Optimization Of Machining Parameters In Milling Of Composite . - Ijmerr
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012ISSN 2278 – 0149 www.ijmerr.comVol. 1, No. 2, July 2012 2012 IJMERR. All Rights ReservedResearch PaperOPTIMIZATION OF MACHINING PARAMETERSIN MILLING OF COMPOSITE MATERIALSM Muthuvel1* and G Ranganath1*Corresponding Author: M Muthuvel, muthuvel.ace@gmail.comIn this paper optimization of End milling has been reported. In recent years GFRP have attractedincreasing use for many purposes. The material has many excellent properties, such as highspecific strength, high specific modulus of elasticity, light weight, good corrosion resistance,etc., the parameters are depth of cut, feed, speed and tool were varied. The experiments weredesigned based on statistical three level full factorial experimental design techniques. BackPropagation Feed Forward Artificial Neural Network (BPFF-ANN) has been used for predictionof surface roughness and Delamination. In the development of predictive models the cuttingspeed, feed, depth of cut and tool type were considered as the model variables. Twenty sevendata were used for training the network. The required datas for predictive model are obtained byconducting a series of test and measuring surface roughness and delamination data. Goodagreement is observed between the predictive model results and the experimentalmeasurements.Keywords: GFRP, ANN, Back propagation, Delamination, Surface roughnessINTRODUCTIONparameters have recent significant contributionon the surface roughness and depth of cut andhardness of material have less significantcontribution on the surface roughness (Sijoand Biju, 2010). In this experiment is executedby using full factorial design. Analysis ofvariances shows that the most significantparameter is feed rate followed by spindlespeed and lastly depth of cut. After thepredicted surface roughness has beenobtained by using both methods, averageSurface roughness is an indicator for thesurface quality is one of the prime customerrequirements for the machined parts. Forefficient use of machine tools, optimum cuttingparameters are required. During Machiningprocess parameter optimization is highlycomplex and time consuming. Taguchiparameter optimization methodology isapplied to optimize cutting parameters. Thenthe results analysis show that the cutting1Department of Mechanical Engineering, Adhiyamaan College of Engineering, Hosur 635109, Tamil Nadu, India.277
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012percentage error is calculated. Themathematical model developed by usingmultiple regression method shows theaccuracy of 86.7% which is reliable to be usedin surface roughness prediction. On the otherhand, artificial neural network technique showsthe accuracy of 93.58% which is feasible andapplicable in prediction of surface roughness(FAb Rashid and Abdul, 2010). In this studymathematical model may be used inestimating the surface roughness withoutperforming any experiments. Finally, predictedvalues of surface roughness by techniques,NN and regression analysis, were comparedwith the experimental values and theircloseness with the experimental valuesdetermined. Results show that, NN is a goodalternative to empirical modeling based on fullfactorial design (Esme et al., 2009). Here todetermining suitable training and architecturalparameters of an ANN still remains a difficulttask. These parameters are typicallydetermined in trial and error procedure, wherea large number of ANN models are developedand compared to one another. Taguchi methodfor the optimization of ANN model trained byLevenberg-Marquardt algorithm. A case studyof a modeling resultant cutting force in turningprocess is used to demonstrateimplementation of the approach. The ANNtraining and architectural parameters werearranged in L18 orthogonal array and thepredictive performance of the ANN model isevaluated using the proposed equation. Usingthe analysis of variance (ANOVA) and analysisof means (ANOM) optimal ANN parameterlevels are identified. Taguchi optimized ANNmodel has been developed and has shownhigh prediction accuracy. Analyses andexperiments have shown that the optimal ANNtraining and architectural parameters can bedetermined in a systematic way, therebyavoiding the lengthy trial and error procedure(Milos and Mirislav, 2011). Numerical andArtificial Neural Networks (ANN) methods arewidely used for both modeling and optimizingthe performance of the manufacturingtechnologies. Optimum machining parametersare of great concern in manufacturingenvironments, where economy of machiningoperation plays a key role in competitivenessin the market. Effects of selected parameterson process variables (i.e., surface roughnessand material removal rate) were investigatedusing Response Surface Methodology (RSM)and