COMPARATIVE STUDY OF TOOL WEAR ESTIMATION IN DRILLING

 

SHRIHARSHA B.R. and H.V. RAVINDRA

Dept. of Mech. Engg., P.E.S.C.E., Mandya – 571 401.  

 

 

ABSTRACT:

 

Tool wear monitoring and estimation are essential for improved productivity of manufacturing systems. In order to study the tool wear phenomenon, Thrust, Torque, Tool - Tip Temperature and Vibration signals were measured during drilling and an attempt was made to obtain a clear insight of the signals involved. Simple functional relations between the parameters have been plotted to derive a basis for more detailed analysis and arrive at possible information on signal-wear relationship. Further to avoid the random variations in material properties more sophisticated method of signal analysis methods like Multiple Regression Analysis and Group Method of Data Handling (GMDH) were used for tool wear estimation. Further, the comparison studies of Multiple Regression Analysis and Group Method of Data Handling were made. It was found that at higher speeds and feed, the multiple regression analysis seems to be the best.

 

 

INTRODUCTION:

 

     In any manufacturing process, the production of a part with specific shape and dimensional accuracy will be intended without damaging the tool. But tool wear is a predominant phenomenon to be predicted, detected for efficient production. Tool wear sensing contributes greatly towards optimization of the cutting process, efficient tool change policies, improved product quality control and lower tool costs. Hence, sensor based monitoring systems have become increasingly useful in improving the efficiency of the manufacturing systems [1]. 

     Several methods have been investigated for monitoring tool wear with signals such as cutting force, power, tool tip temperature, vibration and acoustic emission [2]. Most of these techniques have been applied quite successfully to a certain extent, but with some limitations [3]. The method used for calculations of thrust and torque in drilling has shown that they depend on all the drilling parameters. Also to validate the theoretical model, a range of experimental tests has been undertaken and then compared with the theoretical values of thrust and torque. The comparison shows good agreement between predictions and experimental results [4, 5]. Also, for predicting thrust, torque, radial forces in drilling, an analytical model was developed that includes the effect of the drill bit transverse deflections, which lead to variations from the mean values in the cutting forces. This model when drill vibrations are negligible used for predicting the mean torque and thrust in drilling, as a result the predicted values are in good agreement with the experimental data [6].

     For drill bit transverse vibration, a dynamic model was developed. The model provides good qualitative agreement with experimental data for the effects of design and process parameters on drill vibrations and stability [7].

     Drill bit is a complex tool and it is impossible without using multiple sensors to monitor it. Therefore by using multiple sensor systems for tool condition monitoring in drilling, the productivity can be improved [8, 9].

     The present work aims at devising multi-sensory schemes for tool wear estimation in drilling process, by measurement of thrust, torque, tool tip temperature and vibration. Further to avoid the random variations in material properties more sophisticated method of signal analysis like Multiple Regression Analysis and Group Method of Data Handling (GMDH) were used for tool wear estimation [10]. These methods have been explored for their capability to integrate information from different sensors.

 

 

MULTIPLE REGRESSION ANALYSIS:

 

     The objective of multiple regression analysis is to construct a model to explain, as much as possible, the variability in a dependent variable, using several independent variables.  The model used is usually a linear model, though some times non liner models are also constructed. The least squared estimates are the best linear unbiased estimates of the dependent variable [11].

 

 

GROUP METHOD OF DATA HANDLING (GMDH):

 

     Among the widely used methods for empirical analysis of data and model building multiple regression analysis is well known technique. One of the major problems associated with use of regression has been the need to specify functional formulation.  The linear assumption is not valid in all cases and also an infinite variety of non-linear functional forms exist.  This might not cause many difficulties when model building is used to determine causal relationships.  The problem assumes significance where a dependent variable is to be estimated from measured variables.  In such cases, while it is known that some of the measured variables are to be used, the nature of relationship and relative importance of these variables is unknown.  It would be preferable in such cases to use the data to determine both the nature of function and parameters of the function.  This is the motivation for the development of self-organizing methods in modeling; GMDH is one such method [12, 13].

     The approach is to fit a high degree polynomial using a multilayered network like structure. Each element in the network is a partial polynomial (a quadratic function) of two inputs. The coefficients of the quadratic function are determined by data from training set (certain percentage of data-set is taken as training set to learn the model; the remaining data-set is used for checking the model). All possible combinations of inputs, taken two at a time, are evaluated. The combinations that are allowed to pass to the next year and self organizing is terminated when optimum complexity is reached by evaluation of a criterion function from data in the checking set.

