constraints,we selecteda Taguchi Designmodel.Design MethodologyTaguchiTheoryThedesignfactors developedbytheJapaneseengineer,GenichiTaguchi, are usedto identify factors that control the“melting point."Taguchifocuses on thequality control of designing engineering products and productionprocesses.Taguchiseesthelossafaultyproductimpartstosociety(aftershipment)intermsofmonetaryloss,dissatisfaction,lossoftime,andhazardstotheenvironment[4].Whichof these, ifany,does the Tetracenepotential lossfunctionappeartocover?Taguchiidentifiesthreephasesofthedesignprocess(systemdesign,parameterdesign, and tolerance design) where statistical methods can be used incontrollingtheengineeringprocess[1].Thefirststep,systems design,isnotconsidered since Tetracene has already been developed and used for decades.There were also no available alternatives for the use of the Tetracene.Parameterdesignisthefocusofthismodelingeffort.Weidentifiedthefactorsthat control the design process by reducing fluctuation in variation,anddetermined thenominal target valueforthe qualitycharacteristicbeingtestedThis will ensure qualityintheengineering and productionprocess.Wefocus onthe nominal target as it will be used in actual production cycles.The last phaseistolerance designwhererange specifications are setbased uponeconomicsand a loss function. Although this is not a one of the primary concerns, we willprovide someinsights intothe specification limits onthedecomposition pointoftheTetracene.Since the loss function is the heart of the Taguchi focus, let's explore it in a littlemore detail.Taguchi uses a unique way to illustrate the loss incurred by a firmand societywhentheproductionprocess isnot operatingefficientlyandthe185
185 simply inserts a capillary tube with a sample of Tetracene into a the machine and sets the rate of change and the initial temperature. The computer senses when the Tetracene begins to decompose. The Mettler device replaced the previously used Vanderkamp device. The Vanderkamp device was found to be less reliable during the design analysis. It was the Vanderkamp device that originally was used in the finding that the decomposition point had changed to the 122-125 degree range. The Mettler device confirmed this range of temperature for the decomposition point for the samples of Tetracene. We developed a model, an experimental design test on the manufacturing process of the Tetracene. Many “applied statistics” options can be used. Due to a combination of budgetary and time constraints, we selected a Taguchi Design model. Design Methodology Taguchi Theory The design factors developed by the Japanese engineer, Genichi Taguchi, are used to identify factors that control the “melting point." Taguchi focuses on the quality control of designing engineering products and production processes. Taguchi sees the loss a faulty product imparts to society (after shipment) in terms of monetary loss, dissatisfaction, loss of time, and hazards to the environment [4]. Which of these, if any, does the Tetracene potential loss function appear to cover? Taguchi identifies three phases of the design process (system design, parameter design, and tolerance design) where statistical methods can be used in controlling the engineering process [1]. The first step, systems design, is not considered since Tetracene has already been developed and used for decades. There were also no available alternatives for the use of the Tetracene. Parameter design is the focus of this modeling effort. We identified the factors that control the design process by reducing fluctuation in variation, and determined the nominal target value for the quality characteristic being tested. This will ensure quality in the engineering and production process. We focus on the nominal target as it will be used in actual production cycles. The last phase is tolerance design where range specifications are set based upon economics and a loss function. Although this is not a one of the primary concerns, we will provide some insights into the specification limits on the decomposition point of the Tetracene. Since the loss function is the heart of the Taguchi focus, let’s explore it in a little more detail. Taguchi uses a unique way to illustrate the loss incurred by a firm and society when the production process is not operating efficiently and the
