optimizationofinjection-moldingprocessparametersfor … · 2019. 7. 4. · (pmma, ht55x produced by...

Research ArticleOptimization of Injection-Molding Process Parameters forWeight Control Converting Optimization Problem toClassification Problem

Peng Zhao 123 Zhengyang Dong12 Jianfeng Zhang12 Yi Zhang12 Mingyi Cao12

Zhou Zhu4 Hongwei Zhou5 and Jianzhong Fu12

113e State Key Laboratory of Fluid Power and Mechatronic Systems College of Mechanical Engineering Zhejiang UniversityHangzhou 310027 China2Key Laboratory of 3D Printing Process and Equipment of Zhejiang Province College of Mechanical EngineeringZhejiang University Hangzhou 310027 China3Jiangsu Jianghuai Magnetic Industry Co Ltd Xuyi 211700 China4College of Mechanical Engineering Zhejiang University of Technology Hangzhou 310014 China5Tederic Machinery Co Ltd Hangzhou 311224 China

Correspondence should be addressed to Peng Zhao pengzhaozjueducn

Received 4 July 2019 Revised 12 August 2019 Accepted 16 September 2019 Published 26 March 2020

Academic Editor Leandro Gurgel

Copyright copy 2020 Peng Zhao et al +is is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Product weight is one of the most important properties for an injection-molded part+e determination of process parameters forobtaining an accurate weight is therefore essential +is study proposed a new optimization strategy for the injection-moldingprocess in which the parameter optimization problem is converted to a weight classification problem Injection-molded parts areproduced under varying parameters and labeled as positive or negative compared with the standard weight and the weight errorof each sample is calculated A support vector classifier (SVC) method is applied to construct a classification hyperplane in whichthe weight error is supposed to be zero A particle swarm optimization (PSO) algorithm contributes to the tuning of thehyperparameters of the SVC model in order to minimize the error between the SVC prediction results and the experimentalresults+e proposed method is verified to be highly accurate and its average weight error is 00212+is method only requires asmall amount of experiment samples and thus can reduce cost and time +is method has the potential to be widely promoted inthe optimization of injection-molding process parameters

1 Introduction

Injection molding is regarded as the most important methodfor mass-producing plastic parts [1ndash4] and it is complicatedbecause of steep thermal gradients and complex flow ge-ometries many studies have been conducted in this field[5ndash10]+e product weight of a plastic part is considered oneof the most significant properties especially for plastic partswith high precision requirements such as plastic lenses +eproduct weight can reflect other quality properties such asimpact strength [11] Weight control is also of greatcommercial interest for manufacturers as it reduces the cost

of materials in mass production [12] +erefore the weightcontrol of products is a critical issue in injection molding

For a specific material and product process parametersare the most important factors affecting product weight andmany studies have been conducted to explore the rela-tionship between process parameters and product weightsHassan [13] determined that product weight has a positivecorrelation with packing pressure Lopez et al [14] appliedthe design of experiments (DOE) method to explore theinfluence of parameters on products with different geom-etries Yang and Gao [11] found that the settings for packingpressure barrel temperature and mold temperature have

HindawiAdvances in Polymer TechnologyVolume 2020 Article ID 7654249 9 pageshttpsdoiorg10115520207654249

the most significant effects on the product weight Hence itis of vital importance to select optimal process parametersbefore the part is molded However the procedure forselecting optimal process parameters remains challengingMost parameters are coupled with each other and it istherefore difficult to construct an accurate mathematicalmodel +e traditional parameter optimization method is aninefficient trial-and-error process based on personal expe-rience and it is not suitable for complex plastic parts [15]

Because of these challenges researchers have conductedmany studies to explore the approaches to take in order tooptimize injection-molding parameters +e design of ex-periments (DOE) such as the Taguchi method is applied toimprove the quality of manufactured goods by analyzing thesignal-to-noise (SN) ratio and the analysis of variance(ANOVA) [16ndash21]+e Taguchi method is unsuitable for theproblem in this study because process parameters arecontinuous +us the Taguchi method is unable to assisttechnologists in obtaining optimal process parameters [22]+e DOE approach also requires that technologists un-derstand both statistics and the injection-molding process inexperiment planning [23] Because of the shortcomings ofthe DOE method researchers have proposed new optimi-zation methods by applying surrogate models such as theartificial neural network (ANN) [24] and iterative methodssuch as the genetic algorithm (GA) [25 26] Xu et al [27]developed a backpropagation neural network model com-bined with particle swarm optimization (PSO) to map thecomplex nonlinear relationship between process parametersand the mechanical performance of the product Chen et al[28] applied a backpropagation neural network geneticalgorithms and engineering optimization concepts toachieve competitive advantages in both quality and costHowever defining the structure of a neural network is atime-consuming procedure due to the lack of uniformstructure Technologists need to determine the number oflayers and nodes in the network as well as the connectionrelations between nodes In addition to guarantee accuracyANN usually requires a large number of experiment sam-ples which is difficult to acquire in practice

In this study the optimization objective is to generateprocess parameters for the molding of optimal products of astandard weight Because the weight of the molded producthas an either positive or negative error the optimizationproblem can be converted into a classification problem +esupport vector machine (SVM) method is now widely ap-plied in classification problems such as separating defectiveand nondefective products Yu [29] applied the SVM andDOE methods for predicting process windows to ensurerobust high-quality injectionmoldings Gao [30] proposed amethod using SVM to optimize process parameters thatcould provide more stable product quality than traditionalmethods Shin et al [31] pointed out that SVM generally hasbetter accuracy than ANN as the data size decreases Due tothe advantages of SVM this study proposes a classifiermodel using the support vector machine method called thesupport vector classifier (SVC) to construct a hyperplanewhich separates products that are heavier or lighter than thestandard weight Optimal process parameters are presumed

to be located on the hyperplane constructed by the SVCmodel Having established an SVC model a particle swarmoptimization (PSO) algorithm is applied to improve the SVCmodel by tuning hyperparameters in order to minimize theerror between the experimental result and the SVC predictionvalue +e outline of this paper is as follows Section 2 in-troduces the implementation procedure of the proposedoptimizationmethod and its key algorithms In Sections 3 and4 a case of plastic lenses is carried out to verify the presentedalgorithm +e conclusions are provided in the last section

