·Home ·Table of Contents ·Materials Characterization and testing

Characteristic Parameters from Acoustic Emission Signals In Wear of Cutting Tools J. Masetro,J.E. Ruzzante Ensayos no Destructivos y Estructurales, Grupo de Ondas Elásticas,Grupo Latinoamericano de Emisión Acústica (G.L.E.A Comisión Nacional de Energía Atómica, CAC, Av. Gral. Paz 1499, (1650)) Buenos Aires, Argentina. Email: ruzzante@cnea.gov.ar,Web: http://www.cnea.gov.ar/cac/endye/glea.htm E. Serrano ,R. Piotrkowski Escuela de Ciencia y Tecnología, Universidad Nacional de General San Martín, Buenos Aires, Argentina T. Perez Gallego Dpto. de Ingeniería Mecánica e Ingeniería de Materiales (I.M.E.I.M.), Sección de Tecnología de Fabricación (T.F),E.T.S. de Ingenieros Industriales, Valladolid, España. Contact

Summary

Monitoring of the machining process is essential in order to detect any undesired change in the machining environment and proceed to corrective actions. In this sense, early detection is crucial. A very important method is the analysis of acoustic emission signals during the cutting process, and on line detection of any significant changes in those signals. With this goal in mind we began in this paper with a method to characterize the turning process of a steel material on a lathe. The characteristic variable was the Acoustic Emission (AE) signal from a broad band transducer close to the cutting tool edge. The analysis of the digitized AE signals involved 1/f-type processes and wavelet theory.

1. Introduction

A cylindrical machining, which is a continuous two-dimensional cuttings process, could be simplified by the Orthogonal Cutting Model. But as in any real process, additional effects would appear, such as the chip adhesion or entangling to the tool. This chip congestion could produce random variations in the level and characteristics of AE signals.

Machining involves the removal of the chip by a wedge shaped tool from a larger body. This action causes gradual wear, chipping and the eventual catastrophic failure of the cutting tool. In this paper we intend to correlate characteristics of AE signals with the wear degree of the cutting tool, with the quality of the finished surface and with the cutting rate.

The AE comes from different sources. They are: plastic deformation (dislocation movements), in the Primary Deformation Zone, plastic deformation and sliding friction in the Secondary Deformation Zone, increasing friction between the specimen and the edge of the cutting tool (Contact Area), the stripping, falling and beating of the chip, and the entangling of continuous chip with the tool. It makes sense that AE signals originated in different sources would present identifying features that could change with the increment of tool wear tool [1][2].

In general the available methods for tool wear detection are divided in two main groups: [3]

• Direct methods: volumetric loss connected with tool wear is directly evaluated.
• Indirect methods: detection is based on wear related parameters that are easy to measure such as: shear stresses, mechanical vibrations, dissipated power, tool temperature, generated AE. These methods are important because they allow the monitoring tool wear in real time.

AE presents important advantages in performing the monitoring of the cutting process They are:

• The frequency range of AE signals (50 kHz-1MHz) is higher than the corresponding to machine vibrations and environmental noise. So, the obtained signals are relatively free of deleterious external contributions.
• The simplicity of installation of AE sensors and the low cost of them, as compared with those of dynamic equipment.

Sophisticated signal analysis is nevertheless applied nowadays, such as the methods of chaotic dynamics [4]. In this work, AE is analyzed as a stochastic process which gives place to 1/f-type spectra [5][6][7][8]. Some results are corroborated by wavelet analysis, specially adequate for stochastic processes.

2. Experimental

The experimental setup and the employed materials were:
• Material: steel drawn bars SAE 12L14, hardness: 193 Hv 100, diameter: 38.1 mm, length 4500 mm. For the cutting rate test: steel bar SAE 1010, diameter: 42 mm, length: 6000 mm.
• Lathe: TURRI, type:T-250 , power: 20 kW.

