Int J Adv Manuf Technol (2013) 64:613621 DOI 10.

1007/s00170-012-4039-x

ORIGINAL ARTICLE

Vibration analysis in milling titanium alloy based on signal processing of cutting force
Panling Huang & Jianfeng Li & Jie Sun & Jun Zhou

Received: 3 November 2011 / Accepted: 30 January 2012 / Published online: 25 March 2012 # Springer-Verlag London Limited 2012

Abstract In metal cutting processes, the chatter may cause fast wear of tools and poor surface quality of the processed parts; it can happen on different cutting parameters, but how do we identify the chatter and how do we select suitable cutting parameters to avoid chatter at high material removal rate (MRR). In this paper, the signal processing methods such as time domain, frequency domain, and timefrequency domain analysis are introduced. The signals of cutting force that were collected in milling titanium alloy Ti6Al4Vat variable cutting speeds varied from 80 to 360 m/min; signal analysis methods such as time domain, frequency domain, and time frequency domain were put forward. Further analysis results reveal that the chatter occur when cutting speeds are 240 and 360 m/min, when the maximum value of cutting forces increase by 61.966.8%, the standard deviation increase by 84.186.1%, and the surface roughness increase by 34.2 40.5% compared with that of at 80 m/min. Detail signal d2 is employed to monitor cutting stability state from the result of wavelet analysis.

Keywords Signal analysis . Titanium alloy . Fast Fourier transform . Wavelet analysis . Chatter

1 Introduction High-speed machining has increased in recent years, especially in the aerospace industry due to its advantages such as higher material removal rates, better surface finish and lower cost, and so on [13]. However, the unstable machining (namely chatter vibration) is one of the main limitations for high-speed machining, consequently shortens tool life and decreases machined surface quality. In aerospace industry, titanium alloys are extensively used due to their excellent performance in aerospace environment [4]. However, it is difficult to be machined because of their poor thermal conductivity, low elastic modulus, and high chemical activation. Vibration and high cutting temperature during high-speed machining lead to fast wear of tools and give poor surface quality of the processed parts [58]. Simultaneously, most materials need to be removed from roughcast due to the characteristics of component design in aerospace industry, so machining cost becomes very high when these parts are processed at low cutting speed (traditionally, cutting speeds for titanium alloy are often limited at 60 m/min [9]). These are some constraints in the development of aerospace industry, so how to increase cutting efficiency at stable cutting is very necessary. In order to monitor the cutting status, some processing signals must be measured such as acceleration [10], cutting force [1113], motor current [14], acoustic emission (AE) [15, 16], etc. No matter which signals are chosen, the signal processing methods are essential, namely time domain analysis, frequency domain analysis, and time frequency

P. Huang : J. Li (*) : J. Sun : J. Zhou School of Mechanical Engineering, Shandong University, Jinan 250061, China e-mail: ljf@sdu.edu.cn P. Huang e-mail: hfpl@sdu.edu.cn J. Sun e-mail: sunjie@sdu.edu.cn J. Zhou e-mail: zhoujun@sdu.edu.cn P. Huang : J. Li : J. Sun : J. Zhou Key Laboratory of High Efficiency and Clean Mechanical Manufacture, Shandong University, Ministry of Education, Jinan 250061, China