2 Method Implementation

+e implementation procedure for the proposed optimi-zation method is described in Section 21 Section 22 il-lustrates the SVC algorithm in detail and the method fortuning the hyperparameters of SVC using PSO is introducedin Section 23

21 Implementation Procedure +e standard weight of aplastic product is determined by technologists prior tomolding +e product weight can be regarded as a functionof the process parameters throughout the injection-moldingprocess Each experiment sample under different processparameters could be heavier or lighter than the standardweight and can therefore be labeled as a positive sample(heavier) or a negative sample (lighter) +e molding pa-rameters for the desired product with a standard weightshould be located between the molding parameters for thepositive and negative samples Hence this study intends tosearch for the parameter classification boundary between thepositive and negative classes as shown in Figure 1+e largerthe weight error of a sample is the further away the point isfrom the boundary and the points on the boundary cor-respond to a zero-weight error

In this study an SVC model is employed to classifyproducts by their weights Products under different processparameters are injection-molded and weighed +eseproducts are labeled as either positive or negative samplesand they are employed as training data for the SVC Asillustrated in Figure 1 the SVC constructs a maximum-margin hyperplane which is also known as the classificationboundary to separate the positive and negative samples +edistance between the samples and the hyperplane corre-sponds to weight errors Optimal parameters should belocated on the hyperplane of the trained SVC model Be-cause of the linearly nonseparable problem further modi-fications of the SVC model are required including theselection and calculation of the kernel function [32] as wellas the determination of the slack variable [33] Mathematicalprinciples are illustrated in Section 22 in detail +e kernelparameter of the kernel function and the penalty parameterof the slack variable have a great influence on the perfor-mance of the SVC model +e determination of thesehyperparameters for the SVC is therefore significant In thisstudy a particle swarm optimization (PSO) method isadopted +e implementation procedure for the proposedmethod can be summarized as follows

2 Advances in Polymer Technology

Step 1 initialize parameter range Choose moldingparameters and then set their initial range +e initialparameter range should cover all the feasibleparametersStep 2 perform experiments Carry out experimentsunder each process parameter +en evaluate experi-mental results by comparing the product weight withthe standard weightStep 3 train the SVC model Label these samples andprepare the data sets for the SVC Select proper kernelfunction and train the SVC with sample dataStep 4 tune SVC hyperparameters Apply PSO todetermine hyperparameters A hyperplane that sepa-rates underweight and overweight samples can beobtained +e optimal process parameters are locatedon that hyperplaneStep 5 perform verification experiments

22 Support Vector Classifier Algorithm +e support vectorclassifier is a machine learning algorithm based on statisticallearning theory It minimizes the structural risk and im-proves the generalization ability of the learning machine+e target of a classification is to estimate a function f usingthe training data (x1 y1) middot middot middot (xn yn) wherex (x1 xn) isin RN is the matrix of the process parame-ters in this study and y isin +1 minus 1 is the label indicatingwhether a product is heavier or lighter than the standardweight

Considering a linear case as shown in Figure 1 theclassification function can be written as

f(x) wTx + b (1)

where w is a normal vector and b is the intercept Whenf(x) 0 x is a point located on the hyperplane For allpoints that satisfy f(x)lt 0 the corresponding label yi is minus 1and for all points that satisfy f(x) gt 0 the correspondinglabel yi is 1 It can also be written as

yi times wTx + b1113872 1113873ge 1 (2)

+e margin between two classes is

margin x+rarr

minus xminusrarr

( 1113857 middotwrarr

||wrarr

||1 + b minus (minus 1 + b)

||wrarr

|| (3)

Moreover the target is to find the hyperplane that makesthe margin the largest For mathematic conveniencemaximize 2w

rarr is equal to minimize (12)w2 +us the

optimization problem becomes

min12

w2

st yi middot wTx + b( 1113857ge 1

⎧⎪⎪⎪⎨

⎪⎪⎪⎩

(4)

Because of the linear inseparability of the data set akernel function must be adopted to map the data (inputspace) to a high-dimensional space (feature space) where alinear separating hyperplane can be constructed One kernelcommonly used in training nonlinear SVCs is the Gaussiankernel which maps data to an infinite-dimensional space+e Gaussian kernel on two samples x and y is defined as

K(x y) exp minus||x minus y||2

2σ21113888 1113889 exp minus c||x minus y||

21113872 1113873 (5)

where c 12σ2 represents the width and height of theGaussian function

Although the probability of linear separation is increasedby introducing the kernel function it is still difficult to dealwith the noise in a data set Noisy data also known asldquooutliersrdquo have a great influence on the SVC model espe-cially when outliers become support vectors +e slackvariable ξi is therefore introduced After considering out-liers the constraint condition in equation (4) is

yi wTxi + b1113872 1113873ge 1 minus ξi i 1 2 n (6)

+e slack variable means that the accurate classificationof the outliers is abandoned which is an additional loss tothe classifier and should be added to the objective function+us the optimization problem becomes

min12

w2

+ C 1113944n

i1ξi

st yi wTxi + b( 1113857ge 1 minus ξi

ξi ge 0

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(7)

where C is the penalty parameter which refers to thesensitivity of the SVC model to outliers +e weight errorbetween each sample and the standard weight can be cal-culated which means each sample should have a differentpenalty parameter +e further away the sample is from thestandard the more emphasis the SVC should put on thesample which means the penalty parameter C for thissample should be larger For this reason we shouldmodify C

for each sample according to its own weight error +e

Samples heavier than standardpositive class

Classification boundarypredicted parameters

corresponding to standard weight

Samples lighter than standardnegative class

Figure 1 Schematic diagram of converting the parameter opti-mization problem to a classification problem

