Vous êtes sur la page 1sur 9

Australian Journal of Basic and Applied Sciences, 5(12): 2752-2760, 2011 ISSN 1991-8178

Using of Artificial Neural Networks to Predict Drill Wear in machining processes

1 Aydin Salimi, 2 Samira Abbasgholizadeh, 3 Samad Taghizadeh, 4 Asghar Safarian

1 M.Sc of Mechanic, Lecture of Mechanic, Faculty of Engineering, Department of Mechanic Engineering, University of Payam Noor. Tehran, Iran 2 Mechanics and Industrial Engineering Department, Faculty of Engineering, Payam Noor University of Tabriz-Iran. 3 Dept. of Manufacturing Engineering, Islamic Azad University- Maragheh Branch. Maragheh, Iran. 4 Dept. of Manufacturing Engineering, Islamic Azad University- Maragheh Branch, Iran.

Abstract: In machining operations a hard tool is engaged with work piece along with process. Tool is harder than work piece. However, tool wear occurrence in machining processes is inevitable. Tool wear will results in scraped parts and also it makes tool to weaken and then a tool failer will happen in the end. Therefore, an operator is needed to follow the process and change the tool when it is going to break. But this is a serious problem against automation. To create an automation system, we need to develop a monitoring system to predict tool wear rate by on-line and substitute it with an operator. In this paper by using of a wear model and experimental data and also motor current block diagram ,tool wear rate in drilling process will be predicted .To investigate the results, neural network method is used .The results compared with the real data show that the neural network results have a close fitness with the real data.

Key words: Monitoring, Current signal, Drill wear, Neural Network, Cutting force.

INTRODUCTION

Tool wear detection is one of the most important problems found during manufacturing in automated C.N.C machine tools. The reason for acquiring the drill wear information and its monitoring is to enhance the predictive capability to allow the machine operator to schedule tool change or regrind just in time to avoid underuse or overuse of tools ,avoid shutdown of machines due to damage, and to minimize scrap or rework. On the other hand, drill wear affects the ability of the hole cutting system to satisfy specified performance characteristics, such as hole roundness, centering, burr formation at drill exit, and surface finish. Although, many researchers have been performed for tool condition monitoring, there is no practical method to predict wear rate in drilling process without some problems and limitations (Zhang, M.Z., Liu, Y.B., Zhou, H, 2001). Researchers use two methods for monitoring tool condition .The first is direct methods. In these methods, tool and work piece are in contact with each other. Optical methods are used in direct tool condition monitoring but, because it cannot be used in Real-time tool condition monitoring system, researchers prefer to use indirect methods to monitor tool condition along with the machining processes. Indirect methods that rely on the relationship between tool condition and measurable signals (such as force, acoustic emission, vibration, current, etc.) for detecting tool conditions have been extensively studied. Among these methods ,cutting force has more accuracy because cutting force signals are measured directly on the tool, therefore the structure of machine does not affects the measured data and extracted data accuracy is higher. But on the other hand, dynamometer is needed to measure cutting force signals ,therefore the measurement process cost goes high too. Also, the method based on sensing motor current is considered as one of the major methods (Xiaoli, Li. and Tso, S.K, 1999). Discussed the feasibility of the motor current sensing for adaptive control and tool condition monitoring (Jantunen, Erkki, 2002). Described the use of current signals measured from spindle motor and feed motor to estimate static torque and thrust for monitoring tool conditions. The major advantage of using the measurement of motor current to monitor tool condition in cutting processes is that fixed current sensors do not disturb machining processes, and the cost of the sensors is very low (Xiaoli Li, Shiu Kit Tso, 2000). In this paper, both force and motor current signals are used for monitoring drill wear in drilling process. Using of two methods, force and current data, would result in accurate prediction of tool wear rates. For obtaining the data, empirical models are used. First experimental data of force for a new tool are given to wear model and tool wear rates are estimated for a certain machining parameters. Then machine tool drive block diagram is used to measure motor current. Finally, all of the data are investigated by neural network method to predict tool wear rate. In this paper, PYTIA software is used to design the neural network program. Neural network program is trained by the

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

measured data. Then it can estimate tool wear rate for other inputs. Also tests will be run to prove the reliability of the network.

