Tips & Tricks
Your SlideShare is downloading.
Like this document? Why not share!
Coplanar rectangular patch antenna ...
by IJERA Editor
by jewel billah
GALT Presentation April 2014
by RedChip Companies...
by Melissa Romero
International Journal of Engineerin...
Email sent successfully!
Show related SlideShares at end
Dec 23, 2013
Comment goes here.
12 hours ago
Are you sure you want to
Your message goes here
Be the first to comment
Be the first to like this
Number of Embeds
Flagged as inappropriate
Flag as inappropriate
No notes for slide
Transcript of "20320130406019"
1. International Journal of Advanced Research in Engineering RESEARCH IN ENGINEERING INTERNATIONAL JOURNAL OF ADVANCED and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online)TECHNOLOGY (IJARET) AND Volume 4, Issue 7, November – December (2013), © IAEME ISSN 0976 - 6480 (Print) ISSN 0976 - 6499 (Online) Volume 4, Issue 7, November - December 2013, pp. 161-169 © IAEME: www.iaeme.com/ijaret.asp Journal Impact Factor (2013): 5.8376 (Calculated by GISI) www.jifactor.com IJARET ©IAEME WEDM PROCESS MODELING WITH DATA MINING TECHNIQUES S V Subrahmanyam*, M. M. M. Sarcar** *Assistant Professor, Dept of Mechanical Engg, Gayatri Vidya Parishad College of Engineering, Vizag, A.P. INDIA **Professor and Head of the Department of Mechanical Engg., A.U.College of Engineering, Vizag, A.P. INDIA ABSTRACT To satisfy the needs like Better finish, low tolerance, higher production rate, miniaturization, complex shapes and profiles a different class of modern machining techniques, unconventional in nature, like Wire Electrical discharge Machining (WEDM) emerged. Due to its’ unique machining capabilities WEDM is preferred in various manufacturing industries like Aerospace, nuclear, missile, turbine, automobile, tool and die making. There has been an ongoing research to understand the interaction mechanism between the input parameters Vs output measures like Cutting Speed, Surface Finish and to develop WEDM as a Automated Process System, to avoid as far as possible the operator intervention in machining. In this paper an attempt is made to apply Data mining technique to model the WEDM process and to make the WEDM as Automated Process Control System and Expert System. Using the Experimental data the models are built, trained, tested and validated. It is observed and reported here that desired accuracy is obtained with the built models. INDEX TERMS: C5 for WEDM, QUEST for WEDM, Classification, Data Mining for Wire EDM, WEDM Modelling, Machine Learning, I. INTRODUCTION WEDM involves complex physical and chemical process including heating and cooling. Metal erosion effect takes place in WEDM Process, when electric sparks are generated between the work piece and a wire electrode flushed or immersed in the dielectric fluid. The WEDM machining plays a major role in manufacturing sectors especially industries like aerospace, ordinance, automobile and general engineering etc. This is because the WEDM process provides an effective solution for machining hard materials with intricate shapes, which are difficult to machine through conventional machining methods. [14,15]. 161
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME WEDM provides high accuracy, repeatability, and a better surface finish but the tradeoff is a very slow machining rate and machining tasks take many hours depending on the complexity of the job. Also the cost of machining is very high in WEDM because the high initial investment and cost of the wire–electrode tool. But if WEDM is used to cut difficult to machine materials with complex, precise and accurate contours in low volume and greater variety, then the WEDM process compensates slow machining rate and RI. With Proper selection of suitable input parameter values with a pre-programmed system, which may be termed as Expert System for WEDM, it is possible to predict machining time along with requisite cutting speed, surface finish etc., and increases the Productivity. Improperly selected parameters may result in serious consequences like short-circuiting of wire and wire breakage and in turn reduces productivity. [4,5]. Even though up-to-date CNC-WEDM machines are available, the problem of selection of cutting parameters in WEDM process is not fully solved. Several researchers are trying to develop a method for improving the surface ﬁnish and cutting speed of WEDM process and make it into a Expert System. So far there is no established standard method for predicting machining rate based on the input parameters because of the complex machining mechanism of WEDM. Thus the inability to predict efficient automated machining rate with a method