614

Int J Adv Manuf Technol (2013) 64:613621

domain analysis. Time domain signals are raw signals collected from the sensors. When the chatter occurs, milling forces were found to increase dramatically compared with that of at stable cutting by document [17, 18]. Document [19] thinks the chatter can be considered by the character of an increase of vibration amplitude in time domain. So we can identify the chatter by peak amplitude of the signal collected. Frequency analysis is particularly useful for distinguishing cutting stability status; a signal can be transformed from time domain to frequency domain by using Fourier transform. Document [20] identifies the cutting chatter by fast Fourier transform (FFT) of the once per revolution sampled audio signal in milling. However, Fourier transform has an inherent drawback. Nonstationary transient information in time domain cannot be clearly identified in the frequency domain when Fourier transform is only used. A more flexible approach called wavelets transform has been developed to decompose the signal into various components at different time windows; the wavelet transform for monitoring machining operations has also been reported. Documents [13, 2124] detect cutter breakage based on wavelet transform; it performs multilevel signal decomposition of raw signals collected so that cutter breakage features can be extracted. In this paper, the use of cutting force signal to detect the chatter in milling operations has been studied. Based on Fourier transform, the chatter can be ascertained and the chatter frequency can be obtained. Based on wavelet transform, the cutting force signals are decomposed into a set of approximation coefficient and detail coefficients. The former is a low-frequency component, and the latter are highfrequency components of the signal that is called a multilevel signal decomposition. The paper is organized as follows: In Section 2, signal analysis methods are introduced, namely time domain analysis, frequency domain analysis, and timefrequency domain analysis. The detection method of the chatter is proposed. In Section 3, the experimental setup is described; cutting force signals in end milling are collected. In Section 4, the cutting force signals are analyzed by time domain, frequency domain, and timefrequency domain by MATLAT software. Finally conclusions are given in Section 5.

1. Maximum value: The cutting edge can bear larger load for high milling force, consequently shortening tool life, so often the maximum milling force is used in the time domain analysis. The maximum milling force can be expressed as: Fmax maxF 1; F 2 F n; n 1; 2; :::; N 1

In order to reduce random error in milling course, the average value of many peak values of milling forces during a time period are selected generally. 2. Mean value: the average value of a signal. It is expressed as the following formula: x
N 1 X F n N n 1

3. Standard deviation: the spread of the data about the mean value. It is expressed as the following formula: s n 1 X 2 x 3 xn x 2 N n 1

Normally, a low standard deviation indicates that the data points tend to be very close to the mean value, whereas high standard deviation indicates that the data points are spread out over a large range of values.

2.2 Frequency domain analysis To identify the frequency content of milling force signals, Fourier transform is commonly used to transform from time domain to frequency domain. Because the signals from sensors are discrete, discrete Fourier transform (DFT) usually is used. DFT is given by: F k
N 1 X n 0

F nej N kn ;0  k  N 1

2p

2 Signal analysis methods 2.1 Time domain signals analysis for dynamic milling force In milling, dynamic milling force signal F can be used to monitor cutting status, given a discrete time varying signal F (n) (n 0 1, 2,, N) obtained directly from sensor; the following eigenvalues can be used to describe in time domain.

While DFT transform can be applied to any complex valued series, in practice for large series, it can take considerable time to compute. A much faster algorithm has been developed by Cooley and Tukey around 1965 called FFT. Generally, the frequency spectrum is discrete to periodic signals, the amplitude spectrum appears at its variation frequency and harmonics. In milling process, the signal of cutting force is periodic and its variation frequency is tooth passing frequency (TPF), so the amplitude spectrum of the cutting force shows peaks at TPF and its harmonics. But normally the peak value of milling force will appear at spindle frequency (SF) and its harmonics for the mill runout.

Int J Adv Manuf Technol (2013) 64:613621

615

where a and b are the scaling and shifting parameters in wavelet transform, respectively. A family of scaled and shifted wavelets can be produced through varying the parameters a and b. For a discrete signal F(k), k 0 1, 2, 3,, a discrete wavelet transform can be used. The most commonly used discrete wavelet transform is the scaling and shifting of parameters expressed as follows: 8 > a 2j > < j; k 2 Z 9 > > : b ka k 2j So, the coefficients C(a, b) of wavelet transform of the discrete signal F(n) can be expressed as:
Fig. 1 Experimental setup

