Introduction
Archived in
[pic]
Published in
The International Journal of Advanced Manufacturing Technology, Volume 28, Numbers 5-6 / March, 2006
S S Panda is presently with National Institute of Technology Rourkela
sspanda@nitrkl.ac.in
Drill wear prediction using artificial neural network
A.K. Singh1, S.S. Panda1, D. Chakraborty1 and S.K. Pal2
|(1) |Department of Mechanical Engineering, Indian Institute of Technology, Guwahati, 781039, Assam, India |
|(2) |Department of Mechanical Engineering, Indian Institute of Technology, Kharagpur, 721302, INB, India |
Abstract The present work deals with drill wear monitoring using artificial neural network. A back propagation neural network (BPNN) has been used to predict the flank wear of high speed steel (HSS) drill bit for drilling holes on copper work-piece. Experiments have been carried out over a wide range of cutting conditions and the effect of various process parameter like feeed-rate, spindle speed, drill diameter on thrust force and torque has been studied. The data thus obtained from the experiments have been used to train a BPNN for wear prediction. The performance of the trained neural network has been tested with the experimental data and found to be satisfactory.
Keywords Flank wear[pic] Artificial neural network[pic]Drilling
1 Introduction
Drilling is one of the important machining operations extensively used in manufacturing industries. Tool wear has significant influence on the performance of machining operation. During machining tool wear affects the tool life and surface finish of the machine component. In case of drilling, wear is categorized as flank wear, chisel wear, corner wear, and crater wear. Wear on the drill has a definitive influence on the hole quality and tool life of drill bit. Therefore online monitoring of drill wear is a very important issue in manufacturing industries and thus an emergent area of research. Many works have been reported in the broad field of tool condition monitoring.
A. K. Singh [pic] S. S. Panda [pic] S. K. Pal [pic] D. Chakraborty ([pic])
Department of Mechanical Engineering,
Indian Institute of Technology, Guwahati,
781039, Assam, India
E-mail: chakra@iitg.ernet.in
Noori-Khajavi and Komanduri [1] developed a model for online tool wear monitoring of drilling operation and observed that only one signal is sufficient to monitor the tool wear. Lin and Ting [2] used the force signal to monitor online drill wear. They used the least square method for determining the thrust force and torque as a function of spindle speed, feed-rate, drill diameter and average flank wear. Lin and Ting [3], in another work used back propagation neural network with sample and batch mode, and observed faster convergence of error in the case of sample mode. They also observed that neural network with two hidden layers with same number of nodes converge faster than that with one hidden layer and reported that at higher learning rate error produced is less. Das et al. [4] used back propagation algorithm for measuring flank wear of carbide tool in turning operation. Lee et al. [5] used the abductive network modeling for drilling process for predicting the tool life, tool wear and surface roughness. The network has number of polynomial functional nodes. Optimal network architecture is prepared based on predicted square error criterion. Choudhury et al. [6] developed a three-layer feed forward back propagation neural network for predicting the flank wear in turning operation. He used the geometrical relation in correlating the flank wear on cutting tool with change in work-piece dimension. Kosmol et al. [7] used the finite element method in calculating the wear of drill point, roughness and erroneous surface of bore hole. Liu et al. [8] used the algorithm for synthesis of polynomial network for predicting (ASPNS) the corner wear in drilling operation. Li and Tso [9] used the regression model for monitoring the tool wear based on current signal of spindle motor and feed motor. Choudhury and Raju [10] developed a regression model to measure the flank wear and corner wear of a drill bit in cutting operation. Tsao [11] used the radial basis function network (RBFN) and adaptive based radial basis function network (ARBFN) to predict the flank wear in both the cases and compared their result with experimentally obtained value. Davim and Antonio [12] used the evolution strategy for identifying the type of wear in polycrystalline diamond (PCD) drill bit with metal matrix composite as work-piece. They used the Pareto optimal solution in the genetic algorithm for maximization of tool life and minimization of drill wear. Ertunc and Loparo [13] used decisions fusion center algorithm (DFCA) for monitoring online tool wear condition in drilling process and used the number of numerical methods for predicting the condition of tool wear land. Kim et al. [14] used the William drill model [21] for predicting and validating the progressive drill wear based on spindle motor power consumption. Nouari et al. [15] used the third wave advantedge software for predicting the tool chip interface temperature, which is major factor of drill wear formation in the dry condition. Chien and Tsai [16] used the back propagation neural network for prediction of tool wear and determining the optimum cutting condition in turning operation, They used the genetic algorithm in the optimizing model as well as Taguchi method to find the optimum parameter for both the model. Abbu [17] used the vibration signature analysis for predicting the wear rate in drilling. He estimated three different patterns of vibration signature like harmonic wavelets coefficient, power spectra density and First Fourier Transformation (FFT). All these are inputs to the neural network model. Wong and Hamouda [18] developed the neural network model for representing the machinability data. They used a new type of neuron that is product neuron, which has characteristic of multiplication instead of summation. Ertunc and Oysu [19] used hidden markov model for monitoring drill wear using cutting force signal. Kim and Ramulu [20] used multiple objective linear programming models for optimizing drill hole quality with different cutting condition such as speed and feed-rate.