has been one of the major obstacles in developing automated process control systems and expert systems for WEDM. With WEDM’s inherent complexity it is difficult to model WEDM. From the Literature review it is observed that a limited success is achieved in modeling WEDM using artiﬁcial neural networks, fuzzy logic, etc. Hence to model the WEDM into a Knowledge base system, or Expert System, there is a need to develop a set of IF..Then Rules. With data mining technique, relatively a new technique, an attempt is made in this research to generate the simplest IF…THEN rules, with a very high overall accuracy. The experimental data is classified and analyzed Data mining algorithms coupled with categorizing methods. Classification is used to develop a model that maps a data item into one of several predefined classes. II. EXPERIMENTAL DETAILS En31 alloy steel has been considered in the present set of research work. Brass wire of 0.25 mm diameter was used as tool electrode in the experimental set up. SPRINTCUT (AU) WITH PULSE GENERATOR ELPULS 40A DLX CNC Wire-cut EDM machine was used to conduct the experiments. All the axes of the are servo controlled and through a control panel a preprogrammed CNC program code can be fed into the system and can be programmed to follow a CNC code which is fed through the control panel. The 3 sizes of the work piece considered for experimentation on the wire-cut EDM is 5 mm x 5 mm (24 mm,36mm, 49mm thickness). A small gap of 0.025 mm to 0.05 mm is maintained in between the wire and work-piece. The high energy density erodes material from both the wire and work piece by local melting and vaporizing. The dielectric fluid (de-ionized water) is continuously flashed through the gap along the wire, to the sparking area to remove the debris produced during the erosion. Nine Input process parameters such as Pulse On time (TON), Pulse Off time (TOFF), Peak Current (IP), Spark gap Voltage Setting (SV), Wire tension setting (WT), Wire Feed rate setting (WF), Servo Feed Setting (SF), Flushing pressure of dielectric fluid (WP) and Material Thickness (THICK) used in this study. [1, 4, 5, 6, 13, 14, 17, 9]. 162
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME Table 1 Sl.N o. 1 2 3 4 5 6 7 8 9 PARAMETERS Pulse On time Pulse Off time Peak Current Spark gap Voltage Setting Wire tension setting Wire Feed rate setting Servo Feed Setting Flushing pressure of dielectric fluid Material Thickness SYMBO L TON TOFF IP SV WT WF SF WP THICK LEVEL 1 122 63 130 20 4 4 500 9 24 LEVEL 2 125 58 180 30 8 5 1300 12 36 LEVEL 3 128 53 230 40 12 6 2100 15 49 UNITS µsec µsec Ampere Volts Kg-f m/min mm/min Kg/cm2 mm The range values for the input parameters have been carefully chosen based on the experimental evaluation of one factor at a time method. The results of the experiments given in Table 2 are based on Experiments planned using a Custom Design option of IBM SPSS software. IBM software is statistical design software. Data mining techniques does not require following any statistical design. The use of statistical design of experiments was chosen for the simple reason that if the data is not indicative of the domain, then the machine leaning may not find patterns that may be present in the domain.[2,3] Table 2 EXPNOO 1 2 : . : TON 122 122 : 15 16 17 : : 34 35 : 122 122 122 : : 125 125 36 37 : : 48 49 50 : : 71 72 122 128 : : 128 128 125 : : 128 128 TOFF SV 63 20 63 20 : : : : 53 40 63 40 58 30 : : : : 63 40 58 40 53 53 : : 58 53 63 : : 63 58 40 40 : : 40 20 0 30 0 : : 20 30 IP 130 130 : : 180 180 230 : : 130 130 WF 6 6 : : 4 5 6 : : 5 4 150 150 : : 130 230 130 : : 230 230 6 5 : : 6 5 6 : : 6 6 163 WT 12 12 : : 4 4 8 : : 4 8 SF 500 2100 : : 2100 1300 1300 : : 1300 500 THICK 36 24 : : 49 49 24 : : 49 24 WP 15 12 : : 15 9 12 : : 9 9 4 12 : : 4 8 8 : : 12 4 500 1300 : : 1300 2100 500 : : 2100 500 36 49 : : 24 49 24 : : 24 49 9 12 : : 15 9 12 : : 12 12 Ra(µ) 2.4812 3.2549 : : 1.7728 1.7358 1.8458 : : 2.6059 2.6648 2.7012 3.4276 : : 3.5998 3.7742 3.8852 : : 4.0312 4.1025