Tool Wear Model:

Tool wear has a nonlinear attitude in machine tools. There are many of factors that influence on tool wear such as friction, temperature, chemical materials and etc. The effective value of the factors is not known. Therefore there is no any analytical model to predict tool wear rate in machining process. For a new drill, analytical models of force prediction are developed but for a worn drill it is not easy. We suppose that the

cutting force value for a new drill is

.As the machining process continues; tool starts to wear and becomes

worn little by little. Then the sharpness of tool decreases and tool needs more cutting force to cut the work piece. As mentioned above, this extra force depends on the wear rate, friction factor, cutting length and other parameters. Total force value is shown:

F cut

F F

cut

kwb

(1)

In the formula (12), k, is a constant value, w, is wear rate and b, is cutting length. To find an empirical model for tool wear, experimental data are needed. The block diagram and state equations of wear model has been explained by these researchers (Carrillo, F.J. and Zadshakoyan, M, 1997). In this paper the so called wear model is used to find expected drill wear rate for the certain cutting force values and machining parameters. Fig.1 shows simulated tool wear diagram in MATLAB software.

Fig.1 shows simulated tool wear diagram in MATLAB software. Fig. 1: Simulated tool wear diagram obtained

Fig. 1: Simulated tool wear diagram obtained from wear model for drill diameter 15.9mm, point angle118, helix angle 33, pilot whole diameter 3.2 mm, speed 400rpm, federate .229 mm/rev.

Machine Tool Drive System:

In CNC machine tools, feed motor makes tool to move in relation to work piece. The motion situation depends on the current signal magnitude that feed motor needs. When drilling process is going on, tool wear happens and tool loses its sharpness. Therefore the cutting and thrust force necessary to process continuation would increase. To overcome the force necessary for cutting, motor needs higher current value to increase the power for continuing the machining process. Therefore there is a relationship between tool wear, motor current and cutting and also thrust force. Block diagram of feed drive system of drilling machine tool was simulated in MATLAB software and feed and spindle motor current changes, are obtained along with drilling process by using of this block diagram (fig.2).The block diagram of machine tool drives have been explained and developed by some researchers (Ebrahimi, M.whalley, R, 2000).

Experimental set-up:

The most commonly used drill is the conventional conical point drill and in this paper The experiments are done for this kind of drill only (fig.3). Drilling experiments were conducted on an OKUMA (MC-4VAE) CNC machining center a kistler (9273 A). Four-channel dynamometer was used to measure the thrust and torque forces. Material used is gray cast iron. Data has been sampled at 100Hz and stored in a pc. Table 2 shows the thrust force and torque values for a sharp drill used in this paper (V. Chandrasekharan, S.G. Kapoor and R.E.

2753

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

Devor, 1995). The force and motor current values will be obtained by wear model and machine tool drive block diagrams.

Artificial Neural Network:

Artificial neural network and fuzzy logic have been one of the reliable tools to predict tool wear rate in machining process. (Xiaoli, Li., Xinping, Guan. And Hongrui, Wang, 1999) used fuzzy logic for monitoring tool wear rate.Also some researchers used both of fuzzy and neural network for tool condition monitoring (Xiaoli, Li.,Yingxue, Yao. and Zhejun, Yuan,1997). Artificial Neural networks or parallel distributed processing is an alternative to sequential processing of knowledge as known from symbolic programming. In analogy to the human brain artificial neural networks consist of single units (neurons) that are interconnected by the so-called synapses. The typical network has layers of input and output units that are either connected directly or they are linked through one or several hidden layers. Each unit sends its numeric output modified by inhibitory or excitatory weights (also called transition coefficients) to another unit so that the output of a unit is the input to another or to many other units. Usually, one single unit multiplies the inputs by the individual weights and sums them up. The resulting value is the activation value of the unit which is often modified by applying an appropriate transformation function. The activation value is then preceded to other units by applying the appropriate weight. There are many useful transfer functions. Using of nonlinear transfer function makes a network capable of storing nonlinear relationships between input and output. Choice of the function depends on problem that will be solved. An important stage of a neural network is the training step, in which an input is introduced to the network together with the desired output and the weights are adjusted so that, the network attempts to produce the desired output (Adnan Sözen, Mehmet Özalp, Erol Arcaklio˘glu, 2004). The weights, after training, contain meaningful information whereas, before training they are random and have no meaning. If it has reached a satisfactory level, the train stops, and the network uses the weights to make decisions, to identify patterns, or to define associations in test data. Fig.4 shows the structure of an artificial neural network (ANN).