Tool: Pulvi Metalúrgica Rossi S.A.C.I.
Tool holder: type: CRH.CSDNR-2020K12.
Plate: type: P-20 made of sintered carbide without coating. For the cutting rate test, type P-10.
• Instruments for AE signals detection:
Sensor: Physical Acoustic Group, Model WS-SN-943. Band Pass sensor 100-1000 kHz.
Pre-amplifier: CISE-Tecnologie Innovative, model PA-01.
CNEA AE Equipment with included RMS meter.
Signal Recorder: OMEGA, Model OMEGALINE 790.
Digitizing Oscilloscope: LeCroy-Dual 600 MHz-5 Gs/s.
• Microscope: Leitz Wetzlar with included Vickers hardness meter.
• Roughness meter: HOMMEL-WERKE TESTER P1. Roughness meter recorder HOMMEL-WERKE RECORDER P1.

The cutting parameters were:

• Turning velocity: n=1000 r.p.m. Then for a cylindrical material with a diameter of 38.1 mm the cutting velocity was 120 m/min. When the speed was studied a bar with a 42 mm diameter was used and then the experiment started with a 132 m/min cutting velocity.
• Passing depth: a_p=1.25 mm. In the case of the cutting velocity study the depth was 0.5 mm.
• Advance: f=0.24 mm/rev.
• Mechanized length: L=325 mm. And L=300 mm in the cutting velocity study.

The AE piezo-electrical sensor was set at the end of the tool holder and fastened there with a thermo-plastic adhesive. The AE signal was transformed in an electrical signal and then, after being pre-amplified, it was sent to the AE equipment where it was amplified and transformed in a RMS value as a continuous record. On the other side, the AE amplifier output entered the digitizing oscilloscope.

We intended to correlate these AE data with the VB wear parameter from the cutting tool, which was measured with the microscope, as the mechanical work on each bar proceeded. The connection with the remaining roughness on each bar and with the speed of the mechanical work was also investigated.

The mechanized length on each bar was 325 mm (300 mm for the cutting velocity study. Data were digitized with a sampling 5 MHz frequency. Five to nine files for each bar were obtained, 25000 points each one.

3. Signal analysis

There are two different classes of AE signals: continuous signals, associated with continuous or even constant processes, and discrete signals (bursts or events), associated with punctual facts such as the removal and impact of chips in the machining process. Both are noises, which means non-differential functions that can be studied with the 1/f theory. The 1/f spectra originate in the energy released in relaxation processes when certain barriers, which maintain the system in a metaestable state, are surmounted. When barriers are higher the phenomenon persists during a longer time and energy is released in clusters that give origin to the burst type AE signals and the seismic signals studied in geophysics. Both Fourier and Wavelet methods were used for the signal analysis.

The log-log representation of de Power Spectral Density (PSD), Fourier transform or the correlation function, of temporal noise series, is generally type f^--b , homogeneous over various decades. The b values, attainable by linear regression are in the range (0-4) for different processes. Due to the great number of data (25000 points per file, 5 to 9 files per bar, 12 bars per test) adequate computational algorithms were developed, with adequate software.

Different methods for signal processing based on the Wavelet Transform are being used during the last decade. Wavelet analysis provides a powerful tool to decompose the given signal in time-frequency elements. This process is performed by the selection of a filter bank or multiresolution scheme. This allows characterizing details in different scales with a precision that is inversely proportional to the corresponding frequency range [9][10]. In particular, this method has been recently implemented for the characterization of stochastic non-stationary processes of homogeneous type, more precisely of the 1/f-type. The significant parameters of these processes, can be estimated just from the wavelet coefficients[11].

A method, former introduced in ^[12] based on the power in each multiresolution level of the dyadic wavelet analysis was employed to detect with higher precision the time at which the chip removal occurred. Wavelet analysis can be efficiently implemented for detecting transients in non-stationary processes as is explained in [13][12]. In this work cubic spine orthogonal wavelets have been implemented, which have revealed excellent analytical capabilities as shown in [9][10].