Advances in Polymer Technology 3

determinant of penalty parameter C is as follows firstexpand the value of C to an n times 1 matrix and then calculatethe updated penalty parameter matrix Csample as

Csample C⊙We (8)

whereWe is an n times 1 matrix of the weight error of n samplesand ldquo⊙rdquo is the denotation of componentwise multiplicationFrom equation (7) it can be clearly seen that it is a convexoptimization problem with linear constraint +e final op-timization goal is

minα

1113944

n

i1αi minus

12

1113944

n

i11113944

n

j1αiαjyiyjK xi xj1113872 1113873

st 0le αi leC

1113944

n

i1αiyi 0

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(9)

where αi is a nonnegative Lagrange multiplier and K(xi xj)

is the kernel function as shown in equation (5)It should be noted that the performance of an SVC

model depends largely on the parameters c andC Parameterc determines the sphere of influence of each support vectorA large value of c is conducive to the situation in which thesupport vectors affect only the nearby area which can lead tohigh variance of the model and overfitting Parameter C

controls the tradeoff between achieving a low training errorand the ability to generalize the model to unseen dataBecause of the absence of prior knowledge of the choice ofparameters some types of model selection (parametricsearch) must be conducted Several approaches have beenproposed such as K-folder cross validation [34] and gridsearch [35] K-folder cross validation is often combined withgrid search as a method of parameter evaluation Via thismethod hyperparameters can be modified by improving theclassification accuracy of the model However this criteriononly applies to a model with a large number of samplesBecause of the small data set in this study a new criterion tomodify SVC model must be proposed

23 Determination of SVC Hyperparameters Using ParticleSwarm Optimization Particle swarm optimization is analgorithm that simulates the foraging behavior of a flock ofbirds +e basic idea is to find the optimal solution viacollaboration and information sharing among individuals inthe group A group of particles (candidate solutions) israndomly initialized in the solution space Each particle isevaluated by calculating its fitness value+e optimal solutionis then found by iteration In each iteration the particleupdates its own velocity and position by tracking ldquoPbestrdquo (thebest known position of the current individual) and ldquoGbestrdquo(the best known position of the current group) +e updatedformula for particle i(i 1 2 n) can be written as

vi ω times vi + c1 times rand times Pbesti minus xi( 1113857

+ c2 times rand times Gbesti minus xi( 1113857

xi xi + vi

⎧⎪⎪⎨

⎪⎪⎩(10)

where vi and xi are the velocity and position of the particlerespectively rand is a random value between (minus 1 1) c1 andc2 are learning factors and under normal conditionsc1 c2 2 and ω is known as the inertia weight factor Inorder to achieve better convergence ω is set to linearlydecrease with the iteration

ω ωmin + ωmax minus ωmin( 1113857imax minus i

imax (11)

where ω is the inertia weight for each iteration ωmin andωmax are the minimum and maximum inertia weights re-spectively which are set to 02 and 06 in this study and i

refers to the current iteration step By applying a lineardecrease particles can quickly converge to the approximateoptimal solution with a large step size at the beginning andthen converge to the exact solution with a small step size

In this study PSO is applied to improve the accuracy ofthe SVCmodel by determining its hyperparameters +e keyto applying PSO in this optimization problem is to constructa proper fitness function In order to acquire the mostprecise result the prediction value of the SVC should be thesame as the actual value from experiments By tuning thehyperparameters of the SVC model the position of thehyperplane can be modified +us the distance between thehyperplane and the samples is as close as possible to theactual weight error +e calculation of fitness of particles is

fitness p minus e2 (12)

where p is the normalized vector of the distance between thesample and the hyperplane and e is the normalized vector ofthe product weight error p and e are calculated by

p p minus min(p)

max(p) minus min(p)

e e minus min(e)

max(e) minus min(e)

(13)

+rough PSO all particles converge to the positionwhere the fitness value is reduced to the minimum +eoptimal position corresponds to the modified combinationof hyperparameters of the SVC With this method theparameters c and C can be calculated

3 Experimental Design

To demonstrate the accuracy of the proposed method anexperiment using a plastic lenses is presented in this sectionA high-precision electrical injection-molding machineZhafir VE400 (Zhafir Plastics Machinery GmbH China) isemployed in this study and is depicted in Figure 2(a) +einjection mold and molded plastic lenses are shown inFigure 2(b) and Figure 2(c) respectively A single lens has adiameter of 30mm and a maximum thickness of 2mm +eresin used in this study is poly(methyl methacrylate)(PMMA HT55X produced by Sumipex Japan [36]) Toweigh the product an electronic balance FA2004 (ShanghaiSunny Hengping Scientific Instrument Co Ltd China) isemployed the measurement accuracy of which is 00001 g(see Figure 2)


Many process parameters have large effect on productweight including packing pressure packing time injectionpressure injection time and injection temperature +eeffect of these parameters depends on the type of the resin[13] and the structure of the product [14] According toYang Hassan and Lopez [11ndash14] the injection temperatureand packing pressure are the two parameters which havethe most significant influence on product weight Onpurpose of demonstrating the optimization results moreclearly this study selects only these two parameters asoptimization variables +e initial range of injection tem-perature is divided into 4 levels namely 235degC 245degC255degC and 265degC according to the recommended tem-peratures provided by the material supplier A relativelywide range of packing pressure from 70MPa to 120MPa inincrements of 10MPa is set as the initial range Subse-quently a total number of 24 injection-molding analyseswith different process parameters are carried out Afterobtaining the experiment samples all runners are removedbecause the weight of the additional plastic in runners is notconcerned in this study Only the four lenses of one sampleare weighted +e total standard weight of four lenses is816 g +e experiment results are shown in the followingsection

4 Results and Verification

41 Experimental Results +e product weight under thecondition of each process parameter is presented in Table 1and Figure 3 A positive weight error value indicates that theproduct weight is larger than the standard value and viceversa

