Docstoc

Material Removal

Document Sample
Material Removal Powered By Docstoc
					 International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
       INTERNATIONAL JOURNAL OF MECHANICAL
 6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME
             ENGINEERING AND TECHNOLOGY (IJMET)
ISSN 0976 – 6340 (Print)
ISSN 0976 – 6359 (Online)
Volume 3, Issue 2, May-August (2012), pp. 128-137
                                                                      IJMET
© IAEME: www.iaeme.com/ijmet.html
Journal Impact Factor (2011): 1.2083 (Calculated by GISI)        ©IAEME
www.jifactor.com




        MATERIAL REMOVAL RATE PREDICTION OF C-SIC
        COMPOSITE: COMPARATIVE ANALYSIS OF NEURAL
                NETWORK AND FUZZY LOGIC

         Pallavi.H.Agarwal1 , Prof.Dr.P.M.George 2 and Prof.Dr.L.M.Manocha 3
         1
            Research Scholar, Sardar Patel University, Vallabh Vidyanagar, Gujarat
                                  pallavi_ruhi@yahoo.co.in
     2
       Professor (Mechanical), B.V.M.Engineering College, Vallabh Vidyanagar, Gujarat
 3
   Professor (Material Science), Sardar Patel University, Vallabh Vidyanagar, Gujarat



 ABSTRACT

 Material removal rate is an important objective function in manufacturing
 engineering. It holds the characteristic that is can influence the performance of
 mechanical parts, which is proportional to manufacturing cost. MRR (material
 removal rate) is also an aspect for designing mechanical elements. Material removal
 rate is an essential feature of drilling operation since most of the holes applications
 are required for assembly work. The aim of this experimental and analytical research
 is to identify the parameters which enable the prediction of MRR in drilling. Two
 expert systems are used to analyze the best fit model in predicting the MRR for this
 specific drill job on C-SiC composite. The prediction accuracy is then compared to
 analyze which model could give better results so that it can be recommended for
 machine learning and also future work. It is found that BPN-ANN gives better and
 closer values as compared to the Sugeno ANN model.

 Keywords
  Artificial neural network (ANN), Sugeno fuzzy, fuzzy logic, material removal rate
 (MRR), automatic tool changer (ATC), Back propagation network (BPN)




                                            128
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME

1. INTRODUCTION

In the investigation of drilling of C-SiC, the MRR is considered essential as it influences
the surface finish and the micro geometry of the finished part. Optimum selection of the
process conditions is extremely important as it determines the MRR which in turn affects
the surface finish of the machined parts. Thus in this drilling process an improper
selection of cutting conditions can lead to surfaces with rough finish.
The resurgence of interest in expert systems over the past few decades has opened many
new avenues in various applications. Expert systems lead to greater generality and better
rapport with reality. It is driven by the need for methods of analysis and design, which
can come to grips with imprecision to achieve robustness and low cost production
solution [2]. The use of neural network in machining research has been extensive and
multifaceted. These networks can be trained to recognize arbitrary relations between sets
of inputs and output pairs by adjusting weights of the interconnections. Back
propagation neural network is most commonly used in manufacturing research and
neural network has been used extensively in the past decade to monitor the progress of
tool condition. Fuzzy modeling is based on the idea to find a set of local input-output
relations describing a process. So, the method of fuzzy modeling can express a non-
linear process better than any other ordinary method. As more knowledge about the
system is accumulated the uncertainty diminishes the need for the fuzzy logic treatment
and it can revert to a deterministic or statistical one. The aim of this experimental and
analytical work is to identify suitable parameters, the monitoring of which enable the
prediction of MRR for drilled holes by two expert systems namely Sugeno-Fuzzy and
Neural network. Both have their own ability in determining the output which determines
and maintains the quality of drilled surface. Finally the best expert system has to be
recommended for this drilling job on carbon silicon carbide composite as per their
limitations and advantages by carrying out comparative analysis among the expert
systems within the range of experimental values.