shows the structure of an artificial neural network (ANN). Fig. 2: Simulated feed motor current diagram

Fig. 2: Simulated feed motor current diagram obtained from machine tool drive block diagram model for drill diameter 15.9mm,point angle118,helix angle 33,pilot hole diameter 3.2 mm, speed 400rpm, federate . 229 mm/rev.

Table 1: experimental results of thrust force and torque (V.Chandrasekharan, S.G.Kapoor and R.E.Devor, 1995).

Feed

Speed

Diameter

Point

Web

Pilothole

Cutting lips

Total chisel

Entire drill

 

(mm/rev)

(rpm)

(mm)

angle

Thichness

Diameter

Thrust

Torqe

Thrust

torqe

Thrust

Torqe

Cuttingf

(mm)

(mm)

(N)

(nm)

(N)

(N)

(Nm)

orce(N)

.0.229

400

15.9

118

2.3

3.2

1091.5

13.52

1037

1.29

2228

14.8

1700

0.102

200

15.9

118

2.3

3.2

661.5

7.31

629

.548

1352

8.63

919

0.102

800

15.9

118

2.3

3.2

574.8

6.85

503

.505

1149

7.36

861

0.254

400

12.7

118

2.3

2.8

953.0

9.34

863

.919

1873

10.2

1470

0.102

400

9.5

118

1.5

2.4

373.5

2.7

393

.356

811

3.06

568

0.102

400

12.7

118

2.3

2.8

490

4.9

510

.7

1000

5.7

771

0.102

400

15.9

135

2.3

4.4

539.8

4.42

299

.468

868

6.86

555

0.178

400

9.5

118

1.5

2.4

525.4

8.11

743

.601

1376

8.72

1707

2754

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011 Fig. 3: conical point drill. Fig. 4:

Fig. 3: conical point drill.

Sci., 5(12): 2752-2760, 2011 Fig. 3: conical point drill. Fig. 4: structure of an artificial neural

Fig. 4: structure of an artificial neural network (ANN). (Adnan Sözen, Mehmet Özalp, Erol Arcaklio˘glu, 2004).

Modeling of Tool Wear Rate by Neural Network:

As mentioned above, there is relationship between cutting and thrust forces and drill wear rate .Also as the cutting and thrust force increases, motor current value increases too. For modeling this relationship, neural network is used in this paper. The transfer function used in this paper is Logistic sigmoid (logsig) transfer function:

f

(

z

)

1

1 e

z

(2)

Where z is the weighted sum of the input. PATIA software is used to model data. The data used in neural network obtained from tool wear and machine tool drive block diagrams. Data extracted from the simulation and used in this paper are given in table.2. Tool wear rate for sharp tool is 0 and for the worn tool it is supposed to be 0.8mm.Tool wear condition will be predicted between 0 and 0.8mm by the model. When the wear rate reaches to 0.4, the tool life reaches to half of its expected life. In neural network, there are nine inputs and one output, the output is drill wear rate and the inputs consists of machining parameters and force and motor current values. After giving data, the network is created. In the hidden layer of network, we created 12 neurons. The neurons within the hidden layer perform two tasks: they sum the weighted inputs connected to them and then pass the resulting summations through a non-linear activation function to the output neuron or adjacent neurons of the corresponding hidden layer (in case of more than one hidden neuron layer). There are 48 data to train the network. The data are given to ANN and the network is trained for the given data. The aim of ANN is to estimate interval values accurately. When the inputs are given to the ANN, the results are estimated. After a good training of ANN, the network is tested by some test data to assess the ANN accuracy.