artificial neural networks (Soleymani andKhorram, 2010). So based on these surveysto be selected the work piece material ascomposite because widely used in manypurpose now a days, then during machiningprocess the main failure of the materials dueto the surface roughness and Delamination.Due to these facts, optimum the surfaceroughness and delamination values then onlyable to reduce the material wastage duringmachining process and also the materialwidely used in all purposes.CUTTING CONDITIONSExperimental DesignDesign of experiments is a powerful analysistool for modeling and analyzing the affect ofprocess variable over some specific variablewhich is an unknown function of these processvariables. The experimental design method isan effective approach to optimize the variousmachining parameters. The selection of suchpoints in the design space is commonly calledDesign of Experiments (DoE) or Experimental278
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012powerful design of experiments tool forengineering optimization of a process. It is animportant tool to identify the critical parametersand predict optimal setting of each parameter.Analysis of variance is used to study the effectof process parameters and establishcorrelation among the cutting speed, feed anddepth of cut with respect to the majormachinability factor, cutting forces such ascutting force and feed force. Validations of themodeled equations are proved to be wellwithin the agreement with the experimentaldata (Dinesh et al., 2008). A three level fullfactorial design creates 3n training data,where n is the number of variables. In thesestudy four independent variables such asdepth of cut, speed, feed rate and tool typehas used for experimental runs are shown inthe Table 1. Ranges of process parametersare shown in the Table 2.Design. The choice of the experimental designcan have a large influence on the accuracy andthe construction cost of the approximations.Randomly chosen design points make aninaccurate surface to be constructed or evenprevent the ability to construct a surface at all.Several experimental design techniques havebeen used to aid in the selection of appropriatedesign points. In a factorial design variablerange is divided into levels between the lowestand the highest values (Arbizu and Perez,2003). Experiments were conducted throughthe established Taguchi’s design method. Inthis work, the machining characteristics areinvestigated based on surface roughness andtool wear. The machining parameters are alsooptimized by employing statistical techniques,using the technique of analysis of varianceobtained from regression analysis (Myers andMontgomery, 1995). Taguchi method is aTable 1: Levels of the Variables Used in this WorkFactorsLevel 1Level 2Level 3Cutting velocity244872Feed300600900Depth of cut0.51.52.5Type of tool123Table 2: Experimental Results Obtained from Machining Surface and Cutting ParametersS. No.Input ParametersOutput ResultsSpeedFeedDOCTool face RoughnessDelamination
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012Table 2 (Cont.)S. No.Input ParametersOutput ResultsSpeedFeedDOCTool 7.729001.534.171.154Measurement and ResultSurface RoughnessDelaminationalternative approach. Until today many differentneural network models have been developed.They include perceptrons, Kohonen, Hassoun,Yuille, Hebbian, Oja, Hopfields, Backpropagation and Kolmogorov Networks, tomention a few of the better known networkmodels. Among the various neural networkmodels Back Propagation (BP) is the bestgeneral purpose model and probably the bestat generalization. The typical neural networksarchitecture is shown in the Figure 1. The inputlayer, the hidden layer and the output layerinclude several processing units known asneurons. The input layer is used to present theTerms and UnitsCutting Velocity: m/min, Feed: mm/min,Depth of cut: mm Surface roughness: µm,Delamination: µm.Artificial Neural Network Mode forPrediction of Surface RoughnessArtificial Neural Network is a capablecomputation model for a weight diversity ofproblems. For manufacturing process whereno satisfactory analytic model exist or a loworder empirical polynomial model isinappropriate, Neural networks offer a good280