Wj;k n

1 X

SF is defined as SF n 1000v 60 60p D 5

1 n k 2j F n p y 2j 2j n 0

10

3 Experimental procedure All experiments were performed in a vertical CNC machine centre (DAEWOO ACE-V500) with dry and down milling conditions. The experimental setup is shown in Fig. 1. The experiments were carried out by using solid cemented carbide end mill. The mill diameter is 20 mm with its overhang length 74.3 mm, four cutting edges (N), and 42 helix angle. The workpiece material is titanium alloy Ti6Al4V treated by stress relieving annealing; its main chemical components are shown as Table 1. Fx(t), Fy(t), and Fz(t) are dynamic milling forces at the direction of mill feed, perpendicular to the machined surface and axial directions, respectively; the signals of milling forces were picked up by the dynamometer (Kistler 9257B), and the sampling frequency was set as 7,000 Hz to see the frequency range between 0 and 3,500 Hz. The cutting speeds (v) were varied from 80 to 360 m/min in steps of 40 m/min with constant feed per tooth, axial depth of cut, and radial depth of cut in the experiments, which are listed on Table 2.

where n and v are the spindle speed (in revolutions per minute) and linear speed (in meters per minute), respectively, and D is the diameter of the mill. TPF is defined as TPF N SF 1000Nv 60p D 6

while the appearance of peaks at additional frequencies indicates the chatter. This well-known property of milling dynamics is often exploited for detection of the chatter. 2.3 Wavelet transform Continuous wavelet transform (CWT) is used to divide a continuous time function into wavelets. If a square integral function (t) satisfies the following condition:  2  Z1  _y w dw < 1 7 jw j
1

(t) is called mother wavelet, where

y w is the Fourier 4 The experimental results 4.1 Time domain and frequency domain analysis In order to analyze the milling stability at different cutting speed, first, cutting force signal F y was analyzed at
Al 5.56.75 V 3.54.5 Fe <0.25 O <0.2 C <0.08 N <0.05 H <0.01

transform of (t). Given a time varying signal F(t), the coefficients C(a, b) of wavelet transform of the signal can be expressed as: Z 1 1 tb C a; b dt 8 F t p y a a 1
Table 1 The chemical components of Ti6Al4V

Elements Composition wt.%

Ti Balance

616 Table 2 Cutting parameters Experimental no. 1 2 3 4 5 6 7 8 Cutting speed, v (m/min) 80 120 160 200 240 280 320 360 Spindle frequency (Hz) 21.22 31.83 42.44 53.05 63.66 74.27 84.88 95.49

Int J Adv Manuf Technol (2013) 64:613621

Feed per tooth, fz (mm/tooth) 0.08

Axial depth of cut, ap (mm) 20

Radial depth of cut, ae (mm) 0.5

Fig. 2 FFT analysis on the radial force Fy: a v 0 80 m/min, b v 0 120 m/min, c v 0 160 m/ min, d v 0 200 m/min, e v 0 240 m/min, f v 0 280 m/min, g v 0 320 m/min, and h v 0 360 m/ min

Int J Adv Manuf Technol (2013) 64:613621 Fig. 3 The radial force versus time: a v 0 80 m/min and b v 0 360 m/min

617

frequency domain, radial milling force Fy was treated with fast FFT using MATLAB software, where the time rectangular window was set to 2 s. The results are shown in Fig. 2. When cutting speeds are 80 and 120 m/min, the dominant peak value occurs at the low multiplies of SF (1 and 4), but when cutting speeds are 160, 200, 280, and 320 m/min, besides the low multiplies of SF (1 and 4), high multiplies of SF occur many milling force peak value shown as Fig. 2c, d, f, g. From Fig. 2f, a significant amplitude occurs at 816 Hz (11 SF), the smaller amplitude occurs at 1,114 Hz (15 SF), which can be caused by highfrequency energy increase with wear of the mill during cutting. When cutting speeds are 240 and 360 m/min, milling force peak values occur also at different frequencies from multiplies of TPF and SF, such as 734 Hz at 240 m/ min, and 730, 1,111, 1,268, 1,363, and 1,459 Hz at 360 m/ min. From the former analysis, it is known that the chatter can be assessed from newly occurred frequency which is different from multiplies of TPF and SF, so the chatter occurred when cutting speeds are 240 and 360 m/min. Another remarkable characteristic is that the chatter frequencies are more complex at 360 m/min than at 240 m/min because the mill happens serious wears and more complex vibration occurs when cutting speed is 360 m/min. The whole raw signal of the cutting force at 80 and 360 m/min is shown in Fig. 3. The cutting process is divided to three states: entry