RESULT AND DISCUSSION

Data are given to ANN and after training the predicted results are shown. Also, the error value for any run is obtained too. Square deviation shows the error and unfitness between neural network outputs and experimental results. R2 value of the data is obtained too. As it is given in the figure5, there is a very good fitness between neural network outputs and real data.R2 value is about 0.999 which shows the accuracy of the ANN. Tool wear results obtained from the ANN and the errors for any experiment are given in table3. As it’s seen in the table the maximum square deviation is about 0.000867 that shows the accuracy is very high. Also, variation diagram of the errors has been given in figure 6. It shows that the maximum square deviation value occurs in 45 th set and confirms the above given data. From the error diagram it is found that the error value between the wear rate of 0.2 and 0.5 is higher than others and in this part most of the data have a deviation from the zero line. In fig.7 the diagram of real wear data in relation to data number is given and in fig.8 the diagram of predicted drill wear data is created. From these two diagrams, the accuracy of ANN for prediction of the wear rate is obvious too. By attention to the diagram we find that the maximum unfitness happens in drill wear rate of 0.4 and 0.6 that confirms the results of square deviation diagram. For assessment of the accuracy of the designed neural network to predict drill wear, a test run compared with real value of drill wear. In this test the cutting force was 300N and the diagram is shown in figure5 on the right hand of the main diagram.

2755

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

Table 2: Data used for training ANN.

Data

Feed

Cutting

Drill

point

Webthicknes

pilot hole

Thrust

Feed

Cuttin

Feed

Flank

rate

speed

Diameter

angle

s (mm)

Diameter

force

motor

g force

motor

wear

(m/min)

(m/min)

(mm)

(Degree)

(mm)

(N)

Current

(N)

Current

(mm)

(v)

(v)