From the experimental results we can observe that theproduct weight has a positive correlation with both thepacking pressure and injection temperature +is phe-nomenon agrees well with the research of Hamdy [13] andYang and Gao [11] Due to the high viscosity PMMA has apoor fluidity and the melt viscosity is sensitive to injectiontemperature +e fluidity of the PMMA melt improves withan increase in the injection temperature which leads tomoreplastic melt flowing into the mold In addition the meltsolidifies quickly due to the small thickness of the productWhen the injection temperature is higher the time to reachsolidification increases +erefore more melt can passthrough the gate during a longer time With an increase in

the packing pressure more materials can be packed into thecavity in the packing phase

As the SVC is a supervised machine learning model thedata set must be labeled In this study we regard all thenegative values with a -1 label and the positive values with a+1 label +e two key parameters (C and c) of the SVC areoptimized through PSO and the convergence procedure isshown in Figure 4 From the result of PSO the performanceof the SVC model is determined to be the best with C

24513 and c 574 +e hyperplane that separates theoverweight and underweight samples is exhibited in Fig-ure 5 each dot indicates an experimental sample and thesize of the dot corresponds to the magnitude of the weighterror +e product weight under process parameters that arelocated on the hyperplane should be equal to the standardweight of 816 g

(a) (b) (c)

Figure 2 Experimental setup (a) injection-molding machine (b) injection mold (c) molded plastic lenses

Table 1 Weight under each parameter set

NoProcess parameters

Productweight (g)

Weighterror (g)Packing

pressure (MPa)Injection

temperature (degC)1 70 235 80968 minus 006322 80 235 81324 minus 002763 90 235 81040 minus 005604 100 235 80920 minus 006805 110 235 81456 minus 001446 120 235 81664 000647 70 245 80944 minus 006568 80 245 80896 minus 007049 90 245 81100 minus 0050010 100 245 81248 minus 0035211 110 245 81716 0011612 120 245 82104 0050413 70 255 81316 minus 0028414 80 255 81624 0002415 90 255 82028 0042816 100 255 82404 0080417 110 255 82836 0123618 120 255 83288 0168819 70 265 84116 0251620 80 265 85212 0361221 90 265 87520 0592022 100 265 87608 0600823 110 265 86756 0515624 120 265 88624 07024


120110100

Packing pressure (MPa)90 80 70 235245255

265Injection temperature (degC)

714735756777798819840861882

Wei

ght (

g)

Figure 3 Product weight under varying process parameters

times10ndash4

1000 1500 2000 2500 3000 3500 4000500Penalty parameter C

1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(a)

times10ndash4


1

2

3

4

5

6

7

8

9Ke

rnel

par

amet

er γ

(b)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(c)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ C = 24513

γ = 574

(d)

Figure 4 PSO convergence procedure (a) random initialization (b) 8th iteration step (c) 54th iteration step (d) 261st iteration stepconverge at C 24513 and c 574


42 Verification In order to evaluate the hyperplane ob-tained from the proposed method a set of verification ex-periments are carried out +e verification experimentscheme is illustrated in Figure 6+e points a0 a1 a2 a8and b1 and b2 represent different injection-molding processparameters

To verify that a single point located on the hyperplane isthe best parameter combination among the other pointsaround it the center point (a0) with a packing pressure of95MPa and an injection temperature of 248degC is selected asthe verification point A series of experiments (a1 a2 a3 anda4) is first conducted with a fixed packing pressure of 95MPaand varying injection temperatures from 242degC to 254degC inincrements of 3degC +en several experiments (a5 a6 a7 anda8) were conducted with a fixed injection temperature of248degC and varying packing pressures from 89MPa to101MPa in increments of 3MPa In total 9 sets of exper-iments are conducted For each parameter combination 30products are molded and weighed +e results are providedin Figure 7 and the statistics are listed in Table 2 and 3Although there are fluctuations during the injection-molding process the product weight under the optimizedparameters is the closest to the standard weight +e average

weight error can reach 00212 which is smaller than anyother parameter combinations around it

In the case in which a single point on the hyperplane isknown as an optimal parameter set the second step is toverify that other points on the hyperplane are also feasibleExperiments under two other parameter combinations onthe hyperplane (b1 85MPa2525degC and b2 105MPa2448degC) are carried out +eir weights are 81645 g and81634 g respectively and the results are presented in Fig-ure 8 We can presume that the product weight under otherparameters on the hyperplane is also close to the standardweight Other feasible process parameters offer morechoices so the technologists can pick up the parameter thatfits the current product the best according to the productstructure and the type of the resin

230

235

240

245

250

255

260

265

270

Inje

ctio

n te

mpe

ratu

re (deg

C)

11090 10080 12070Packing pressure (MPa)

Figure 5 Classification hyperplane (green curve) red dots areoverweight samples blue dots are underweight samples

a0

a1

a2

a3

a4

a5a6 a7

a8

b1

b2

270

265

260

255

250

245

240

235

23070 80 90

Packing pressure (MPa)

Inje

ctio

n te

mpe

ratu

re (deg

C)

100 110 120

Figure 6 Verification experiments scheme a0 b1 and b2 areprocess parameters on the hyperplane while a1 a2 a8 areprocess parameters not on the hyperplane

0 5 10 15 20 25 30

808

812

816

820

824

828

a1

a2

a0a3

251degC 254degCStandard

Prod

uct w

eigh

t (g)

Test number

242degC245degC248degC

a4

(a)

0 5 10 15 20 25 30

808

812

816

820

824

828

a8

a7

a0

a6

a5

89MPa92MPa95MPa

Prod

uct w

eigh

t (g)

Test number

98MPa101MPaStandard

(b)

Figure 7 Verification results product weight under (a) varyinginjection temperature (b) varying packing pressure