cutting, steady cutting, and exit cutting, the data during steady cutting state are used to analyze in time domain. Figure 4 presents the effect of cutting speed on the maximum value, mean value, and standard deviation of cutting force Fy and surface roughness. It was observed that the mean value of cutting forces has no significant change even at chatter cutting (at 240 and 360 m/min), while the maximum value and standard deviation of cutting force change significantly with different cutting speeds. The maximum value and standard deviation increase slightly with cutting speed varied from 80 to 160 m/min because the impact force and strain hardening rate of the materials increase with the milling speed growing, which cause increasing milling force. Meanwhile, the cutting temperature also increases with increasing cutting speed, which, however, results in thermal softening of the workpiece material, consequently causing the decreasing of milling force. Finally, milling forces only slightly increase under the combined effects of those two sides. Then, the maximum value and standard deviation increase by 36.1% and 67.6% at 200 m/min compared with that of at 80 m/min, respectively; the reasons can be known from Fig. 2d, a bigger milling force peak value occurs at 790 Hz (15SF), so the energy in high frequencies increases

Fig. 4 The radial force, surface roughness versus cutting speed. max the maximum value, mean the mean value, stand the standard deviation, Sa surface roughness

Fig. 5 Wavelet decomposition diagram

618

Int J Adv Manuf Technol (2013) 64:613621

Fig. 6 Wavelet analysis of cutting force versus data point number: a v 0 80 m/min and b v 0 360 m/min

amplitude of the milling force. When cutting speed is 240 m/min, the maximum value and standard deviation increase by 61.9% and 84.1% compared with that of at 80 m/min. When cutting speed is 360 m/min, they increase by 66.8% and 86.1% compared with that of at 80 m/min. So the chatter is the main reason that causes a significant increase of milling force, and an increase of high frequency energy for mill wear is another reason. The surface roughness curve measured has similar variation tread with that of the maximum value and standard deviation especially with the latter. When the cutting speeds are 240 and 360 m/min, the surface roughness increase by 34.2% and 40.5% compared with that of at 80 m/min. The reasons leading to this tread of roughness curve are similar to the reasons of milling force, namely the chatter and the increase of high-frequency energy for the mill wear. 4.2 Wavelet analysis of dynamic milling force signals Usually, the Daubechies wavelets (written dbN) are suitable for detection of a sudden frequency change in a signal [11], the higher order N of the wavelet, the more complicated signal processing is required. In this paper, db3 is selected to

use wavelet analysis. Figure 5 shows a six-level wavelet decomposition diagram. The analysis frequency is 3,500 Hz because the sampling frequency of cutting force is set to 7,000 Hz. Where a6 is low-pass approximation coefficient, d1d6 are high-pass detail coefficients, a6 belongs to the frequency bandwidth [054.6875 Hz], d6d1correspond to the frequency bandwidths [54.6875109.375 Hz], [109.375 218.75 Hz], [218.75437.5 Hz], [437.5875 Hz], [875 1,750 Hz], and [1,7503,500 Hz], respectively. So the raw signals can be expressed by: s a6 d 6 d 5 d 4 d d 2 d 1 11

Fig. 7 Standard deviation of cutting force at different wavelet coefficients

Int J Adv Manuf Technol (2013) 64:613621

619

Fig. 8 Range of cutting force at different wavelet coefficients

Figure 6 presents the wavelet decomposition of cutting force signals at 80 and 360 m/min, respectively. From Fig. 6a, when cutting speed is 80 m/min, the cutting force waves s are mainly distributed in approximation signal a6 (054.68755 Hz), the amplitudes of a6 are bigger than that of other frequency bands signals, and the amplitudes of d6 (54.6875109.375 Hz) are second to a6 because frequency spectrum is dominated by the SF (21.22 Hz) and TPF (82.44 Hz) shown as Fig. 2a. Another remarkable characteristic is that there are some burst events at exit cutting state pointed out by circle because there are some high-frequency frictions between mill and workpiece at exit cutting state, its energy is dominated at frequency bands d3d1 namely