     Three different criterion functions, viz., regularity, unbiased and combined criterions were attempted. Regularity criterion has good predictive power but sensitive to noise. Unbiased criterion selects models that are insensitive to data from which it is built and hence gives good noise immunity but may not have good predictive power. Combined criterion is the combination of regularity and unbiased criterions [14]. GMDH has been applied to the modeling and optimization of machining processes. Also it was observed that among multiple regression analysis, best results could be obtained from GMDH and neural networks [15].

     In the present paper, applications of multiple regression and GMDH to tool wear estimation is discussed and compared.  Various heuristics of GMDH appropriate for the present work are detailed in Appendix.

 

 

EXPERIMENTAL WORK:

 

     The experimental work consisted of drilling Mild Steel plate using High Speed Steel Drill Bit. The drilling was carried out in Automatic Drilling Machine. The parameters like thrust, torque, tool tip temperature and vibration were measured for different operating conditions. Digital Drill Tool Dynamometer is used to measure both the thrust and torque of the drill bit. The temperature readings were obtained by using Infrared Thermometer (WAHL HEAT SPY). The vibration readings were recorded using Oscilloscope (100 MHz).

     The experiments were performed to obtain progressive Flank wear. These experiments were conducted with different speed and feed combinations. Drilling was stopped at regular intervals and the width of the flank wear was measured using Toolmakers Microscope.

     The composition of work material is given in the Table 1. The tool material specifications are given in Table 2. The cutting conditions used during the experiments are given in Table 3.

 

 

Table 1: WORK MATERIAL COMPOSITION

 

Work Material

C-60 Steel

Hardness

BHN 225

Composition in Percentage

 

C-0.55 to 0.65

Mn-0.50 to 0.80

 

 

 

 

 

 

 

 

 

 

Table 2: TOOL MATERIAL SPECIFICATION

 

Tool material

High speed steel

Diameter of the drill bit used

10mm.

Chisel edge angle

120º to 135º

Helix angle or rake angle

30º

Point angle

118º

Lip clearance angle

12º

 

 

 

 

 

 

 

 

 

 

 

 

Table 3: EXPERIMENTAL CONDITIONS

 

a) Effect of thrust, torque, temperature on flank wear

Speed (m/min)

9, 18, 22

Feed  (mm/rev)

0.190

b) Effect of thrust, torque, temperature on flank wear

Feed  (mm/rev)

0.095, 0.190, 0.285

Speed (m/min)

15

c) Effect of thrust, torque, temperature and vibration on flank wear

Feed (mm/rev)

0.190

Speed (m/min)

18, 22

 

 

 

 

 

 

 

 

 

 

 

 

 

 

RESULTS AND DISCUSSIONS:

 

     Initially, Simple functional relations between the parameters have been plotted to derive a basis for more detailed analysis and arrive at possible information on signal-wear relationship.

 

Effect of Speed and Feed on Flank Wear

 

     Fig. 1 gives the  measured flank wear at three different speeds (15 m/min, 18 m/min, and 22 m/min) and at a constant feed of 0.190 mm/rev and Fig. 2 gives the measured wear at three different feeds (0.095 mm/rev, 0.190 mm/rev. and 0.285 mm/rev.) and at a constant speed of 15 m/min. It can be observed that the wear curves have clearly defines the regions of running in, steady state and rapid wear curves. Hence, this implies that the drilling is done at near optimal regions. Also, it was observed that at higher speeds and feeds the tool wear is maximum.

 

 

 

Fig. 1 Flank Wear v/s Machining Time

 

 

 

 

 

Fig. 2 Flank Wear v/s Machining time

 

Effect of Thrust, Temperature, Vibration on Flank Wear

 

     Fig. 3 gives the plot of thrust, torque, temperature and maximum measured wear for different duration of drilling at 9 m/min cutting speed and 0.190 mm/rev feed. There is an increase in thrust and temperature with progressive increase in flank wear. But there is a negligible increase in torque. Therefore it is neglected because of least variations. These effects were also same for the data’s taken at higher speeds and constant feed. Also, similar effects were observed for increasing feed and constant speed.

     Fig. 4 shows the vibration signals [Vavg, Vrms, Vp (in voltages) along with thrust, temperature and measured wear for 22 m/min cutting speed and 0.190 mm/rev feed.  It was observed that the flank wear increases with the increase in vibration signals such as VP, thrust and temperature. But VAVG and VRMS become constant throughout and therefore they are neglected. These features are also same for the data’s taken at higher speeds.