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME III. M E T HO D O L O G Y Using data mining techniques the experimental results were analyzed. During the process models are constructed from the experimental data pattern with the data mining techniques. The same models are validated with the obtained trained, tested data from the experimental data. The system uses top down induction of decision tree method to analyze the data. The advantage of decision tree method is that decision trees represent rules that can be easily understood by everyone and gives the actual process knowledge to develop a knowledge base system. A set of rules can be derived from this decision tree representing the relationship between the various input parameters of WEDM like, Pulse on, Pulse off, Thickness of material etc. and the output measures namely surface finish and material removal rate [1,2]. 3.1 THE DATA MINING PROCESS Data mining process has the following four steps to follow: 1. Data collection, Data Cleaning, and Data Transformation are the three steps used to gather data, clean the data and to transform the collected data for further processing. This is known as DATA PRE PROCESSING. 2. To find and compare the patterns in data this method was used. This is the most crucial part of the data mining process. Neural networks, Machine Learning and statistical methods are the different data mining algorithms that are used for this purpose. This step is known as PATTERN SEARCH. 3. In the ANALYSIS step the output of the pattern search is analyzed and investigated to decide to stop or to perform a revised search. 4. Finally investigated findings are further interpreted, in the INTREPRETATION step. 3.1.1. DATA PREPARATION In the data mining process Data preparation is an essential step. The data must be reformulated in terms that can suitable to apply the data mining algorithm. The following classification strategy is used: (a) The WEDM process target output values are surface roughness “Ra” and cutting speed “CSPEED”, Material Removal Rate MRR. The Ra and CSPEED, MRR are treated separately. In this process the effect of the input parameters analyzed on CSPEED and then repeated the exercise for surface roughness Ra and MRR. For the present paper only the Ra is considered. (b) A set of attributes/input parameters with their ranges are predefined as shown in table1. (c ) The experiments are carried out with the input parameters and their ranges. In each experiment a work piece of size 5x5x24(or36 or49) is removed by the Machine. The Ra values of each piece measured by Mitutoyo surftest SJ-201P, by taking the average readings on all over the entire cut-off length. (d) The input parameters/attributes values in WEDM can only be changed on the machine in a step-wise manner; hence they will be treated as categorical data. The target values are categorized into four (2, 3, 4, 5) classes by using an incremental categorization approach. The incremental approach is something like a common man’s language which helps in classifying the data based on whether the output is below or above a threshold value. The categorizations of target values were performed by the above two, three, four, and five classes using the mean and standard deviation method. There are several such methods to process the data in data mining. 164
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME (e) In the standard deviation method the threshold values are chosen with the aid of mean and standard deviation by applying some of the basic statistical rules to the outputs in the training set, testing set and validation set. 3.1.2. SEARCH FOR PATTERNS 188.8.131.52. DATA MINING ALGORITHM: There is several rule induction algorithms for classification, and C4.5, ID3, C5, CN2 and QUEST, MARS are examples of such algorithms. Some studies have compared the prediction accuracy, complexity and training time of different classification algorithms. In the present study C5, QUEST (Quick Unbiased Efficient Statistical Test) algorithms have been chosen for the following reasons: One of the main differences between these algorithms is the method of splitting criteria that they use in selecting attributes for root and internal decision nodes in building the decision tree. Also, another difference is that whether an algorithm can split the node into two or more child nodes. The C5 node builds either a decision tree or a rule set. The model works by splitting the sample based on the filed that provides the maximum information gain at each level. The target field must be categorical. Multiple splits into more than two subgroups are allowed. QUEST is a relatively new binary tree-growing algorithm. It deals with split field selection and split-point selection separately. The univariate split in QUEST performs approximately unbiased field selection. QUEST is a binary-split decision tree a1agorithm for classification and data mining. QUEST stands for Quick, Unbiased and Efficient Statistical Tree. The QUEST node provides a binary classification method for building decision tree, designed to reduce the processing time required also reducing the tendency found in classification tree methods to favor predictors that allow more splits. Predictor fields can be ranges, but the target field must be categorical AIl splits are