Fig. 9 Detail coefficients d2 of cutting force versus data point number: a v 0 80 m/min, b v 0 120 m/min, c v 0 160 m/min, d v 0 200 m/min, e v 0 240 m/ min, f v 0 280 m/min, g v 0 320 m/min, and h v 0 360 m/min

620

Int J Adv Manuf Technol (2013) 64:613621

437.53,500 Hz. Figure 6b shows the different characteristic from Fig. 6a at 360 m/min when the chatter happens, the approximate signal a6 cannot reflect the raw signal s, the cutting force is dominated by the frequency bands d2 namely 8751,750 Hz; this can be seen from Fig. 2h. The chatter frequencies are concentrated in the frequency band between 700 and 2,400 Hz; d3 (437.5875 Hz) and d1 (1,750 3,500 Hz) have the bigger amplitudes than other frequency bands. Figures 7 and 8 show the standard deviation and range of cutting force at different wavelet coefficients. The range is the difference value of the maximum cutting force and minimum cutting force. From the two figures, detail coefficients d2 [875, 1,750 Hz] have the similar variation tread to raw signals s about standard deviation and range of cutting force, so d2 can be employed to monitor stability state. Figure 9 shows the detail coefficients d2 of cutting force at different cutting speeds. From Fig. 9a, b, cutting forces have obvious period when cutting speed is at 80 and 120 m/min and their cutting force peak values are about 50 and 60 N, respectively. The periods become bad when cutting speeds at other cutting speeds, especially when cutting speeds are beyond 200 m/min. The amplitude values significantly increased when the chatter occurs namely at 240 and 360 m/min. There are medium amplitude values when cutting speeds are 200, 280, and 320 m/min. The reasons can be the mill wear which causes the increase of high-frequency energy.

to monitor cutting stability state and mill wear state because it has similar tread to the standard deviation and range of cutting force.
Acknowledgments The authors acknowledge the support from the National High Technology Research and Development Program ("863"Program) of China (Grant no. 2009AA044303), Shandong Provincial Natural Science Foundation, China (Grant no. ZR2011EEZ001), and Independent Innovation Foundation of Shandong University (Grant no. 2011JC015).