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012Figure 1: Neural Network ArchitectureHidden LayerInput utputDepth of CutSurfaceRoughnessTool Useddata in the network model and the output tocreate the ANN’s response.NNet j w ijJ 0There are several transfer functions such asthreshold function, piece wise-Linear function,sigmoid/hyperbolic function and logarithmicused in neural network models. Tangenthyperbolic activation function was selected inthis work. For the prediction of surfaceroughness, in this study a multilayerperceptrons consisting of an input, two hiddenlayers and an output layer was used as shownin Figure 1. The optimal ANN architecture wasdesigned by means of MAT Lab NeuralNetwork toolbox. Neurons in the input layercorrespond to depth of cut, cutting speed andfeed rate. The output layer corresponds tosurface roughness and Delamination. In thismodel, the inputs are fully connected to theoutputs. Input and output layers have 4-36-2neuron, respectively as shown in Figure 1. Inthe neural network model, the output neuronson the input layer reach the jth neuron on thenext layer and become its input as stated asin Equation (1).(1)Where N is the number of neurons of theinputs to the jth neuron in the hidden layer andNetj is the total or net input. Xi is the input fromthe ith neuron in the preceding layer and wij isthe weight of between the ith neuron on the inputlayer and the j-th neuron on the next layer. Atangent hyperbolic function (f) that transformsthe input value of the hidden layer to produceits output (outj)The back propagation algorithmX k 1 X k k g k.(2)The back propagation is used as learningprocedure for multi layer perception network.The algorithm makes it possible to propagateerror from the output layer to the input layerand correct the weight vectors, which willresult in minimum error. The back propagationalgorithm minimizes the square of thedifferences between actual output anddesired output units and for all training pairs.281
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012Figure 2: A Typical Network ArchitectureSUMOutput to OtherNeuronsLimiter(Sigmoidal Function)Weighted InputsThe activation function f(x) is a non linearfunction and is given byThe error obtained when the training pair(pattern) consisting of both input and outputgiven to the input layer of the network is givenby equation (MSE).1Ep n T O 2f(x) a tan sig(n) 2/(1 exp(–2*n)) –1.(6)where f(x) is differentiable.(3)piiPurelin is a neural transfer function. Transferfunctions calculate a layer’s output from its netinputwhere,Tpi is the ith component of the desired outputvector;a purel in (n) nOpi is the calculated output of ith neuron inthe output layer.Thus, the result found after the developmentof the ANN model the result comparison to begiven in the Table 3. In that comparison themachine data and the ANN output having tooless variation between them.The overall error of all patterns is given byE Ep.(4)Training Function and learning functions aremathematical procedures used toautomatically adjust the network’s weights andbiases. The training function dictates a globalalgorithm that affects all the weights andbiases of a given network. The learningfunction can be applied to individual weightsand biases within a network. X k 1 X k J T J I 1 JT e.(7)ANN APPROACH: RESULTSAND COMPARISONTraining of neural network model wasperformed using twenty seven experimental.The trained network model was tested usingother experimental data points, which were notused in the training process. The resultspredicted from the ANN model are comparedwith those obtained by experimental test in.(5)282
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012Table 3: Machining Output vs. ANN Output for Surface Roughness and DelaminationTestNo.Surface RoughnessActual OutputANN OutputDelaminationErrorActual OutputANN 0.00611.1541.3023–0.14830Table 3 and the training set patterns in theTable 2 that ANN prediction is in goodagreement with the experimental results.Figures 3 and 4 compare the neural networksurface roughness and Delaminationprediction with the experimental test result andthe ANN result.Here that the Simulated ANN Output forSurface roughness and Delamination werefound for the (Table 4) sample Data’s and thosevalues are found with good interpolation andthen this method is very useful for predictionof various combinations of input data’s withoutundergoing the Experimental process. With283
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012Figure 3: Machining Output vs. ANN Output for Surface RoughnessMachiningANNFigure 4: Machining Output vs. ANN Output for DelaminationMachiningANNTable 4: Simulated ANN Output for Surface Roughness and DelaminationS. No.Input for ANNSimulated ANN ResultsSpeedFeedDOCTool 727000.525.021.728284Surface RoughnessDelamination