     By observing the above signals it was found that, there was random increase of thrust, vibration and tool wear with the increase in speed and feed. This can be attributed to random variations in material properties. Under such conditions, any system based on simple threshold criterion to estimate the end of tool life can produce wrong diagnosis. Thus, there is a requirement for more sophisticated methods of signal analysis like Multiple Regression Analysis and Group Method of Data Handling (GMDH). These methods are robust for random variations in the variables and is capable of integrating


 

Fig. 3 Flank wear, thrust, torque and temperature with machining time

 

Fig. 4 Flank wear, thrust, torque, temperature and vibration with machining time 


information such as, measured drilling parameters (thrust, torque, temperature and vibration), with feed and speed.

 

 

TOOL WEAR ESTIMATION

 

     For effective and reliable monitoring and thereby estimating tool wear, the signals due to thrust, torque, temperature and vibration of the tool can be supplemented. With the introduction of a variety of data, MULTIPLE REGRESSION ANALYSIS and GMDH become more appropriate for estimation of tool wear.

     The two methods of specifying input data were used. This consisted of considering flank wear as a function of the following combinations.

(a)    Temperature, Thrust, Torque.

(b)    Temperature, Thrust and vibration (VP).

 

MULTIPLE REGRESSION ESTIMATES

 

     >From Fig. 5, the estimates from the regression analysis at 22 m/min cutting speed and 0.190 mm/rev. feed were plotted. It was observed that the measured wear was well correlated with the predicted wear. The same effects were observed for other cutting conditions. These effects were observed for the first set input combinations. Also, the same effects were observed for second set input combination as shown in Fig. 6.

 

 

Fig. 5 Multiple Regression Estimates

 


Fig. 6 Multiple Regression Estimates

 

 


GMDH ESTIMATES

    

     Here three criterions namely, regularity, combined and unbiased criterion were used for guiding the self-organization procedure. Fig. 7 gives the plot of estimates for 9 m/min cutting speed and 0.190 mm/rev. feed for the flank wear. The estimates from regularity, combined and unbiased criteria were plotted. It was observed that the estimates of regularity and combined criteria are correlating with the measured wear. Due to poor correlation, the unbiased criterion has not been considered. The same effects were also observed for varying cutting speeds and constant feed. These effects were observed for the first set input combinations. Also, the same effects were observed for second set input combination as shown in Fig. 8.

     By observing the plots, it was concluded that the unbiased criterion does not have good predictive power and usually tends to wrongly estimate the variations in the dependant variable. The Regularity criterion, which has better predictive ability, works well in the absence of noise.  It has seen that GMDH precisely estimate the tool wear by presenting a closed representation with the observation especially at high drilling conditions i.e., with higher drilling conditions, the tool exhibited rapid wear in relatively shorter duration of drilling, maintaining the correlation between the tool wear and the drilling parameters. Also it was concluded that the tool wear estimates are required only after the tool has passed through the running in stage of wear.

     Fig. 9 gives the plot of estimates from regularity and combined criteria for different drilling conditions. Also the regression estimates those are correlating well are considered


 

 

Fig. 7 GMDH ESTIMATES

 

 

Fig. 8 GMDH ESTIMATES

for the comparison purpose. It was observed that the estimates from regression analysis were well correlated compared to GMDH estimates. At higher speeds and feed the multiple regression seems to be the best. But considering other variables such as work material properties, tool material, GMDH with regularity criterion may be better. Good correlation was obtained for all by using regularity criterion.

 

                        Fig. 9. Comparison of Regression and GMDH Estimates

 

 

CONCLUSION

 

·         Experimental work was designed to use multiple sensors and to obtain data for sharp tool and for different stages of flank wear.

·         Experimental data was used to establish the effect of drilling conditions on drilling parameters, and tool wear.

·         The multiple regression analysis and group method of data handling has been attempted for tool wear estimation using two types of input variable combinations.  This has resulted in successful implementation of tool wear estimation.

·         For GMDH 3 different criterion functions have been used. Based on the present work, it can be concluded that the regularity criterion functions well for input variables like steady state drilling parameters like Thrust, Torque, Temperature and vibration.

·         Multi sensory approaches have been proposed and implemented.  The ability of above-mentioned algorithms to effect integration of sensor information has been established. 

APPENDIX

 

Heuristics Used

     Several heuristics used to guide the self-organization are described below.

    The order of data: In the Present work, data with the largest variance is put in the training set.  The variance for ith data point is given by

 

        m

Di2 = å  (Xij – Xj)2/sj2                                      (1)

        I = 1

Where Xj = means for jth variable and

 

                 n

sj2 = (1/n) å   (Xij – Xj )2                               (2)

                i = 1

 

Number of Data in Training Set

 

     Estimates were obtained for 25%, 50% and 62.5% of total in the training set.  The best was selected from these.