5 Conclusions

+is study proposed a new method for optimizing processparameters to mold products with a standard weight +eparameter optimization problem is converted into aweight classification problem according to whether thesample is heavier or lighter than the standard weight +esupport vector classifier and particle swarm optimizationalgorithm are adopted to construct the classification hy-perplane which separates samples A new criterion for theclassification model is introduced in order to improveaccuracy +e product weight under parameter sets lo-cated on the classification hyperplane should be the sameas the standard weight Based on the results obtained inthis study the following conclusions can be drawn (1)+e

product weight under optimized parameters is ratherclose to the standard weight Experimental results indicatethat the weight error can reach 00212 +e hyperplaneobtained from the SVC has a high level of correspondencewith the verification result (2) +e idea of converting anoptimization problem to a classification problem is provenuseful for process parameter optimization (3) +e ac-curacy of the classification model can be improved byapplying the PSO method and the criterion proposed inthis study (4) In contrast to the shortcomings in thetraditional Taguchi method this approach for the opti-mization of process parameters can deal with the situationin which the process parameters are continuous andnonlinear In general the proposed method has the ad-vantages of a small data set requirement high accuracyand the ability to deal with nonlinear problems

+e ongoing studies will serve to solve the followingproblems (1) apply this method to multiple parametersoptimization problems (2) drive the mathematical model toa tool for design algorithms to determine the processwindow (3) promote the proposed method to other pa-rameter optimization problems for other polymer pro-cessing techniques such as extrusion molding and blowmolding

Data Availability

+e data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+e authors would like to acknowledge the financial supportof the Zhejiang Provincial Natural Science Foundation ofChina (No LZ18E050002) the Key Research and Devel-opment Plan of Zhejiang Province (No 2020C01113) andthe National Natural Science Foundation Council of China(No 51875519 and No 51635006)

References

[1] H Zhou Computer Modeling for Injection Molding JohnWiley amp Sons Hoboken NJ USA 2013

[2] B Ozcelik and T Erzurumlu ldquoComparison of the warpageoptimization in the plastic injection molding using ANOVAneural network model and genetic algorithmrdquo Journal ofMaterials Processing Technology vol 171 no 3 pp 437ndash4452006

[3] J Yao ldquoA review of current developments in process andquality control for injection moldingrdquo Advances in PolymerTechnology vol 24 no 3 pp 165ndash182 2010

[4] C Fernandes A J Pontes J C Viana and A Gaspar-CunhaldquoModeling and optimization of the injection-molding processa reviewrdquo Advances in Polymer Technology vol 37 no 2pp 429ndash449 2018

[5] Y Zhao P Zhao J Zhang J Huang N Xia and J Fu ldquoOn-line measurement of clamping force for injection molding

Table 2 Product weight under varying injection temperature

Injection temperature (degC) Averageweight (g) Error () Standard

deviation (g)242 81050 minus 06746 000401245 81194 minus 04973 000445248 81617 00212 000453251 81823 02736 000564254 82147 06705 000705

Table 3 Product weight under varying packing pressure

Packing pressure (MPa) Averageweight (g) Error () Standard


828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters

85MPa2525degC95MPa248degC105MPa2448degC

b2

Wei

ght (

g)

Figure 8 Product weight under three parameters on thehyperplane


machine using ultrasonic technologyrdquo Ultrasonics vol 91pp 170ndash179 2019

[6] G Shan ldquoFlow properties of polymer melt in longitudinalultrasonic-assisted microinjection moldingrdquo Polymer Engi-neering amp Science vol 57 no 8 pp 797ndash805 2017

[7] J Zhang P Zhao Y Zhao J Huang N Xia and J Fu ldquoOn-line measurement of cavity pressure during injection moldingvia ultrasonic investigation of tie barrdquo Sensors and ActuatorsA Physical vol 285 pp 118ndash126 2019

[8] N Xia P Zhao J Xie C Zhang J Fu and L-S TurngldquoDefect diagnosis for polymeric samples via magnetic levi-tationrdquo NDT amp E International vol 100 pp 175ndash182 2018

[9] N Xia P Zhao T Kuang Y Zhao J Zhang and J FuldquoNondestructive measurement of layer thickness in water-assisted coinjection-molded product by ultrasonic technol-ogyrdquo Journal of Applied Polymer Science vol 135 no 33p 46540 2018

[10] C Fernandes A J Pontes J C Viana J M Nobrega andA Gaspar-Cunha ldquoModeling of plasticating injectionmolding - experimental assessmentrdquo International PolymerProcessing vol 29 no 5 pp 558ndash569 2014

[11] Y Yang and F Gao ldquoInjection molding product weightonline prediction and control based on a nonlinear principalcomponent regression modelrdquo Polymer Engineering amp Sci-ence vol 46 no 4 pp 540ndash548 2006

[12] M R Kamal A E Varela and W I Patterson ldquoControl ofpart weight in injection molding of amorphous thermo-plasticsrdquo Polymer Engineering amp Science vol 39 no 5pp 940ndash952 1999

[13] H Hassan ldquoAn experimental work on the effect of injectionmolding parameters on the cavity pressure and productweightrdquo 13e International Journal of AdvancedManufacturing Technology vol 67 no 1-4 pp 675ndash686 2013

[14] A Lopez J Aisa A Martinez and D Mercado ldquoInjectionmoulding parameters influence on weight quality of complexparts by means of DOE application case studyrdquo Measure-ment vol 90 pp 349ndash356 2016

[15] H GAO Y Zhang X Zhou and D Li ldquoIntelligent methodsfor the process parameter determination of plastic injectionmoldingrdquo Frontiers of Mechanical Engineering vol 13 no 1pp 85ndash95 2018

[16] M D Azaman S M Sapuan S Sulaiman E S Zainudin andA Khalina ldquoOptimization and numerical simulation analysisfor molded thin-walled parts fabricated using wood-filledpolypropylene composites via plastic injection moldingrdquoPolymer Engineering amp Science vol 55 no 5 pp 1082ndash10952015

[17] M Cao F Gu C Rao J Fu and P Zhao ldquoImproving theelectrospinning process of fabricating nanofibrous mem-branes to filter PM25rdquo Science of 13e Total Environmentvol 666 pp 1011ndash1021 2019