2 Back propagation neural network
Back propagation neural network (BPNN) is used in the present work. A generalized delta rule algorithm is used in this proposed model. Basic structure of back propagation neural network having input, hidden and output layers, is shown in Fig. 1. Input layer receives information from the external sources, and passes this information to the network for processing. Hidden layer receives information from the input layer, and does all the information processing, and output layer receives processed information from the network, and sends the results out to an external receptor. The number of hidden layer and the number of node in a hidden layer is a variable quantity, which depends upon the convergence criteria of results.
The input signals are modified by interconnection weight, known as weight factor Vij, which represents the interconnection of ith node of the first layer to jth node of the second layer. The calculation performed in a particular node is shown is Fig. 2. The sum of modified signals (total activation) is then modified by a sigmoidal transfer function.
Batch mode type supervised learning has been used in the present case, where, all input-output pattern sets are presented to the neural network one by one, and then adjusted using average gradient information. Generalized delta rule algorithm is used. During training the calculated output is compared with the desired output and the mean square error is calculated. If mean square error is more than a prescribed limiting value error it is back propagated i.e., from output to input, weights are further modified till the error is within a prescribed limit.
Mean square error[pic], is calculated by the formula,
[pic] [pic]
where,
[pic]Number of pattern
[pic]Number of node in output layer
[pic]=Desired output of kth node of pth pattern
[pic]=Calculated out put of kth node of pth pattern
3 Experimental set-up
Large number of experiments over a wide range of cutting conditions has been performed. Among all types of existing drill wear, flank wear is the predominant and hence it has been considered in this present work. Fig. 3 shows a schematic representation of the experimental set up used in present work.
Radial drilling machine (Batliboi Limited, BR618 model) is used for the drilling operation. HSS drill bits with different diameters have been used for drilling in copper work-piece at different cutting condition.
Thrust force and torque are recorded through a piezo-electric Kistler 9272 dynamometer. Signal from the dynamometer is amplified through charge amplifier, and is stored in the computer through data acquisition system. Charge amplifiers of B&K 2525 model have been used in this work. Advantech PCL 818 HG model data acquisition system is used in present work.
Flank wear is measured by digital microscope with the help of Karl-Zeiss software interfacing. The maximum flank wear is used as the criterion to characterize the drill condition and is obtained by measuring the wear at different points on either of the cutting edge.
Photographs of gradual wear build-up process for three different feed-rates are shown in Fig. 4(a)-4(c).
Fig. 3 Schematic diagram of experimental set-up
[pic]
Fig. 4(a) Flank wear at diameter 10 mm, spindle speed 500 rpm and feed-rate 0.13 mm/rpm.
[pic]
Fig. 4(b) Flank wear at diameter 10 mm, spindle speed 500 rpm and feed-rate 0.18 mm/rpm.
[pic]
Fig. 4(c) Flank wear at diameter 10 mm, spindle speed 500 rpm and feed-rate 0.25 mm/rpm.
4 Results and discussion
Drilling operation has been conducted over a wide a range of cutting condition. Spindle speed has been varied in the range 315 rpm to 1000 rpm in six steps. Feed-rate has been varied from 0.13 to 0.71 mm/rpm in six steps. HSS drill bit of three different diameters have been used for drilling hole in a copper plate. Various combination of spindle speed, feed-rate and drill diameter has been used to perform 49 different drilling operations. For each of these condition, thrust force and torque have been measured using the data acquisition system, and is stored in the computer. Also corresponding to each cutting condition, maximum flank wear has been measured using digital microscope with interface of Karl-Zeiss software. The results of the experiment are tabulated in Table 1.