0.229

400

15.9

118

2.3

3.2

2228

24.8

1700

24.1

0

0.102

200

15.9

118

2.3

3.2

1352

23.6

919

23

0

0.102

800

15.9

118

2.3

3.2

1149

23.35

861

22.9

0

0.254

400

12.7

118

2.3

2.8

1873

24.3

1470

23.75

0

0.102

400

9.5

118

1.5

2.4

811

22.8

568

22.54

0

0.102

400

12.7

118

2.3

2.8

1000

23.1

771

22.8

0

0.102

400

15.9

135

2.3

4.4

868

22.9

555

22.52

0

0.178

400

9.5

118

1.5

2.4

1376

23.65

1707

24.1

0

0.229

400

15.9

118

2.3

3.2

3728

26.85

3200

26.12

0.2

0.102

200

15.9

118

2.3

3.2

2851

25.65

2419

25

0.2

0.102

800

15.9

118

2.3

3.2

2647

25.37

2360

24.9

0.2

0.254

400

12.7

118

2.3

2.8

3368

26.25

2969

25.81

0.2

0.102

400

9.5

118

1.5

2.4

2311

24.9

2067

24.57

0.2

0.102

400

12.7

118

2.3

2.8

2499

25.1

2270

24.85

0.2

0.102

400

15.9

135

2.3

4.4

2368

24.9

2055

24.56

0.2

0.178

400

9.5

118

1.5

2.4

2875

25.67

3201

26.12

0.2

0.229

400

15.9

118

2.3

3.2

5229

28.9

4700

28.17

0.4

0.102

200

15.9

118

2.3

3.2

4353

27.7

3918

27.1

0.4

0.102

800

15.9

118

2.3

3.2

4149

27.43

3860

27.02

0.4

0.254

400

12.7

118

2.3

2.8

4873

28.4

4470

27.86

0.4

0.102

400

9.5

118

1.5

2.4

3810

26.96

3567

26.62

0.4

0.102

400

12.7

118

2.3

2.8

4000

27.2

3770

26.9

0.4

0.102

400

15.9

135

2.3

4.4

3866

27.03

3554

26.61

0.4

0.178

400

9.5

118

1.5

2.4

4376

27.7

4706

28.1

0.4

0.229

400

15.9

118

2.3

3.2

5977

29.93

5450

29.2

0.5

0.102

200

15.9

118

2.3

3.2

5103

28.73

4669

28.1

0.5

0.102

800

15.9

118

2.3

3.2

4898

28.45

4610

28.05

0.5

0.254

400

12.7

118

2.3

2.8

5621

29.4

5218

28.8

0.5

0.102

400

9.5

118

1.5

2.4

4428

27.8

4318

27.6

0.5

0.102

400

12.7

118

2.3

2.8

4748

28.2

4520

27.9

0.5

0.102

400

15.9

135

2.3

4.4

4614

28.06

4305

27.6

0.5

0.178

400

9.5

118

1.5

2.4

5125

28.7

5454

29.2

0.5

0.229

400

15.9

118

2.3

3.2

6728

30.95

6200

30.2

0.6

0.102

200

15.9

118

2.3

3.2

5850

29.7

5420

29.1

0.6

0.102

800

15.9

118

2.3

3.2

5648

29.48

5359

29.07

0.6

0.254

400

12.7

118

2.3

2.8

6371

30.4

5968

29.9

0.6

0.102

400

9.5

118

1.5

2.4

5310

29.01

5068

28.6

0.6

0.102

400

12.7

118

2.3

2.8

5499

29.2

5270

28.9

0.6

0.102

400

15.9

135

2.3

4.4

5367

29.01

5050

28.6

0.6

0.178

400

9.5

118

1.5

2.4

5875

29.7

6206

30.23

0.6

0.229

400

15.9

118

2.3

3.2

8226

32.99

7699

31.6

0.8

0.102

200

15.9

118

2.3

3.2

7351

31.8

6920

31.2

0.8

0.102

800

15.9

118

2.3

3.2

7150

31.5

6861

31.9

0.8

0.254

400

12.7

118

2.3

2.8

7870

32.5

7466

31.9

0.8

0.102

400

9.5

118

1.5

2.4

6810

31.06

6568

30.7

0.8

0.102

400

12.7

118

2.3

2.8

6999

31.3

6774

31

0.8

0.102

400

15.9

135

2.3

4.4

6866

31.1

6696

30.98

0.8

0.178

400

9.5

118

1.5

2.4

7373

31.8

7706

32.28

0.8

Table 3: ANNs predicted data and errors

Data

Feed

Cutting

Drill

point angle

Web

pilot

hole

Thrust

Feed

Cutting

Feed

Flank

ANNs

Square

rate(m/

speed

Diameter

(Degree)

thickness

Diameter

force

motor

force

motor

wear

Predicted

deviation

min)

(m/min)

(mm)

(mm)

(mm)

(N)

Current

(N)

Current

(mm)

flank

(mm^2)

(v)

(v)

wear(mm)