[18] P Zhao M Cao H Gu et al ldquoResearch on the electrospunfoaming process to fabricate three-dimensional tissue engi-neering scaffoldsrdquo Journal of Applied Polymer Sciencevol 135 no 46 p 46898 2018

[19] H Oktem T Erzurumlu and I Uzman ldquoApplication ofTaguchi optimization technique in determining plastic in-jection molding process parameters for a thin-shell partrdquoMaterials amp Design vol 28 no 4 pp 1271ndash1278 2007

[20] X Sanchez-Sanchez A Elias-Zuntildeiga and M Hernandez-Avila ldquoProcessing of ultra-high molecular weight polyeth-ylenegraphite composites by ultrasonic injection mouldingTaguchi optimizationrdquo Ultrasonics Sonochemistry vol 44pp 350ndash358 2018

[21] J Zhao and G Cheng ldquoAn innovative surrogate-basedsearching method for reducing warpage and cycle time ininjection moldingrdquo Advances in Polymer Technology vol 35no 3 pp 288ndash297 2016

[22] C-T Su and H-H Chang ldquoOptimization of parameter de-sign an intelligent approach using neural network andsimulated annealingrdquo International Journal of Systems Sci-ence vol 31 no 12 pp 1543ndash1549 2000

[23] S L Mok C K Kwong andW S Lau ldquoReview of research inthe determination of process parameters for plastic injectionmoldingrdquo Advances in Polymer Technology vol 18 no 3pp 225ndash236 1999

[24] W-C Chen and D Kurniawan ldquoProcess parameters opti-mization for multiple quality characteristics in plastic injec-tion molding using Taguchi method BPNN GA and hybridPSO-GArdquo International Journal of Precision Engineering andManufacturing vol 15 no 8 pp 1583ndash1593 2014

[25] C Fernandes A J Pontes J C Viana and A Gaspar-CunhaldquoUsing multiobjective evolutionary algorithms in the opti-mization of operating conditions of polymer injectionmoldingrdquo Polymer Engineering amp Science vol 50 no 8pp 1667ndash1678 2010

[26] C Fernandes A J Pontes J C Viana and A Gaspar-CunhaldquoUsing multi-objective evolutionary algorithms for optimi-zation of the cooling system in polymer injection moldingrdquoInternational Polymer Processing vol 27 no 2 pp 213ndash2232012

[27] Y Xu Q Zhang W Zhang and P Zhang ldquoOptimization ofinjection molding process parameters to improve the me-chanical performance of polymer product against impactrdquo13e International Journal of Advanced Manufacturing Tech-nology vol 76 no 9ndash12 pp 2199ndash2208 2015

[28] W-C Chen G-L Fu P-H Tai and W-J Deng ldquoProcessparameter optimization for MIMO plastic injection moldingvia soft computingrdquo Expert Systems with Applications vol 36no 2 Part 1 pp 1114ndash1122 2009

[29] S Yu ldquoOffline prediction of process windows for robustinjection moldingrdquo Journal of Applied Polymer Sciencevol 131 no 18 2014

[30] H Gao ldquoProcess parameters optimization using a novelclassification model for plastic injection moldingrdquo 13e In-ternational Journal of Advanced Manufacturing Technologyvol 94 no 1ndash4 pp 357ndash370 2018

[31] K-S Shin T S Lee andH-j Kim ldquoAn application of supportvector machines in bankruptcy prediction modelrdquo ExpertSystems with Applications vol 28 no 1 pp 127ndash135 2005

[32] B E Boser I M Guyon and V N Vapnik ldquoA trainingalgorithm for optimal margin classifiersrdquo in Proceedings of theFifth Annual Workshop on Computational Learning 13eoryJuly 1992

[33] C Cortes and V Vapnik ldquoSupport-vector networksrdquo Ma-chine Learning vol 20 no 3 pp 273ndash297 1995

[34] K Duan S S Keerthi and A N Poo ldquoEvaluation of simpleperformance measures for tuning SVM hyperparametersrdquoNeurocomputing vol 51 pp 41ndash59 2003

[35] F Friedrichs and C Igel ldquoEvolutionary tuning of multipleSVM parametersrdquo Neurocomputing vol 64 pp 107ndash1172005

[36] Ltd SCAP PMMAMMA httpswwwsumitomo-chemcomsgpetrochemicals-plasticspmma-mma


the most significant effects on the product weight Hence itis of vital importance to select optimal process parametersbefore the part is molded However the procedure forselecting optimal process parameters remains challengingMost parameters are coupled with each other and it istherefore difficult to construct an accurate mathematicalmodel +e traditional parameter optimization method is aninefficient trial-and-error process based on personal expe-rience and it is not suitable for complex plastic parts [15]

Because of these challenges researchers have conductedmany studies to explore the approaches to take in order tooptimize injection-molding parameters +e design of ex-periments (DOE) such as the Taguchi method is applied toimprove the quality of manufactured goods by analyzing thesignal-to-noise (SN) ratio and the analysis of variance(ANOVA) [16ndash21]+e Taguchi method is unsuitable for theproblem in this study because process parameters arecontinuous +us the Taguchi method is unable to assisttechnologists in obtaining optimal process parameters [22]+e DOE approach also requires that technologists un-derstand both statistics and the injection-molding process inexperiment planning [23] Because of the shortcomings ofthe DOE method researchers have proposed new optimi-zation methods by applying surrogate models such as theartificial neural network (ANN) [24] and iterative methodssuch as the genetic algorithm (GA) [25 26] Xu et al [27]developed a backpropagation neural network model com-bined with particle swarm optimization (PSO) to map thecomplex nonlinear relationship between process parametersand the mechanical performance of the product Chen et al[28] applied a backpropagation neural network geneticalgorithms and engineering optimization concepts toachieve competitive advantages in both quality and costHowever defining the structure of a neural network is atime-consuming procedure due to the lack of uniformstructure Technologists need to determine the number oflayers and nodes in the network as well as the connectionrelations between nodes In addition to guarantee accuracyANN usually requires a large number of experiment sam-ples which is difficult to acquire in practice