Table 1 Experimental data
|Serial number |Drill diameter (mm) |Spindle speed |Feed rate |Thrust |Torque |Maximum wear (mm) |
| | |(rpm) |(mm/rpm) |force |(N-m) | |
| | | | |(N) | | |
|1 |10 |500 |0.13 |1925 |19.253 |0.11 |
|2 |7.5 |500 |0.13 |510 |7.1 |0.06 |
|3 |5 |500 |0.13 |245 |2.5 |0.03 |
|4 |10 |500 |0.18 |3860 |25.1 |0.195 |
|5 |7.5 |500 |0.18 |595 |4.41 |0.08 |
|6 |5 |500 |0.18 |275 |2.75 |0.06 |
|7 |10 |500 |0.25 |3740 |27.44 |0.21 |
|8 |7.5 |500 |0.25 |539 |5.39 |0.105 |
|9 |5 |500 |0.25 |386 |2.9 |0.095 |
|10 |10 |400 |0.13 |2518 |23.52 |0.185 |
|11 |7.5 |400 |0.13 |853 |11.27 |0.085 |
|12 |5 |400 |0.13 |267 |3.1 |0.05 |
|13 |10 |400 |0.18 |3921 |26.78 |0.2 |
|14 |7.5 |400 |0.18 |646 |12.64 |0.1 |
|15 |5 |400 |0.18 |451 |3.96 |0.085 |
|16 |10 |400 |0.25 |4010 |29.25 |0.26 |
|17 |7.5 |400 |0.25 |1051 |16.54 |0.1 |
|18 |5 |400 |0.25 |505 |1.96 |0.07 |
|19 |10 |630 |0.13 |1258 |10.11 |0.12 |
|20 |7.5 |630 |0.13 |488 |5.86 |0.09 |
|21 |5 |630 |0.13 |186 |2.94 |0.08 |
|22 |10 |630 |0.18 |1470 |13.23 |0.125 |
|23 |7.5 |630 |0.18 |524 |3.95 |0.1 |
|24 |5 |630 |0.18 |187 |2.64 |0.07 |
|25 |10 |630 |0.25 |3077 |15.68 |0.18 |
|26 |7.5 |630 |0.25 |441 |4.41 |0.1 |
|27 |5 |630 |0.25 |285 |2.15 |0.025 |
|28 |10 |800 |0.36 |2234 |22.34 |0.16 |
|29 |7.5 |800 |0.36 |1666 |8.66 |0.13 |
|30 |10 |800 |0.5 |2548 |24.1 |0.19 |
|31 |7.5 |800 |0.5 |1440 |19.3 |0.13 |
|32 |5 |800 |0.5 |1087 |11.27 |0.09 |
|33 |10 |315 |0.36 |3303 |25.33 |0.20 |
|34 |7.5 |315 |0.36 |1866 |12.74 |0.16 |
|35 |5 |315 |0.36 |592 |7.72 |0.1 |
|36 |10 |315 |0.5 |3413 |29.54 |0.205 |
|37 |7.5 |315 |0.5 |1688 |15.19 |0.14 |
|38 |5 |315 |0.5 |1210 |13.39 |0.08 |
|39 |10 |315 |0.71 |3920 |36.22 |0.24 |
|40 |7.5 |315 |0.71 |1828 |17.15 |0.12 |
|41 |5 |315 |0.71 |1282 |16.66 |0.1 |
|42 |10 |1000 |0.36 |1460 |12.25 |0.14 |
|43 |7.5 |1000 |0.36 |554 |5.39 |0.095 |
|44 |5 |1000 |0.36 |421 |4.21 |0.06 |
|45 |10 |1000 |0.5 |1960 |18.13 |0.13 |
|46 |7.5 |1000 |0.5 |784 |7.35 |0.1 |
|47 |5 |1000 |0.5 |651 |6.17 |0.07 |
|48 |10 |1000 |0.71 |2009 |20.58 |0.17 |
|49 |7.5 |1000 |0.71 |970 |8.05 |0.12 |
4.1 Effect of important parameters on thrust force and torque
Figs. 5-10 shows the effect of important cutting parameters on thrust force and torque during drilling operation. From Figs. 5-6 and Figs. 8-9 it could be observed that thrust force and torque increase as drill diameter and feed rate increase. This is due to increase of the un-deformed chip thickness, which is known as size effect [2]. It has also been observed that drill diameter has more effect on thrust force and torque than that of feed-rate, and as a result both thrust force and torque increase sharply beyond 7.5 mm drill diameter. It may be due to increase of thickness of un-deformed chip with increase of drill diameter than that of feed rate. Increase of drill diameter along the circumferential direction imparts more chip load than that of increase in feed rate along the axial direction.