 

Number of Variables Selected at Each Layer

        

This is usually taken as a fixed number of a constantly increasing numbers [usually given as a fractional increase in number of independent variables present in the precious level].  In this work, a fixed number, equal number of input variables, was taken.  This was done to simplify the computational requirements.

 

The Criterion Function Evaluated on Checking Set

 

Three different criterion functions – regularity criterion, unbiased criterion and combined criterion were attempted.

Regularity criterion is given by the equation

                      n

             å         (Yi - Zij ) 2

i= nt+1

rj2  =                                                     (3)       

                        n

               å      Yi2

                i= nt+1

 

j = 1, 2……….m(m-1)/2

 

Where Zij refers to estimate of ith dependant variable using jth equation.

Regularity Criterion has good prediction power but sensitive towards noise.

 

     Unbiased criterion selects models that are insensitive to data from which it is built and hence gives good noise immunity but may not have good predictive power.  The criterion value is given by

                                       n

                        å    (Yi - Zij ) 2

                        i= 1

Uj2  =                                                         (4)

                         n

                         å     Yi2

                                  i= 1

 

j = 1,2…..m(m-1)/2

 

Combined criterion is combination of both of these and it is given by the equation

Cj2= rj2+ uj2.

 

In the present work all the three criterions were considered.

 

NOTATIONS USED

 

MW                             Measured wear (microns)

PW                              Predicted wear (microns)

Vp                               Peak voltage (volts)

Vrms                            Root mean square voltage (volts)

Vavg                            Average voltage (volts)

M/C Time                    Machining time (seconds)

Temp                           Temperature (°C)

Thrust                           Thrust in Kgf

Torque                         Torque in Kg-m

 

 

 

 

 

REFERENCES

 

1.       Tonshoff,H.K., Wulsbedg, J.P.., Kals, H.J.J., Konig, W., and Vam Luytervelt, C.A., Developmentsand trends in monitoring of machining process, CIRP Annals, 37, 611-622, 1988.

2.       Lister P. M., and Barrow, G., Tool condition monitoring systems, Proc. 27th MTDR Conf., pp.271-288.1986.

3.       Cook N.H. Tool wear sensors, wear, 62, pp.49-57, 1980.

4.       Mustapha Elhachimi, Serge Torbaty, Pierre Joyot, “Mechanical Modeling of high Speed Drilling. 1: predicting torque and thrust”, International Journal of machine tools and Manufacture, 39, 553-568, 1999.

5.       Mustapha Elhachimi, Serge Torbaty, Pierre Joyot, “Mechanical Modeling of high Speed Drilling. 2: predicted and experimental results”, International Journal of machine tools and Manufacture, 39, 569-581, 1999.

6.       Diana M. Rincon and A.galip Ulsoy, ‘Effects of Drill Vibrations on Cutting Forces and Torque’, Annals of CIRP Vol. 43/1/1994.

7.       A. Galip Ulsoy and Ozan Tekinalp, “ Dynamic Modeling of transverse Drill bit Vibrations”, Annals of CIRP Vol. 33/1/1984

8.        A. Noori-Khajvi, R.Komanduri, “ On multisensor approach to drill wear monitoring”, Annals of CIRP Vol.       42/11993.

9.       G.byrne, D. Donfeld, I.Inasaki, “Tool condition monitoring – the status of research and industrial application”. Annals of CIRP Vol. 44/2/1995.

10.   Uematsu, T., and Mohri, N.  , Prediction and detection of cutting tool failure by modified group method of data handling. INT. Journal of machine tool design and research, 26(1), 69-80, 1986.

11.   T.H. Wannacott and R.J. Wanancott, Regression- A second course in statistics, John Wiley, 1981.

12.   Farlow, S.J. (ED) , Self organizing methods in modeling. (USA: Marcel Dekker), 1984.

13.   Ivaknenko, A.G., polynomial theory of complex systems. IEEE Transactions on systems, man and cybernatics, pp. 346-378, 1971.

14.   Ravindra H.V., Tool condition monitoring in turning using cutting force, vibration and acoustic emission, Ph.D Thesis, IIT Madras (India), 1993.

  1. Chryssolouoris, G. , and Gullot, M., A comparison of statistical and AI 

approaches to the selection of process parameters in intelligent machining. Transactions of ASME, Journal of Engineering for Industry, 112 (2), 122-131, 1990.