In this study the optimization objective is to generateprocess parameters for the molding of optimal products of astandard weight Because the weight of the molded producthas an either positive or negative error the optimizationproblem can be converted into a classification problem +esupport vector machine (SVM) method is now widely ap-plied in classification problems such as separating defectiveand nondefective products Yu [29] applied the SVM andDOE methods for predicting process windows to ensurerobust high-quality injectionmoldings Gao [30] proposed amethod using SVM to optimize process parameters thatcould provide more stable product quality than traditionalmethods Shin et al [31] pointed out that SVM generally hasbetter accuracy than ANN as the data size decreases Due tothe advantages of SVM this study proposes a classifiermodel using the support vector machine method called thesupport vector classifier (SVC) to construct a hyperplanewhich separates products that are heavier or lighter than thestandard weight Optimal process parameters are presumed

to be located on the hyperplane constructed by the SVCmodel Having established an SVC model a particle swarmoptimization (PSO) algorithm is applied to improve the SVCmodel by tuning hyperparameters in order to minimize theerror between the experimental result and the SVC predictionvalue +e outline of this paper is as follows Section 2 in-troduces the implementation procedure of the proposedoptimizationmethod and its key algorithms In Sections 3 and4 a case of plastic lenses is carried out to verify the presentedalgorithm +e conclusions are provided in the last section

2 Method Implementation

+e implementation procedure for the proposed optimi-zation method is described in Section 21 Section 22 il-lustrates the SVC algorithm in detail and the method fortuning the hyperparameters of SVC using PSO is introducedin Section 23

21 Implementation Procedure +e standard weight of aplastic product is determined by technologists prior tomolding +e product weight can be regarded as a functionof the process parameters throughout the injection-moldingprocess Each experiment sample under different processparameters could be heavier or lighter than the standardweight and can therefore be labeled as a positive sample(heavier) or a negative sample (lighter) +e molding pa-rameters for the desired product with a standard weightshould be located between the molding parameters for thepositive and negative samples Hence this study intends tosearch for the parameter classification boundary between thepositive and negative classes as shown in Figure 1+e largerthe weight error of a sample is the further away the point isfrom the boundary and the points on the boundary cor-respond to a zero-weight error

In this study an SVC model is employed to classifyproducts by their weights Products under different processparameters are injection-molded and weighed +eseproducts are labeled as either positive or negative samplesand they are employed as training data for the SVC Asillustrated in Figure 1 the SVC constructs a maximum-margin hyperplane which is also known as the classificationboundary to separate the positive and negative samples +edistance between the samples and the hyperplane corre-sponds to weight errors Optimal parameters should belocated on the hyperplane of the trained SVC model Be-cause of the linearly nonseparable problem further modi-fications of the SVC model are required including theselection and calculation of the kernel function [32] as wellas the determination of the slack variable [33] Mathematicalprinciples are illustrated in Section 22 in detail +e kernelparameter of the kernel function and the penalty parameterof the slack variable have a great influence on the perfor-mance of the SVC model +e determination of thesehyperparameters for the SVC is therefore significant In thisstudy a particle swarm optimization (PSO) method isadopted +e implementation procedure for the proposedmethod can be summarized as follows





f(x) wTx + b (1)


yi times wTx + b1113872 1113873ge 1 (2)


margin x+rarr

minus xminusrarr


||wrarr


||wrarr

|| (3)




min12

w2


⎧⎪⎪⎪⎨

⎪⎪⎪⎩

(4)




21113872 1113873 (5)





min12

w2

+ C 1113944n

i1ξi


ξi ge 0

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(7)










Csample C⊙We (8)


minα

1113944

n

i1αi minus

12

1113944

n

i11113944

n


st 0le αi leC

1113944

n

i1αiyi 0

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(9)








xi xi + vi

⎧⎪⎪⎨

⎪⎪⎩(10)



imax (11)






p p minus min(p)

max(p) minus min(p)

e e minus min(e)

max(e) minus min(e)

(13)












(a) (b) (c)




Productweight (g)





120110100



714735756777798819840861882

Wei

ght (

g)


times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(a)

times10ndash4


1

2

3

4

5

6

7

8

9Ke

rnel

par

amet

er γ

(b)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(c)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ C = 24513

γ = 574

(d)







230

235

240

245

250

255

260

265

270

Inje

ctio

n te

mpe

ratu

re (deg

C)



a0

a1

a2

a3

a4

a5a6 a7

a8

b1

b2

270

265

260

255

250

245

240

235

23070 80 90


Inje

ctio

n te

mpe

ratu

re (deg

C)

100 110 120


0 5 10 15 20 25 30

808

812

816

820

824

828

a1

a2

a0a3


Prod

uct w

eigh

t (g)

Test number


a4

(a)

0 5 10 15 20 25 30

808

812

816

820

824

828

a8

a7

a0

a6

a5

89MPa92MPa95MPa

Prod

uct w

eigh

t (g)

Test number

98MPa101MPaStandard

(b)



5 Conclusions




Data Availability




Acknowledgments


References












828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters


b2

Wei

ght (

g)







































f(x) wTx + b (1)


yi times wTx + b1113872 1113873ge 1 (2)


margin x+rarr

minus xminusrarr


||wrarr


||wrarr

|| (3)




min12

w2


⎧⎪⎪⎪⎨

⎪⎪⎪⎩

(4)




21113872 1113873 (5)





min12

w2

+ C 1113944n

i1ξi


ξi ge 0

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(7)










Csample C⊙We (8)


minα

1113944

n

i1αi minus

12

1113944

n

i11113944

n


st 0le αi leC

1113944

n

i1αiyi 0

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(9)








xi xi + vi

⎧⎪⎪⎨

⎪⎪⎩(10)



imax (11)






p p minus min(p)

max(p) minus min(p)

e e minus min(e)

max(e) minus min(e)

(13)












(a) (b) (c)




Productweight (g)