Fig. 7 and Fig. 10 show that thrust force and torque decrease with increasing spindle speed. This is due to high temperature generation at the tool chip interface, and thus the strength of the work material reduces [2].
[pic]
Fig. 5 Comparison of average thrust with drill diameter at different feed-rate and constant speed
[pic]
Fig. 6 Comparison of average thrust force with drill diameter at different speeds and constant feed-rate
[pic]
Fig. 7 Comparison of average thrust force with spindle speed at different drill diameters and constant feed-rate
[pic]
Fig. 8 Comparison of average torque with drill diameter at different feed-rates and constant speed
[pic]
Fig. 9 Comparison of average torque with drill diameter at different speeds and constant feed-rate
[pic]
Fig. 10 Comparison of average torque with spindle speed at different drill diameters and constant feed-rate
4.2 Wear prediction by neural network
Back propagation neural network algorithm has been used in the present work. To train the neural network thrust force, torque, feed-rate, drill diameter and spindle speed are used as input parameters and corresponding maximum flank wear has been used as the output parameter. From the 49 data sets obtained from the experiment, 34 have been selected at random for training the network and remaining 15 are used for testing. The normalized data sets are used for training the network. The data sets are normalized in the range of 0.1 to 0.9 by using the equation (2).
[pic] [pic]
where,
[pic]= Actual value
[pic][pic]Maximum value of [pic]
[pic]=Minimum value of [pic]
[pic]=Normalized value corresponding to [pic] [pic]
The number of hidden layer, number of nodes in the hidden layer, learning rate [pic] and momentum coefficient [pic] are decided by trial and error. Large number of neural network architecture has been tried and based on the convergence rate of mean square error shown in Table 2, the optimal network for the present case has been selected as 5-4-1 with momentum coefficient [pic] is equal to 0.3 and learning rate [pic] is equal to 0.3. Fig. 11 shows the mean square error for training and testing of the data for the neural network (5-4-1 with [pic]=0.3 and[pic]=0.3). The wear predicted by the neural network compared with the actual values is shown in Fig. 12. It has been observed that wear predicted is within (7.5 percent of the experimental value.
Table 2 Training and testing error for different neural network architectures
|Serial |Neural |Learning |Momentum Coefficient[pic]|Mean Square |Number of |Maximum |Minimum Predicted |
|Number |Network |Rate[pic] | |Error |Iteration |Predicted Error |Error (%) |
| | | | | | |(%) | |
|1 |5-3-1 |0.7 |0.8 |0.11 |1119 |28.8 |4.09 |
|2 |5-3-1 |0.8 |0.9 |0.11 |4903 |39.4 |2.05 |
|3 |5-3-1 |0.5 |0.6 |0.11 |4141 |24.2 |4.5 |
|4 |5-3-1 |0.6 |0.4 |0.00937 |4639 |25.4 |0.34 |
|5 |5-3-1 |0.3 |0.3 |0.00857 |16034 |31.5 |4.1 |
|6 |5-4-1 |0.7 |0.8 |0.0079 |234 |39.2 |0.778 |
|7 |5-4-1 |0.8 |0.9 |0.0075 |1251 |48.1 |0.74 |
|8 |5-4-1 |0.5 |0.6 |0..075 |448 |31.2 |0.64 |
|9 |5-4-1 |0.6 |0.4 |0.00799 |220 |31.2 |0.55 |
|10 |5-4-1 |0.3 |0.3 |0.006 |875 |28.1 |1.1 |
|11 |5-4-1 |0.7 |0.8 |0.00998 |3081 |25.4 |4.1 |
|12 |5-4-1 |0.8 |0.9 |0.0112 |1545 |27.6 |2.3 |
|13 |5-4-1 |0.5 |0.6 |0.1 |573 |35.1 |1.01 |
|14 |5-4-1 |0.6 |0.4 |0.008 |534 |33.2 |0.8 |
|15 |5-4-1 |0.3 |0.3 |0.006 |1790 |27.1 |1.17 |
[pic]
Fig. 11 Convergence of mean square error with number of iterations
[pic]
Fig. 12 Comparison of predicted value and experimental value of flank wear.
5 Conclusion
Back propagation neural network based drill wear prediction methodology has been adopted using various important parameters like thrust force, torque, drill diameter, spindle speed and feed-rate influencing the drill wear. Effect of various parameters on the thrust force and torque has been studied which agrees well with the earlier published results [2]. It has been observed that neural network could learn well the pattern and could be used for future prediction of drill wear.