0.229

400

15.9

118

2.3

3.2

2228

24.8

1700

24.1

0

0.000996

0.000002

0.102

200

15.9

118

2.3

3.2

1352

23.6

919

23

0

0.000808

0.000001

0.102

800

15.9

118

2.3

3.2

1149

23.35

861

22.9

0

0.003233

0.000016

0.254

400

12.7

118

2.3

2.8

1873

24.3

1470

23.75

0

0.001129

0.000002

0.102

400

9.5

118

1.5

2.4

811

22.8

568

22.54

0

0.001834

0.000005

0.102

400

12.7

118

2.3

2.8

1000

23.1

771

22.8

0

0.005168

0.000042

0.102

400

15.9

135

2.3

4.4

868

22.9

555

22.52

0

0.001405

0.000003

0.178

400

9.5

118

1.5

2.4

1376

23.65

1707

24.1

0

0.00398

0.000025

0.229

400

15.9

118

2.3

3.2

3728

26.85

3200

26.12

0.2

0.201998

0.000006

0.102

200

15.9

118

2.3

3.2

2851

25.65

2419

25

0.2

0.200827

0.000001

0.102

800

15.9

118

2.3

3.2

2647

25.37

2360

24.9

0.2

0.197461

0.00001

0.254

400

12.7

118

2.3

2.8

3368

26.25

2969

25.81

0.2

0.194176

0.000053

0.102

400

9.5

118

1.5

2.4

2311

24.9

2067

24.57

0.2

0.200397

0

0.102

400

12.7

118

2.3

2.8

2499

25.1

2270

24.85

0.2

0.197746

0.000008

2756

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

0.102

400

15.9

135

2.3

4.4

2368

24.9

2055

24.56

0.2

0.199299

0.000001

0.178

400

9.5

118

1.5

2.4

2875

25.67

3201

26.12

0.2

0.200933

0.000001

0.229

400

15.9

118

2.3

3.2

5229

28.9

4700

28.17

0.4

0.407462

0.000087

0.102

200

15.9

118

2.3

3.2

4353

27.7

3918

27.1

0.4

0.406456

0.000065

0.102

800

15.9

118

2.3

3.2

4149

27.43

3860

27.02

0.4

0.412872

0.000259

0.254

400

12.7

118

2.3

2.8

4873

28.4

4470

27.86

0.4

0.419513

0.000595

0.102

400

9.5

118

1.5

2.4

3810

26.96

3567

26.62

0.4

0.402399

0.000009

0.102

400

12.7

118

2.3

2.8

4000

27.2

3770

26.9

0.4

0.411971

0.000224

0.102

400

15.9

135

2.3

4.4

3866

27.03

3554

26.61

0.4

0.407914

0.000098

0.178

400

9.5

118

1.5

2.4

4376

27.7

4706

28.1

0.4

0.39311

0.000074

0.229

400

15.9

118

2.3

3.2

5977

29.93

5450

29.2

0.5

0.484785

0.000362

0.102

200

15.9

118

2.3

3.2

5103

28.73

4669

28.1

0.5

0.489416

0.000175

0.102

800

15.9

118

2.3

3.2

4898

28.45

4610

28.05

0.5

0.488414

0.00021

0.254

400

12.7

118

2.3

2.8

5621

29.4

5218

28.8

0.5

0.49032

0.000146

0.102

400

9.5

118

1.5

2.4

4428

27.8

4318

27.6

0.5

0.484988

0.000352

0.102

400

12.7

118

2.3

2.8

4748

28.2

4520

27.9

0.5

0.489971

0.000157

0.102

400

15.9

135

2.3

4.4

4614

28.06

4305

27.6

0.5

0.490368

0.000145

0.178

400

9.5

118

1.5

2.4

5125

28.7

5454

29.2

0.5

0.491737

0.000107

0.229

400

15.9

118

2.3

3.2

6728

30.95

6200

30.2

0.6

0.604951

0.000038

0.102

200

15.9

118

2.3

3.2

5850

29.7

5420

29.1

0.6

0.603796

0.000023

0.102

800

15.9

118

2.3

3.2

5648

29.48

5359

29.07

0.6

0.597945

0.000007

0.254

400

12.7

118

2.3

2.8

6371

30.4

5968

29.9

0.6

0.593409

0.000068

0.102

400

9.5

118

1.5

2.4

5310

29.01

5068

28.6

0.6

0.611409

0.000203

0.102

400

12.7

118

2.3

2.8

5499

29.2

5270

28.9

0.6

0.602067

0.000007

0.102

400

15.9

135

2.3

4.4

5367

29.01

5050

28.6

0.6

0.602713

0.000011

0.178

400

9.5

118

1.5

2.4

5875

29.7

6206

30.23

0.6

0.614925

0.000348

0.229

400

15.9

118

2.3

3.2

8226

32.99

7699

31.6

0.8

0.795966

0.000025

0.102

200

15.9

118

2.3

3.2

7351

31.8

6920

31.2

0.8

0.796361

0.000021

0.102

800

15.9

118

2.3

3.2

7150

31.5

6861

31.9

0.8

0.797059