If s(x) is the sample signal of the analyzed process each d _j [n] , where

is the wavelet coefficient that condenses the information in the scale j and localization around

If we assume that the information of the signal in the scale j is concentrated in N_j coefficients, then a power function can be defined as

The behavior of these parameters at successive levels j, allows estimating relevant features of the involved process.

4. Results

The frequency range where the linear trend in the log-log graph was more pronounced is 200 to 900kHz. It is important to observe that this frequency interval was coincident with the linear range of the AE sensor. The fundamental frequency associates with a rotation of 1000 rpm was near 104.72 kHz.

4.1.AE and chip formation in turning processes.
Fig. 1 shows calculated b values for each sample corresponding to one of the bars. As the Le Croy equipment acquired the samples as long as the mechanical work proceeded, these values are correlated with time.

Fig 1: b values for bar 6, extracted from the processing of the six samples recorded.

Turning is a continuous cutting process. Nevertheless, eventually, due to chip break down, separation and ulterior collision against the tool or the piece, a burst signal adds to the AE continuous signal. This happens when a certain critical length value is achieved.

These discrete events were clearly observed in the temporal signal records. The question is if the b values are sensitive to them. The answer seems to be affirmative if we observe in Fig. 1 the zigzag variation of b vs. t. b varies in a rather cyclic way, correlated with the cyclic chip detachment.

When chips are detached, lower frequency contributions appear in the spectra. However, the highest b values appear a little later in time, as if they were more connected with the ulterior relaxation process. This fulfills the requirements of the SOC theory concerning the relation between the 1/f spectra and the material relaxation processes involving several energy barriers[5].

4.2. A.E and tool wear in turning processes.
As the whole cutting process, involving all the bars, was performed with the same tool, wear can be studied considering signals from different bars. In this paragraph, we study the relation between the variation of the b parameter and the wear degree (VB) of the cutting tool. A clear relation is not obtained. However a trend is observed, in the sense that b diminishes as long as wear degree increases. In order to avoid transient changes due to chip formation, mean b values were considered.

The grater the wear degree, the greater the geometry detriment which leads to worse the working conditions and then to increase the temperature of the contact area. So, the energetic contents associated with different friction and deformation sources are enhanced and AE signals tend to be continuous. In this way, discrete signals, associated with chip formation and chip detachment are in a great deal masked by the continuous ones and accordingly, the mean b values are lower and less scattered. This is observed in Fig. 2.

Fig 2: a) Shows the sequential machining of each bar. b) Shows the wear degree in VB units vs. time, c) Shows the value of the mean value of the b parameter for each bar vs. the mean wear degree of each bar.

4.3. AE and surface properties in turning processes
As long as the turning process proceeds, the surface quality worsens because the mechanical work is performed with a tool with an increasing wear degree. This can be observed in Fig 3. The bad tool condition produces a loss of adequate geometry properties and a temperature increase at the contact area. Both facts give place to the incorrect chip separation and the ulterior chip trapping and accumulation in the recently mechanized surface, so generating a surface roughness increase. From this, the importance of early tool replacement is easily understood

It seems reasonable that the b parameter would behave in a similar way related with roughness or wear degree and this is what really happens as can be corroborated in Fig. (3),

Fig 3: a) surface roughness vs. tool wear degree, b) bvalues vs. wear degree.

This encouraging result shows that the b parameter is a helpful sign for the loss of surface quality. This result could not be accomplished when the RMS signal values was intended as surface quality indicator.

4.4. AE and cutting rate in turning processes
At higher velocities, the tool conditions are more demanding due to the fact that material is removed promptly, more energy is released per unit time, and then more AE signals are produced.

It was observed, as is shown in Fig. 4, that. b values augment with the cutting rate. An explanation could be that while the continuous signal number increases, the increase of the number of discrete signals or bursts, due to chip separation events is more important.