References
1. Noori-Khajavi A, Komanduri R (1995) Frequency and time domain analyses of sensor signal in drilling-II, Investigation on some problems associated with sensor signal. International journal of machine tool manufacturer 35: 795-815
2. Lin SC, Ting CJ (1995) Tool wear monitoring in drilling using force signals. International journal of machine tool manufacturer 180: 53-60
3. Lin SC, Ting CJ (1996) Drill wear monitoring using neural network. International journal of machine tool manufacturer 36: 465-475
4. Das S, Roy R, Chottopadhyay AB (1996) Evaluation of wear of turning carbide inserts using neural network. International journal of machine tool manufacturer 36:789-797
5. Lee BY, Liu H S, Tarng YS (1998) Modelling and optimization of drilling process. Journal of material processing technology 74: 149-157
6. Choudhury SK, Jain VK, Rama Rao ChVV (1999) On-line monitoring of tool wear in turning using neural network. International journal of machine tool and manufacture 39: 489-504
7. Kosmol J, Czech M, Klarecki K, Sliwka J (1999) The optimization of drills for machining of austenitic steel. Journal of material processing technology 80-90: 117-122
8. Liu HS, Lee BY, Tarng YS (2000) In-process prediction of corner wear in drilling operations. Journal of material processing technology 101: 152-158
9. Li Xiaoli, Tso S.K (1999) Drill wear monitoring based on current signals. Wear 231: 172-178
10. Choudhury SK, Raju G (2000) Investigation into crater wear in drilling. International journal of machine tool manufacturer 40: 887-898
11. Tsao CC (2002) Prediction of flank wear of different coated drills for JIS SUS 304 stainless steel using neural network. Journal of material processing technology 123: 354-360
12. Davim JP, Antonio CAC (2001) Optimal drilling particulate metal matrix composites based on experimental and numerical procedures. International journal of machine tool and manufacture 41:21-31
13. Ertunc HM, Loparo KA (2001) A decesion fusion algorithm for tool wear condition monitoring in drilling. International journal of machine tool manufacturer 41: 1347-1362
14. Kim HY, Ahnn JH, Kim SH, Takata S (2000) Real-time drill wear estimation based on spindle motor power. Journal of material processing technology 124: 267-273
15. Nouari M, List G, Girot F, Coupard D (2003) Experimental analysis and optimization of tool wear in dry machining of aluminium alloys. Wear 255:1359-1368
16. Chien W T, Tsai C S (2003) The investigation on prediction of tool wear and the determination of optimum cutting conditions in machining17-4 PH stainless steel. Journal of material processing and technology 140: 340-345
17. Abbu-Mahfouz I (2003) Drilling wear detection and classification using vibration signals and artificial neural network. International journal of machine tool manufacturer 43: 707-720
18. Wong SV, Hamouda AMS (2003) Machinability data represention with artificial neural network. Journal of material processing technology 138: 538-544
19. Ertunc HM, Oysu C (2003) Drill wear monitoring using cutting force signal. Mechatronics
20. Kim D, Ramulu M (2004) Drilling process optimization for graphite/ bismaleimide-titanium alloy stack. Composite structures 63: 101-114
21. Williams RA (1974) A study of drilling process. Journal of engineering for industry 96: 1207-1215
Fig. 1 Three-layer neural network
Fig. 2 Calculation of a particular node
-----------------------
Node
W2j
[pic]
[pic]
an
a1
Wnj
Work-piece
Job fixture
Digital microscope (drill wear)
bj
Tj
W1j
Dynamometer
Drill bit
Spindle
Drill wear monitoring
Neural network model
f(xj)
Computer
Data acquisition system
Charge amplifier
Radial drilling machine
Output layer
Hidden layer
Input layer
C1
V11
V12
V13
V51
V52
V53
B1
B2
B3
W31
W21
W11
TCi
TBi
TAi
A1
A2
A3
A4
A5
I5
I4
I3
I2
I1
................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.
Related searches
- introduction to financial management pdf
- letter of introduction sample
- argumentative essay introduction examples
- how to start an essay introduction examples
- introduction to finance
- introduction to philosophy textbook
- introduction to philosophy pdf download
- introduction to philosophy ebook
- introduction to marketing student notes
- introduction to marketing notes
- introduction to information systems pdf
- introduction paragraph examples for essays