References
1. Duncan GS, Tummond MF, Schmitz TL (2005) An investigation of the dynamic absorber effect in high-speed machining. Int J Mach Tool Manuf 45:497507 2. Lpez de Lacalle LN, Prez-Bilbatua J, Snchez JA, Llorente JI, Gutirrez A, Albniga J (2000) Using high pressure coolant in the drilling and turning of low machinability alloys. Int J Adv Manuf Technol 16:8591 3. Lopez de Lacalle LN, Perez J, Llorente JI, Sanchez JA (2000) Advanced cutting conditions for the milling of aeronautical alloys. J Mater Process Technol 100:111 4. Ezugwu EO, Wang ZM (1997) Titanium alloys and their machinabilitya review. J Mater Process Technol 68:262274 5. Sun S, Brandt M, Dargusch MS (2009) Characteristics of cutting forces and chip formation in machining of titanium alloys. Int J Mach Tool Manuf 49:561568 6. Basturk S, Senbabaoglu F (2010) Titanium machining with new plasma boronized cutting tools. CIRP Ann Manuf Technol 59:101 104 7. Sun J, Guo YB (2009) Material flow stress and failure in multiscale machining titanium alloy Ti6Al4V. Int J Adv Manuf Technol 41:651659 8. Chen G, Ren CZ, Yang XY, Jin XM, Guo T (2011) Finite element simulation of high-speed machining of titanium alloy (Ti6Al4V) based on ductile failure model. Int J Adv Manuf Technol 56:1027 1038 9. Rahman M, Wang ZG, Wong YS (2006) A review on high-speed machining of titanium alloys. Jpn Soc Mech Eng Intern J 49 (1):1120 10. Al-Zaharnah IT (2006) Suppressing vibrations of machining processes in both feed and radial directions using an optimal control strategy: the case of interrupted cutting. J Mater Process Technol 172:305310 11. Armando ISA, Anselmo ED, Robson P (2010) Vibration analysis of cutting force in titanium alloy milling. Int J Mach Tool Manuf 50:6574 12. Liao YS, Young YC (1996) A new on-line spindle speed regulation strategy for chatter control. Int J Mach Tool Manuf 36(5):651 660 13. Lee BY, Tarng YS (1999) Milling cutter breakage detection by the discrete wavelet transform. Mechatronics 9:225234 14. Li XL, Guan XP (2004) Time-frequency-analysis-based minor cutting edge fracture detection during end milling. Mech Syst Signal Process 18:14851496 15. Cao HR, Chen XF, Zi YY, Ding F, CheHX TJY, He ZJ (2008) End milling tool breakage detection using lifting scheme and Mahalanobis distance. Int J Mach Tool Manuf 48:141151 16. Kim JS, Kang MC, Ryu BJ, Ji YK (1999) Development of an online tool-life monitoring system using acoustic emission signals in gear shaping. Int J Mach Tool Manuf 39:17611777

5 Conclusions In this paper, cutting force variation is described based on time domain analysis, frequency domain analysis, and wavelet analysis to ascertain the effect of the cutting speeds on cutting stability. The following conclusions were drawn: 1. Three analysis methods of machining signals were introduced, namely time domain analysis, frequency domain analysis, and timefrequency analysis. 2. Chatter happens when cutting speed are 240 and 360 m/ min according to the results of frequency analysis. The maximum value, the standard deviation of cutting force, and machined surface roughness obviously increased by 61.966.8%, 84.186.1%, and 34.240.5 % compared with at stable cutting, respectively, so the chatter should be prevented in the cutting process. 3. When cutting speeds are 160, 200, 280, and 320 m/min, high multiplies of SF occur many milling force peak values, which can be because high-frequency energy increases with mill wear or breakage during cutting. 4. Daubechies wavelets (db3) are used to analyze the cutting force signals. The detail coefficient d2 is employed

Int J Adv Manuf Technol (2013) 64:613621 17. Huang PL, Li JF, Sun J, Song LY (2010) Vibration analysis in high speed machining titanium alloy with solid cemented carbide end mill. Mater Sci Eng Powder Metall 15:574579 18. Huang PL, Li JF, Sun J, Ge MJ (2012) Milling force vibration analysis in high-speed-milling titanium alloy using variable pitch angle mill. Int J Adv Manuf Technol 58(1):153160 19. Yao ZH, Mei DQ, Chen ZC (2010) On-line chatter detection and identification based on wavelet and support vector machine. J Mater Process Technol 210:713719 20. Schmitz T, Medicus K, Dutterer B (2002) Exploring once-perrevolution audio signal variance as a chatter indicator. Mach Sci Technol 6:207225

621 21. Cao HR, Chen XF, Zi YY, Ding F (2008) End milling tool breakage detection using lifting scheme and Mahalanobis distance. Int J Mach Tool Manuf 48:141151 22. Li XL (2001) Detection of tool flute breakage in end milling using feed-motor current signatures. ASME Trans Mechatron 6:491498 23. Li WH, Gong WG (2005) A method of recognizing tool-wear states based on a fast algorithm of wavelet transform. J Mater Process Technol 170:374380 24. Zhu KP, Wong YS, Hong GS (2009) Wavelet analysis of sensor signals for tool condition monitoring: a review and some new results. Int J Mach Tool Manuf 49:537553