0.000014

0.254

400

12.7

118

2.3

2.8

7870

32.5

7466

31.9

0.8

0.79517

0.000036

0.102

400

9.5

118

1.5

2.4

6810

31.06

6568

30.7

0.8

0.77644

0.000867

0.102

400

12.7

118

2.3

2.8

6999

31.3

6774

31

0.8

0.796463

0.00002

0.102

400

15.9

135

2.3

4.4

6866

31.1

6696

30.98

0.8

0.797495

0.00001

0.178

400

9.5

118

1.5

2.4

7373

31.8

7706

32.28

0.8

0.782508

0.000478

7373 31.8 7706 32.28 0.8 0.782508 0.000478 Fig. 5: Comparison of real wear and predicted drill

Fig. 5: Comparison of real wear and predicted drill wear rate by ANN.

2757

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011 Fig. 6: Square deviation of the data

Fig. 6: Square deviation of the data from the zero line.

2752-2760, 2011 Fig. 6: Square deviation of the data from the zero line. Fig. 7: Diagram

Fig. 7: Diagram of real drill wear data.

2758

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011 Fig. 8: Diagram of predicted drill wear

Fig. 8: Diagram of predicted drill wear data by ANN.

Conclusion:

In this paper artificial neural network was used to predict drill wear rate by using of cutting and thrust force and motor current value. We conclude that by measuring of cutting force and motor current values, estimating the drill wear rate is possible because there is a relationship between them. Also Artificial Neural Network is a reliable method to predict nonlinear incidents in machining process such as drill wear and etc. By using of ANN we can process and investigate many of the data that without using of this kind of tools accurately prediction of tool wear in machining will be impossible.

REFERENCES

Adnan Sözen, Mehmet Özalp, Erol Arcaklio˘glu, 2004." Investigation of thermodynamic properties of

refrigerant absorbent couples using artificial neural networks" Chemical Engineering and Processing, 43: 1253-

1264.

Carrillo, F.J. and M. Zadshakoyan, 1997." Adaptive observers for on-line tool wear estimation and monitoring in turning, using hybrid identification approach" ECC’97 European Control Conference. Bruxelles, Belgique, 1: 1-4. Chandrasekharan, V., S.G. Kapoor and R.E. Devor., 1995." A mechanistic approach to predicting the cutting forces in drilling:with application to fiber-reinforced composite materials" Journal of Engineering for Industry, l,117: 559-570. Ebrahimi,M.whalley, R., 2000. "Analyses, modeling and simulation of stiffness in machine tool drives" Computer & Industrial Engineering, 38: 93-105. Jantunen, Erkki., 2002. " A summary of methods applied to tool condition monitoring in drilling", International Journal of Machine Tools & Manufacture, 42: 997-1010. Xiaoli Li, Shiu Kit Tso, 2000." Real-Time Tool Condition Monitoring Using Wavelet Transforms and

Fuzzy Techniques, IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C:

APPLICATIONS AND REVIEWS, 30(3). Xiaoli, Li. and S.K. Tso, 1999. "Drill wear monitoring based on current signals", Elsevier Wear,

231:172-178.

Xiaoli, Li., Xinping, Guan. And Hongrui, Wang. 1999." Identification of tool wear states with fuzzy classification", International Journal of Computer Integrated Manufacturing, 12: 503-509.

2759

Aust. J. Basic & Appl. Sci., 5(12): 2752-2760, 2011

Xiaoli, Li.,Yingxue,Yao. and Zhejun, Yuan. 1997." On-line tool condition monitoring system with wavelet fuzzy neural network",Journal of Intelligent Manufacturing, 8: 271-276. Zhang, M.Z., Y.B. Liu and H. Zhou, 2001. " Wear mechanism maps of uncoated HSS tools drilling die- cast", Aluminum Alloy Tribology International, 34: 727-731.

2760