process does not mitigatethevariation in theproduction process.Mostengineeringprocesseshave specification limits that givea rangeforvaluesfortheprocesscharacteristicforwhichanyvalueinthisrangeisacceptable.AccordingtoRyan [6],lmplicit inthis viewis the ideathat all valueswithin thisrangeareequallygood."Taguchidisagrees withthisassessmentand insteadpromotestheuseofaspecificnominaltargetvalueinthecontrolofqualitycharacteristics,suchasthedecompositionpoint.The nominal target valuehelps focus thequalityof theproduct asopposedtotheproduct justneedingtomeeta specification range.Thus,as the qualitycharacteristicvariesfromthetargetvaluethefirmandsocietyexperiencesaloss.Thesquareofthedifferenceisreferredtoasthemeansquaredeviation(MSD).Byfocusingonthetargetvalueandreducingthevariationinthedesignprocess,the ammunition plant can reduce its loss because the production goaloftheammunitionplantismet.Thisreductionintheexpectedloss(attributedtothereduction of thevariance)is themain reason whyTaguchi stresses that allmanufacturingfirms should strive to reducethevarianceforall process [6].The loss function is an excellent tool to determine the magnitude of the qualitycharacteristicfrom thetarget value and to determinetherange ofthetolerancelevel[5].LosstotheArmyoccurswhentheTetracene cannotbeused,resultinginthe5.56MMand7.62MMsmallcaliberproductionlinesbeingshutdown.Theloss function isgenerallyknown and as a result can bemodeled [6]Applicationtothe DesignModelNowthatTaguchi's approachhasbeen described,wefocus onthe statisticalcomputationsandconceptsusedinparameterdesign.Inordertominimizethesensitivityofaprocess,Taguchihasestablishedorthogonal arraysforhisdesign of experiments.The useof orthogonalarrays is similartothe conceptusedinmoreclassical factorial design ofexperiments.There areseveraldifferences in both designs but bothhavethe goal ofidentifying whichfactorscontroltheexperimentandwhatlevelsshouldbeusedtometthecharacteristicdesired.Taguchi'sdesignincorporatesbothcontrollableanduncontrollablefactors.Theuncontrollable factors are referred to as noise factors. In our design set up, thedifferent employees making theTetracene,the climate,manufacturingimperfectionsproductdeterioration,anddaytodayvariationsareexamplesofnoisefactors.Nospecificnoisefactorsweresetforourdesign,insteadanumbrellaapproachcoveringallthesevariationswasusedbyrepeatingthematrixcreatingBatch1andBatch2samples.Controlfactorsare thevariables that can be controlled inthe productionprocess.Theselectionofthesevariablesiscriticaltoensurethatthecorrectfactorsthatdominatetheprocesscanbeidentified.Thesefactors,asinfactorial186
186 process does not mitigate the variation in the production process. Most engineering processes have specification limits that give a range for values for the process characteristic for which any value in this range is acceptable. According to Ryan [6], “Implicit in this view is the idea that all values within this range are equally good." Taguchi disagrees with this assessment and instead promotes the use of a specific nominal target value in the control of quality characteristics, such as the decomposition point. The nominal target value helps focus the quality of the product as opposed to the product just needing to meet a specification range. Thus, as the quality characteristic varies from the target value the firm and society experiences a loss. The square of the difference is referred to as the mean square deviation (MSD). By focusing on the target value and reducing the variation in the design process, the ammunition plant can reduce its loss because the production goal of the ammunition plant is met. This reduction in the expected loss (attributed to the reduction of the variance) is the main reason why Taguchi stresses that all manufacturing firms should strive to reduce the variance for all process [6]. The loss function is an excellent tool to determine the magnitude of the quality characteristic from the target value and to determine the range of the tolerance level [5]. Loss to the Army occurs when the Tetracene cannot be used, resulting in the 5.56 MM and 7.62 MM small caliber production lines being shut down. The loss function is generally known and as a result can be modeled [6]. Application to the Design Model Now that Taguchi’s approach has been described, we focus on the statistical computations and concepts used in parameter design. In order to minimize the sensitivity of a process, Taguchi has established orthogonal arrays for his design of experiments. The use of orthogonal arrays is similar to the concept used in more classical factorial design of experiments. There are several differences in both designs but both have the goal of identifying which factors control the experiment and what levels should be used to met the characteristic desired. Taguchi’s design incorporates both controllable and uncontrollable factors. The uncontrollable factors are referred to as noise factors. In our design set up, the different employees making the Tetracene, the climate, manufacturing imperfections, product deterioration, and day to day variations are examples of noise factors. No specific noise factors were set for our design, instead an umbrella approach covering all these variations was used by repeating the matrix creating Batch 1 and Batch 2 samples. Control factors are the variables that can be controlled in the production process. The selection of these variables is critical to ensure that the correct factors that dominate the process can be identified. These factors, as in factorial