2. EXPERIMENTAL PROCEDURE

 The work piece material used is carbon silicon carbide composite .The machine used
 is Den ford CNC machine with Fanuc controller. The VMC has a maximum cross
 travel of 170 mm, maximum long travel of 290 mm and head travel of 235 mm with a
 six station ATC. The spindle speed range available is 0-4000 rpm and the feed range
 available is 0-1000 mm/min. Factorial approach [16] which is very effective to deal
 with responses influenced by multi variables is used to conduct the experiments. This
 method is a powerful design of experiment tool, which provides a simple, effective and
 systematic approach to experimentation. This method reduces the number of
 experiments that are required to model the response functions. Traditional
 experimentation is one factor at a time experiment, where one variable is changed
 while the rest are held constant. The major disadvantage of traditional experimentation
 is that it fails to consider any possible interactions between the parameters. An
 interaction is the failure of one factor to produce the same effect on the response at
 different levels of another factor. The factorial design approach has many advantages.
 It allows the interactions to be evaluated so that misleading conclusions can be

                                            129
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME

 avoided. This approach also allows the effects of a factor to be estimated at several
 levels of the other factors, yielding conclusions that are valid over a range of
 experimental conditions. It is also impossible to study all the factors and determine
 their main effects in a single experiment. The machining parameters used are as shown
 in Table 1. The experiments are planned according to 23 with 4 center points. The
 design matrix with the response parameter MRR is as shown in Table 2.

Table 1: Machining parameters and their levels
            Sample     Machining        Unit      -1         0         1
                       Parameters
               A         Spindle        Rpm      1500      2250      3000
                          speed
               B        Feed rate      mm/min     30        35         40

               C         Drill size     Mm        1          2         3



Table 2: Design Matrix with response parameter MRR

Experiment No                Factors             MRR(g)
                     A           B        C
       1             -           -        -      0.0145
       2             +           -        -      0.0159
       3             -           +        -      0.0132
       4             +           +        -      0.0248
       5             -           -        +      0.0222
       6             +           -        +      0.0003
       7             -           +        +      0.0004
       8             +           +        +      0.0529
       9             0           0        0      0.0206
      10             0           0        0      0.0482
      11             0           0        0      0.0141
      12             0           0        0      0.0003


3. EXPERT SYSTEMS

A. Artificial Neural Network

Recent research activities in artificial neural networks (ANNs) have shown that ANNs
have powerful pattern recognition and classification. ANNs are suited for problems
which require knowledge that is difficult to specify but for which there are enough data
or observations. They learn from the training data and capture information about
relationships among the data even if underlying relationships are unknown or hard to

                                           130
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME

describe. ANNs are universal approximations [3] it has been shown that a network can
approximate any continuous function to any desired accuracy by many researchers [4],
[5]. Neural network which uses back propagation algorithms for modeling has been
developed using machining process parameters, spindle speed, feed rate and drill size as
input parameters and material removal rate as output parameter. Making connections
from the input layer to the output layer improves the learning efficiency. Out of the
experimental data generated, the training data is used to train the model and this data is
not used for testing and validation.
The BP algorithm was developed by Paul Werbos in 1974 and rediscovered by
Rumelhart and Parket. Since its rediscovery, the back propagation algorithm has been
widely used as a learning algorithm in the feed forward multilayer neural network. The
back –propagation algorithm is applied to feed forward ANN’s with one or more number
of hidden layers as shown in Fig.1.Based on this algorithm, the network learns a
distributed associative mapping between the input and output layers. What makes this
algorithm different than the others is the process by which weights are calculated during
the learning phase of the network. The input layer is used to feed the data in the network.
The inputs are subsequently modified based on the interconnection weights between
layers. The net input to each neuron from the preceding layer will be as:
        N
  netj= ∑ wi jx i + b j
        i =1