binary.. Then after the 2,3,4,5 classification, the model was built using data from Table 2. Then the data was subjected to the two data mining methods (learning methods to validate the applicability of the rules to the pattern found in the data though the training, testing. The software provides the option of selecting a part of the data for training and another part of data for testing and some another part of the data for validating the suitability of algorithm, classification etc. This is in terms of %. The following % allocation of data chosen by the author as per the standards: For eg., In QUEST algorithm 70% of data were used as training data and 20% and 10% for validating the algorithm were employed as test ones and the a factor equal to 0.05 was used to split data and also the tree was pruned at level five. Time for executing the algorithm is 1 second and in 91.67% of cases algorithm arrival to correct answers. In case of C5 also it is 87.5%. The results are tabulated in Table3, Table 4. IV. RESULTS AND DISCUSSIONS Table 3 165
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME Table 4 From the decision trees it is seen that TON, TOFF, IP, SV, SF, WF and Thickness of the material are the important parameters influencing the Ra. Surface roughness increases with the increase of TON, IP, SF and THICKNESS and decreases with increase in TOFF, SV, and WF and WP. Table 5 2CLSD 1_Training 2_Testing 3_Validation QUEST 100.00% 100.00% 100.00% C5 94.23% 100.00% 91.67% Table 6 3CLSD 1_Training 2_Testing 3_Validation QUEST 100.00% 100.00% 91.67% C5 100.00% 100.00% 91.67% Table 7 4CLSD 1_Training 2_Testing 3_Validation QUEST 94.23% 83.33% 83.33% C5 94.23% 100.00% 83.33% 166
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME Table 8 5CLSD 1_Training 2_Testing 3_Validation QUEST 88.46% 75.00% 75.00% C5 92.31% 86.75% 83.33% Table 9 2CL SD 3CL SD 4CL SD 5CL SD QUEST 100.00% C5 100.00% QUEST 91.67% C5 91.67% QUEST 83.33% C5 83.33% QUEST 75.00% C5 83.33% From the results of data mining with four classes, it is seen that material thickness (THICK) is influencing the material removal rate. As the thickness is increasing the Ra is increasing. The result is obtained from both the C5 and QUEST confirms this. After the Thickness equally important parameters which influence the Ra are TON, IP, SF, SV, WF, WP in the order written in C5 algorithm. In C5 the Order differs but the final conclusion is agreeable with QUEST. It may be noted from the trees and also from the above tables 5, 6, 7, 8 that as the number of classes are increasing the importance of accuracy in pattern search is increasing as result the validation and the misclassifying the training and test data is also increasing. Hence it may be worth noting that as the classification increase into number of classes then the categorizing methods followed should be more math oriented and care should be taken in framing the rules for classification. There are total 8 models (Table 9) for the 4 classes(2,3,4,5 classes) with standard deviation method and 2 data mining machine learning algorithms. Out of 8 models 2 models yield 100% , 2 models yield 91.67% 3models yield 83.33%.Whereas for the last class one model yields 75.0%. Thus an overall basis the success rate is 7/8=85% (75% and above is considered as threshold). Thus if such a model is developed then the trial and error methods to be adopted by the operator will reduce and the machining time reduces and increases productivity and the can be more compatible to CAPP and knowledgebase system. The invalidated cases in the classification are 16.33% up to 16 to 25% in case of QUEST and it is 16.33% in case of C5. It is also to be noted from the results that data misclassifying is only to the tune of 7 to 11.54% (table3) of the cases hence it is presumed that the model performed reasonably well on the test data misclassifying only 16.33% of the cases. If we observe the Trees generated by both the algorithms we understand what parameters to be used to get a better output. For eg. If we take Class 2 to get a low Surface Roughness of High Surface finish we have to choose TON values <=125.332 and corresponding thickness 32.526mm, corresponding to this IP=230, SV=30,SF=1300, WF=4, TOFF=58 for QUEST MODEL. Similarly for C5 Model TON values <=125.332 and corresponding thickness 32.526mm, corresponding to this IP=180, SV=40, WF=5, TOFF=63, SF=2100. 167