120110100



714735756777798819840861882

Wei

ght (

g)


times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(a)

times10ndash4


1

2

3

4

5

6

7

8

9Ke

rnel

par

amet

er γ

(b)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(c)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ C = 24513

γ = 574

(d)







230

235

240

245

250

255

260

265

270

Inje

ctio

n te

mpe

ratu

re (deg

C)



a0

a1

a2

a3

a4

a5a6 a7

a8

b1

b2

270

265

260

255

250

245

240

235

23070 80 90


Inje

ctio

n te

mpe

ratu

re (deg

C)

100 110 120


0 5 10 15 20 25 30

808

812

816

820

824

828

a1

a2

a0a3


Prod

uct w

eigh

t (g)

Test number


a4

(a)

0 5 10 15 20 25 30

808

812

816

820

824

828

a8

a7

a0

a6

a5

89MPa92MPa95MPa

Prod

uct w

eigh

t (g)

Test number

98MPa101MPaStandard

(b)



5 Conclusions




Data Availability




Acknowledgments


References












828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters


b2

Wei

ght (

g)





































Csample C⊙We (8)


minα

1113944

n

i1αi minus

12

1113944

n

i11113944

n


st 0le αi leC

1113944

n

i1αiyi 0

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(9)








xi xi + vi

⎧⎪⎪⎨

⎪⎪⎩(10)



imax (11)






p p minus min(p)

max(p) minus min(p)

e e minus min(e)

max(e) minus min(e)

(13)












(a) (b) (c)




Productweight (g)





120110100



714735756777798819840861882

Wei

ght (

g)


times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(a)

times10ndash4


1

2

3

4

5

6

7

8

9Ke

rnel

par

amet

er γ

(b)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(c)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ C = 24513

γ = 574

(d)







230

235

240

245

250

255

260

265

270

Inje

ctio

n te

mpe

ratu

re (deg

C)



a0

a1

a2

a3

a4

a5a6 a7

a8

b1

b2

270

265

260

255

250

245

240

235

23070 80 90


Inje

ctio

n te

mpe

ratu

re (deg

C)

100 110 120


0 5 10 15 20 25 30

808

812

816

820

824

828

a1

a2

a0a3


Prod

uct w

eigh

t (g)

Test number


a4

(a)

0 5 10 15 20 25 30

808

812

816

820

824

828

a8

a7

a0

a6

a5

89MPa92MPa95MPa

Prod

uct w

eigh

t (g)

Test number

98MPa101MPaStandard

(b)



5 Conclusions




Data Availability




Acknowledgments


References












828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters


b2

Wei

ght (

g)











































(a) (b) (c)




Productweight (g)





120110100



714735756777798819840861882

Wei

ght (

g)


times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(a)

times10ndash4


1

2

3

4

5

6

7

8

9Ke

rnel

par

amet

er γ

(b)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(c)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ C = 24513

γ = 574

(d)







230

235

240

245

250

255

260

265

270

Inje

ctio

n te

mpe

ratu

re (deg

C)



a0

a1

a2

a3

a4

a5a6 a7

a8

b1

b2

270

265

260

255

250

245

240

235

23070 80 90


Inje

ctio

n te

mpe

ratu

re (deg

C)

100 110 120


0 5 10 15 20 25 30

808

812

816

820

824

828

a1

a2

a0a3


Prod

uct w

eigh

t (g)

Test number


a4

(a)

0 5 10 15 20 25 30

808

812

816

820

824

828

a8

a7

a0

a6

a5

89MPa92MPa95MPa

Prod

uct w

eigh

t (g)

Test number

98MPa101MPaStandard

(b)



5 Conclusions




Data Availability




Acknowledgments


References












828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters


b2

Wei

ght (

g)




































120110100



714735756777798819840861882

Wei

ght (

g)


times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(a)

times10ndash4


1

2

3

4

5

6

7

8

9Ke

rnel

par

amet

er γ

(b)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ

(c)

times10ndash4


1

2

3

4

5

6

7

8

9

Kern

el p

aram

eter

γ C = 24513

γ = 574

(d)







230

235

240

245

250

255

260

265

270

Inje

ctio

n te

mpe

ratu

re (deg

C)



a0

a1

a2

a3

a4

a5a6 a7

a8

b1

b2

270

265

260

255

250

245

240

235

23070 80 90


Inje

ctio

n te

mpe

ratu

re (deg

C)

100 110 120


0 5 10 15 20 25 30

808

812

816

820

824

828

a1

a2

a0a3


Prod

uct w

eigh

t (g)

Test number


a4

(a)

0 5 10 15 20 25 30

808

812

816

820

824

828

a8

a7

a0

a6

a5

89MPa92MPa95MPa

Prod

uct w

eigh

t (g)

Test number

98MPa101MPaStandard

(b)



5 Conclusions




Data Availability




Acknowledgments


References












828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters


b2

Wei

ght (

g)








































230

235

240

245

250

255

260

265

270

Inje

ctio

n te

mpe

ratu

re (deg

C)



a0

a1

a2

a3

a4

a5a6 a7

a8

b1

b2

270

265

260

255

250

245

240

235

23070 80 90


Inje

ctio

n te

mpe

ratu

re (deg

C)

100 110 120


0 5 10 15 20 25 30

808

812

816

820

824

828

a1

a2

a0a3


Prod

uct w

eigh

t (g)

Test number


a4

(a)

0 5 10 15 20 25 30

808

812

816

820

824

828

a8

a7

a0

a6

a5

89MPa92MPa95MPa

Prod

uct w

eigh

t (g)

Test number

98MPa101MPaStandard

(b)



5 Conclusions




Data Availability




Acknowledgments


References












828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters


b2

Wei

ght (

g)




































5 Conclusions




Data Availability




Acknowledgments


References












828

824

820

816

812

808

804

804b1 a0

81645 81617 81634

Process parameters


b2

Wei

ght (

g)




































optimizationofinjection-moldingprocessparametersfor … · 2019. 7. 4. · (pmma, ht55x produced by...

Documents