                          Fig.1 Feed Forward Multilayer Perceptron

Where, netj is the net input, N is the number of neurons of the inputs to the jth neuron in
the hidden layer. Wij is the interconnected weight from the ith neuron in the forward layer
to the jth neuron in the hidden layer, xi is the input from the ith neuron in the hidden layer
and bj is the bias value corresponding to the neuron. The output signal is obtained by
applying activations to the net input. In this particular analysis, multilayer back
propagation neural network model is developed using MATLAB neural network toolbox
for predicting the material removal rate in drilling C/SiC material. There is one input
vector with three elements. The value of the first element of the input vector ranges
between 1500 to 3000 rpm, the values of the second element of the input vector ranges
between 30-40 mm/min and the value of the third element of the input vector ranges
between 1-3 mm. There are twenty neurons in the hidden layer and one neuron in the
output layer. The number of neurons in the hidden layer was chosen based on the MSE
value which is least between 20 to 25 nodes as shown in Fig.2


                                             131
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME

The transfer function in the hidden layer and the output layer is tan-sigmoid. The
training function is trainlm which updates weight and the bias values according to the
Levenberg-Marquadrt optimization rule.




Fig 2 Effect of number of nodes in the hidden layer

Network learning function is learngdm which is the gradient descent with momentum
weight and bias learning function. The data is divided as training data to train the neural
network. The testing data is used to test the model; it does not take part in the training of
the model. The mean squared error (MSE) is the criterion for selecting the network
structure. Here the error is calculated as the difference between the target output and the
network output
The experimental data is used to train and validate the ANN. The prediction of the MRR
by neural network is shown in Table 3.

B.FUZZY SYSTEM

Fuzzy logic has lot of application in the real world. Basically the system will accept the
input or some inputs and pass the inputs to a process called fuzzification. In the
fuzzification process, the input data will undergo some translation into the linguistic
quantity as low, medium, high of the physical properties. The translated data will be sent
to an inference mechanism that will apply the predefined rules. The inference engine
generates the results in the linguistic form. The linguistic output will go through
defuzzification process to be in numerical form. Defuzzification is defined as the
conversion of a fuzzy membership function to precise or crisp quantity [9], [10]. Fuzzy
modeling and approximation are the most interesting fields where the fuzzy theory can
be effectively applied. As far as modeling and approximation is concerned one can say
that the main interest is towards the applications when we intend to apply fuzzy
modeling and approximation to an industrial process. One of the key problems to be
solved is to find the fuzzy rules.

       Sugeno – Type fuzzy inference

The most commonly known or used fuzzy inference methodology is Mamdani. But this
paper discussed the Sugeno or Takagi-Sugeno-Kang, method of fuzzy inference. The
main difference between Mamdani and Sugeno is that the Sugeno output membership
functions are either linear or constant but can be excellently suited for modeling non-
linear systems by interpolating between multiple linear models.
A typical rule in a Sugeno fuzzy model has the form:
If Input 1=x, Input 2=y, then output z=ax+by+c.

                                             132
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME

For a zero –order Sugeno model,
The output level Z is a constant (a=b=0). The output level Zi of each rule is weighted by
the firing strength Wi of the rule. For example, for an AND rule with Input 1=x and
Input 2=y the firing strength is
Wi= and method [F1(x), F2(y)]
Where F1 and F2 are the membership functions for the input 1 and 2, the final output of
the system is the weighted average of all rule outputs, computed as shown by equation
stated below and the Sugeno rule operates as shown in Figure 3

Final Output = (           )/(         )




                            Fig.3 Sugeno-Fuzzy Output Rule

Due to linear dependence of each rule on the input variables of the system, the Sugeno
method is ideal for acting as an interpolating supervisor of multiple linear controllers
that are to be applied, respectively, to different operating conditions of a dynamic non-
linear system. A Sugeno fuzzy inference system is extremely well suited to the task of
smoothly interpolating the linear gains that would be applied across the input space, it is
a natural and efficient gain scheduler. A Sugeno system is well suited for modeling non-
linear systems by interpolating between multiple linear models [10]. The plot of
membership functions and the input-output variables fed into the Fuzzy Inference
System [FIS] is shown in Figure 4, where L,M and H denotes Low, Medium and High
respectively.




                                            133
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME




                      Fig.4 Membership function of Sugeno-Fuzzy

Upon developing the membership function, precise rules have been fed into the system
relating the FIS input-output variables. Each of these rules plays an important role in
generating the fuzzy logic controller model and the accuracy of the numerical output.
Upon the rules determination, the fuzzy logic controller will simulate the FIS variables
with respective rules and modeling of the controller toolbox will take place. The model
controller toolbox to the system is shown in the Figure 5.