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME V. CONCLUSION In this paper WEDM domain was dealt as classification problem. The data mining approach provides the important decision parameters that are useful in obtaining the desired output results and making the WEDM process much more automation process to incorporate into CAPP. The QUEST, C5 algorithms has been used to automatically learn rules from WEDM data. Similarly other algorithms can also be tried with suitable categorization, classification approaches. This approach can be applied to any manufacturing process and can significantly reduce cycle times by increasing the level of automation and also providing insight into the process. For improved performance and prediction accuracy Hybrid approaches in combination of artificial neural networks and decision trees may also be employed. However there will be question when to use C5 and when to use QUEST analysis. QUEST analysis is especially useful to reduce the tendency found in classification tree methods to favor inputs that allow more splits, that is, continuous (numeric range) input fields or those with many categories. The QUEST node provides a binary classification method for building decision trees, designed to reduce the processing time required for large. All splits are binary. For C5 the model works by splitting the sample based on the field that provides the maximum information gain at each level. The target field must be categorical. Multiple splits into more than two subgroups are allowed. DECISION TREES FOR C5 ALGORITHM 2CLASS 3CLASS 4CLASS 5CLASS DECISION TREES FOR QUEST ALGORITHM 2CLASS 3CLASS 4CLASS 168 5CLASS
International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 7, November – December (2013), © IAEME REFERENCES                    Shajan Kuriakose, Kamal Mohan, M.S. Shunmugam Data mining applied to wire-EDM process, Journal of Materials Processing Technology 142 (2003) 182–189. C. Zhou, P.C. Nelson, W. Xiao, T.M. Tirpak, S.A. Lane, An intelligent data mining system for drop test analysis of electronic products, IEEE Trans. Electr. Packaging Manuf. 24 (3) (2001) 222–231 S. Gordon, Linoff, J.A. Michael, Berry, Data Mining Techniques for Marketing, Sales and Customer Support, Wiley, New York, 1997. L. Breiman, J. Friedman, R. Olshen, C. Stone, Classification and Regression Trees, Wadsworth International, Belmont, CA, 1984. S V Subrahmanyam, M. M. M. Sarcar "Evaluation of Optimal Parameters for machining with Wire cut EDM Using Grey-Taguchi Method", International Journal of Scientific and Research Publications, Volume 3, Issue 3, March 2013 S V Subrahmanyam and M. M. M. Sarcar "Parametric Optimization for cutting speed – a statistical regression modeling for WEDM", International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 1, pp. 142 – 150, January - February (2013), © IAEME S.S. Hiremath, P.K. Mishra, On some aspects of erosion rate in wirecut electro-discharge machining, in: Proceedings of the 13th AIMTDR Conference, Calcutta, India, 1988, pp. F-07–F-10. S V Subrahmanyam, M. M. M. Sarcar "Statistical Analysis of Wire Electrical Discharge Machining On Surface Finish",International Journal of Engineering Research & Technology (IJERT)Vol. 2 Issue 3, March – 2013 C. Apte and S.M. Weiss, Data Mining with Decision Trees and Decision Rules D. Scott, S. Boyna, K.P. Rajurkar, Analysis and optimization of parameter combinations in WEDM, Int. J. Prod. Res. 29 (1991) 2189–2207. Lin Zhang, Yan Chen, Yan Liang “Application of data mining classification algorithm for Customer membership Card Model” College of Transportation and Management China 2008. Jiawei Han, Michelin Kamber,“Data Mining Concepts and Techniques” [M], Morgan Kaufmann publishers, USA, 2001, 70-181. J.A. Mc Geough, Advanced Methods of Machining, Chapman & Hall, London, 1988. M.I. Gokler, A.M. Ozanozgu, Experimental investigation of effects of cutting parameters on surface roughness in the WEDM process, Int. J. Mach. Tools Manuf. 40 (2000) 1831–1848. M.S. Shunmugam, S. Sai Kumar, I.K. Kaul, Fuzzy logic modeling of wire-cut EDM process, Proc. SPIE 4192 (2000) 417–425. S.S. Hiremath, P.K. Mishra, On some aspects of erosion rate in wirecut electro-discharge machining, in: Proceedings of the 13th AIMTDR Conference, Calcutta, India, 1988, pp. F07–F-10. SPRINTCUT (AU), Wire EDM of ELECTRONICA Technologies, Reference Manual, vol. 1, 1994. Y.S.Sable, R.B.Patil and Dr.M.S.Kadam, “Investigation of MRR in WEDM for Wc-Co Sintered Composite”, International Journal of Mechanical Engineering & Technology (IJMET), Volume 4, Issue 3, 2013, pp. 349 - 358, ISSN Print: 0976 – 6340, ISSN Online: 0976 – 6359. Mane S.G. and Hargude N.V., “An Overview of Experimental Investigation of Near Dry Electrical Discharge Machining Process”, International Journal of Advanced Research in Engineering & Technology (IJARET), Volume 3, Issue 2, 2012, pp. 22 - 36, ISSN Print: 0976-6480, ISSN Online: 0976-6499. 169