design, are assigned two levels.The purpose of the experiment is to determinenotonly whichfactors control thedesign (evenwiththenoiselevels exertingtheir influence)but also which of thetwo assigned levels will minimizethevarianceintheproductionprocesswhiledrivingthequalitycharacteristictowardthe nominal target value.Theparameters identified aspotential controllablefactorsintheprocessofmanufacturingTetracenewerelotvariationsinaminoguanidine bicarbonate,sodium nitrite and sulfuric acid; water; agitation of theTetracenemix;sodiumnitriteconcentration;andtemperaturecontrolofthemixTwovariationlevelswerechosenforeachvariable.ThevariableswerefitintooneofTaguchi'sorthogonal arraysthatallowsforsevenparameters orvariables,withatotal of eightexperimentalruns.Werepeatthematrixtoallowfornoisevariation,thusperformingsixteenruns.TESTMATRIXFORTETRACENEINVESTIGATIONLEVEL2CONTROLLABLEFACTORSLEVEL1ALOT2AMINOGUANIDINELOT1BICARBONATEBPRODUCTIONREAGENTSODIUMNITRITECREAGENTSULFURICACIDPRODUCTIONDWATERDEIONIZEDDISTILLEDEWITHAGITATIONOFWITHOUTMIXFSODIUMNITRITEEXCESSSTARVEDCONCENTRATIONGTEMPERATURECONTROLLED(95UNCONTROLLEDCONTROL101DEGC)NOISEFACTORSB1&B2Classical statistical modelsusedinfactorialdesignprocesses use theaverageresponseto calculate the main effects as well as the analysis of variance(ANOVA)toidentify ifthecontrolled variables were significant intheprocess.Incontrast, Taguchiusesa signal tonoise ratio (S/N)to measurethemaineffectsoftheexperimentaldesign.The signal to noise ratioisthechange inthequalitycharacteristic,thedecompositionpointofTetracene.Therefore,theS/Nmeasures the sensitivity of the quality characteristic being investigated in acontrolled manner, to external influencing factors not under control [5].AhighS/Nasthemeasureof performancestatisticisbetterthanalowratiobecauseahighS/N impliesthatthesignal isstrongerthanthenoisefactors.Italsoimpliesthatthequalitycharacteristichasminimumvariance.lnordertounderstand thesetwo concepts, we must look at how the S/Nis calculated(1)S/N= -10LOG10 (MSD)The MsDis thesquare of the responsemeasureminus thetarget value.(2)MSD=(Y-T)InorderfortheMsDtobesmall,theresponsemeasuredfromthedesignofexperimentsmustbeclosetothetargetvalue.Thusindicatingasmallvariancefrom the target value.The small variance also indicates that the noise factors187
187 design, are assigned two levels. The purpose of the experiment is to determine not only which factors control the design (even with the noise levels exerting their influence) but also which of the two assigned levels will minimize the variance in the production process while driving the quality characteristic toward the nominal target value. The parameters identified as potential controllable factors in the process of manufacturing Tetracene were lot variations in amino guanidine bicarbonate, sodium nitrite and sulfuric acid; water; agitation of the Tetracene mix; sodium nitrite concentration; and temperature control of the mix. Two variation levels were chosen for each variable. The variables were fit into one of Taguchi’s orthogonal arrays that allows for seven parameters or variables, with a total of eight experimental runs. We repeat the matrix to allow for noise variation, thus performing sixteen runs. TEST MATRIX FOR TETRACENE INVESTIGATION CONTROLLABLE FACTORS LEVEL 1 LEVEL 2 A AMINO GUANIDINE BICARBONATE LOT 1 LOT 2 B SODIUM NITRITE PRODUCTION REAGENT C SULFURIC ACID REAGENT PRODUCTION D WATER DEIONIZED DISTILLED E AGITATION OF MIX WITH WITHOUT F SODIUM NITRITE CONCENTRATION EXCESS STARVED G TEMPERATURE CONTROL CONTROLLED (95- 101 DEG C) UNCONTROLLED NOISE FACTORS B1 & B2 Classical statistical models used in factorial design processes use the average response to calculate the main effects as well as the analysis of variance (ANOVA) to identify if the controlled variables were significant in the process. In contrast, Taguchi uses a signal to noise ratio (S/N) to measure the main effects of the experimental design. The signal to noise ratio is the change in the quality characteristic, the decomposition point of Tetracene. Therefore, the S/N measures the sensitivity of the quality characteristic being investigated in a controlled manner, to external influencing factors not under control [5]. A high S/N as the measure of performance statistic is better than a low ratio because a high S/N implies that the signal is stronger than the noise factors. It also implies that the quality characteristic has minimum variance. In order to understand these two concepts, we must look at how the S/N is calculated. S/N = - 10 LOG10 (MSD) (1) The MSD is the square of the response measure minus the target value. MSD = (Y-T)2 (2) In order for the MSD to be small, the response measured from the design of experiments must be close to the target value. Thus indicating a small variance from the target value. The small variance also indicates that the noise factors