                         Fig.5 Controller tool box for each rule.

4. RESULTS AND DISCUSSION

Fig. 6 and 7 are the Sugeno-Fuzzy based surface model showing an excellent
relationship between the two sets of input variables: speed and feed rate and drill size.
Fig.6 with the output material removal rate plotted against speed and feed rate and in the
Fig.7 the output material removal rate is plotted against speed and drill size. The



                                           134
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME

inference drawn is that for medium feed rate, drill size and spindle speed the MRR rate
is higher for carbon silicon carbide composite.




                Fig.6 Sugeno Fuzzy model (spindle speed & feed rate)




                Fig.7 Sugeno Fuzzy model (spindle speed & drill size)




                Fig. 8 Sugeno Fuzzy model(spindle speed and drill size)




                                          135
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME


Table 3 Output Result Table

Experiment                1                  2                   3
Actual MRR             0.0222             0.0529              0.0482

Predicted              0.0282             0.0205              0.0206
MRR(BPN)
Error(BPN)             -0.0065           -0.0323             -0.0276

Fuzzy MRR              0.0010             0.0025              0.0206
Error(fuzzy)           0.0212             0.0504              0.0276

 From the numerical data, it is clear that the BPN system has produced closer output as
compared to Sugeno-fuzzy system. It has been studied that Fuzzy has the ability of
predicting the future (forecasting) based on the membership function of the input and the
output variables, limits and rules fed. Although its values are not the best, but it also
matches closely to the actual.

5. CONCLUSION

           •   BPN has shown the capability of generalization and prediction of material
               removal rate in drilling within the range of experimental data.
           •   The maximum deviation observed and estimated by BPN is minimal.
           •   The present work can be extended with different process parameters,
               material thickness and type to test the ability of the expert systems in
               prediction of the output and these findings can then be applied to indirect
               tool condition monitoring in unmanned manufacturing system.
           •   The predicted values of the ANN output can be further improved by
               increasing the weights of the experiments.

6. ACKNOWLEDGEMENTS

I am thankful to Mr.Sreeram R. who introduced me to the domain of fuzzy logic and
neural network.

7. REFERENCES

[1] Rot Berg and Ber, “Methods for drilling parameters evaluation applied for drill point
development “, Annals of CIRP, vol. 36, pp.49-51, 1997.

[2] Matsumura T. and Obikawa T, “On the development of expert system for selecting
the optimum cutting “, Journal of the Japan Society of Precision Engineering, 58,
pp.1691-1696, 1992.



                                           136
International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 3, Issue 2, May-August (2012), © IAEME

[3] Kuo. R.J.,”Intelligent tool wear system through artificial neural networks and fuzzy
modelling “, Journal of Artificial Intelligence in Engineering, 5, pp.229-242, 1998.

[4] Choudhury S.K. and Jain V.K., “Online monitoring of tool wear in turning using a
neural network “, International Journal of Machine tools and Manufacturing, 39, pp.489-
504, 1999.

[5]Rahaman M, and Zhou Q, “Online cutting state recognition using a neural
network,”International Journal of Advanced Manufacturing Technology, vol.2, pp.87-
92, 1995.

[6] Eshima T and Shibasaka,”Estimation of cutting tool life by processing tool images
data with neural network”, CIRP Annals, vol.42, pp.59-62, 1993.

[7] Rangwala.S, and Dornfeld, “Sensor integration using neural networks for intelligent
tool condition monitoring,”ASME Journal of Engineering and Industry, 112, pp.219-
228, 1990.

[8] Masory, “Monitoring machining processes using multisensory readings fused by
artificial neural networks “, 7th International Conference on Computer aided production
Engineering, vol.28, pp.231-240, 1991.

[9] Li.P.G. and S.M.Wu,”Monitoring –drilling wear states by a fuzzy pattern recognition
technique,” Journal of Engineering for Industry, 110, pp.297-302, 1998.




                                          137

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:11/19/2012
language:Unknown
pages:10