Int. J. Mech. Eng. & Rob. Res. 2012M Muthuvel and G Ranganath, 2012help of these method will be reduce themachining time as well as getting goodmachinability.and Materials Sciences, Vol. 16,pp. 44-50.3. Esme U, Sagbas A and Kahraman F (2009),“Prediction of Surface Roughness in WireElectrical Discharge Machining UsingDesign of Experiments”, Iranian Journal ofScience & Technology, Transaction B,Engineering, Vol. 33, No. B3, pp. 231-240.It is found that the developed ANN modelhas good interpolation capability and can beused as an efficient predictive thecombinations for good surface roughness andDelamination. Increasing the number of nodesincreases the computational cost anddecreases the error.4. Milos J Madic and Mirislav R Radocanovic(2011), “Optimal Selection of ANN Trainingand Architectural Parameters UsingTaguchi Method: A Case Study”, FMETransactions, Vol. 39, pp. 79-86.CONCLUSIONThe experimental observations wereincorporated into the ANN model. A feedforward neural network was developed topredict surface roughness and Delamination.Good agreement was shown between thepredictive model results and the experimentalmeasurements. As in future without undergoingthe machining process able to get goodmachining data’s and its very useful ANNmodel for getting good Optimum machiningprocess.5. Myers R H and Montgomery D C (1995),“Response Surface Methodology Processand Product Optimization Using DesignedExperiments”, Wiley, NewYork, USA.6. Rashid M F and Abdul Lain M R (2010),“Surface Roughness Prediction of CNCMilling Proces Using Artifical NeuralNetwork”, Proceedings of the WorldCongress on Engineering, Vol. III, WCE.REFERENCES7. Sijo M T and Biju N (2010), “TaguchiMethod of Optimization of CuttingParameters in Turning Operations”, DOI:02.AMAE.2010.01.536.1. Arbizu I P and Perez C J L (2003), “SurfaceRoughness Prediction by Factorial Designof Experiments in Turning Processes”, J.Mater. Process Technol., pp. 143-144.8. Soleymani Yazdi M R and Khorram A(2010), “Modelling and Optimization ofMilling Process by Using RSM and ANNMethods”, IACSIT International Journal ofEngineering and Technology, Vol. 2,No. 5, October, ISSN 1793-8236.2. Dinesh Thakur, Ramamoothy B andVijayaraghavan L (2008), “Optimization ofHigh Speed Turning Parameters of SuperAlloy Inconel 718 Material Using TaguchiTechnique”, Indian Journal of Engineering285
efficient use of machine tools, optimum cutting parameters are required. During Machining process parameter optimization is highly complex and time consuming. Taguchi parameter optimization methodology is applied to optimize cutting parameters. Then the results analysis show that the cutting ISSN 2278 - 0149 www.ijmerr.com Vol. 1, No. 2, July .
equation 2, the S/N values of machining performance of the obtained MRR values are computed. In order to obtain the effects of machining parameters for each level, the S/N values of each fixed parameter and level in each machining performance were summed up. 1.2 Process Parameters Selection In this analysis, WEDM parameters such as T on
There are different types of machining process used for sapphire material. The fig. 1 shows a graphical representation of sapphire machining processes i.e. laser machining process, grinding process, polishing process, lapping process, new developed machining process, compound machining process and electro discharge machining process. Fig.1.
Machining metals follows a predictable pattern with minimal creep. When machining plastics, quick adjustments must be made to accommodate substantial creep — not to mention that the material has a strong propensity for chipping and melting during machining. Simply stated, the basic principles of machining metals do not apply when machining
PCI Machining - Absolute Machine Tools 7944 Innovation Way Mason OHIO 44053 USA 1 (513) 701-5550 sales-usa@pci-machining.com PCI Machining Rue Copernic 42030 - Saint-Étienne FRANCE 33 4 77 42 60 88 sales@pci-machining.com services@pci-machining.com PCI Machining Vertriebsbüro Deutschland Lebacher Straße 4
Machining metals follows a predictable pattern with minimal creep. When machining plastics, quick adjustments must be made to accommodate substantial creep — not to mention that the material has a strong propensity for chipping and melting during machining. Simply stated, the basic principles of machining metals do not apply when machining
where the use of 5-axis simultaneous machining brings unequalled surface quality. Moreover, it is targeted at prototype machining, 5-axis trimming and special machining where full 5-axis machining is the requirement for quick and accurate manufacturing. Multi-Axis Surface Machining is also an add-on product to Prismatic Machining and Lathe .
on the parameter optimization of CNC lathe machining for surface roughness using the Taguchi method, where surface roughness generated during machining. In the parameter optimization, the parameters are cutting speed, feed, and depth of cut. After selecting parameters turning on CNC lathe is to be done and selected orthogonal array and parameters used for the optimum set of combined controlled .
Optimization of machining parameters using Taguchi coupled GRA 167 Table 3: Input parameters with Taguchi design Machining Parameters Levels of Parameters 1 (low) 2 (medium) 3 (high) Cutting speed (v) m/min 42 60 108 Feed rate (f) mm/rev 0.1 0.2 0.3 Depth-of-cut (d) mm 0.25 0.5 0.75 Evaluation of optimal cutting parameters