Fig 4: Values of the parameter b vs. cutting velocity.

4.5. Wavelet analysis
Wavelet analysis was applied to signals in order to corroborate the detection of transient processes as the chip removal obtained in Section 4.1. The power of the signals, for levels corresponding to frequencies lying within the AE sensor linear range was calculated, for each sequential sample of the sample. Fig. 5 shows one of these results, for the sequential samples corresponding to bar 6. Fig.5-a, 5-b correspond respectively to levels j=-5 (D w @ 150-300kHz) and j=-6 (D w @ 80-150 kHz). The transient event is in this case detected earlier at lower frequencies. As was obtained in previous work ^[12], the detection with wavelet analysis is earlier, which means that it improves the Fourier methods as failure indicator.

Fig 5: Power obtained by wavelet transform vs. number of sample for bar 6. Left: level j=-5, right: level j=-6. Compare with Fig. 1.

5. Conclusions

• The chip removal events were clearly identified in the signals because they produced more abundant individual AE bursts. In the (1/f)^b type of power spectra higher b values were subsequently obtained.
• The increase in tool wear degree and/or the surface roughness was identified in the signals by an increase in the continuous AE emission, which masks the individual bursts. In the (1/f)^b type of power spectra lower mean b values were then obtained.
• The increase in cutting rate produced a greater number of both continuous and discrete signals but the augmentation of the last number prevailed and so, larger b values were obtained.
• The results obtained for transient signals as those obtained from chip formation and detachment, were corroborated and improved with wavelet analyses.
• The results obtained until now are promising in the sense of finding a useful method for the early detection of a hazardous wear degree of the cutting tool or the misbehavior of the machine tool.

References

1. E. Kannatey-Asibu and D.A. Dornfeld, "Quantitative relationship for acoustic emission from orthoghonal metal cutting", Transactions of the ASME, 103(1981)330-340.
2. L.A. Kendall, "Tool wear and tool life" in "Machining", Vol. 16, Metals Handbook, ASM, 1993.
3. T Pérez Gallego, M. San Juan Blanco, "Aplicación de la emisión acústica a la monitorización de procesos de corte en máquina-herramienta fresadora", Internal Report, E.T.S. de Ingenieros Industriales, Valladolid, España.
4. D.Zupancic and I. Grabec, "Identification of a manufacturing process by AE analysis", Progress in Acoustic Emission VI, The Japanese Society for NDI, 1992, 377-384.
5. M. Schoreder, "Fractals, chaos, power laws", W.H. Freeman and Company, 1991.
6. M.S. Keshner, "1/f noise", Proc. IEEE, 70(1993)N°1.
7. B.B- Mandelbrot, and J.J. van Ness, "Fractional brownian motion, fractional noises and applications", SIAM Rev. 40(1968)422-437.
8. R. Piotrkowski, M.Gómez and J.E. Ruzzante, "1/f noise in acoustic emission signals", Proc. VII RPIC, San Juan, Argentina, 1997, 53-58.
9. Y. Meyer, "Wavelets, algorithms & applications", SIAM, Philadelphia, 1993.
10. S. Mallat, "A wavelet tour of signal processing", Academic Press, San Diego, 1998.
11. G.A. Hirchoren and C.E. D'Attellis, "On the optimal number of scales in the estimation of fractional brownian motion using wavelets", Signal Processing 63(1997) N°1.
12. R. Piotrkowski, G.A. Hirchoren, M.P. Gómez, C.E. D´Attellis and J.E. Ruzzante, Proc. First Meeting of the Acoustic Emission Latin-American Group, Buenos Aires, Argentina September 1999. To be published.
13. C.E. D'Attellis, S.I. Isaacson and R.O. Sirne, "Detecting epileptic events in electroencephalograms using wavelets", Annals of Biomedical Engineering 25(1997)286-293.

© AIPnD , created by NDT.net

|Home|    |Top|