The Correlation of Vibration Signal Features to Cutting Tool

Int J Adv Manuf Technol(2002)19:705–713 Ownership and Copyright2002Springer-Verlag LondonLimitedThe Correlation of Vibration Signal Features to Cutting Tool Wear in a Metal Turning OperationD.E.Dimla,Snr.School of Engineering and Technology,De Montfort University,Leicester,UKThis paper describes a tool-wear monitoring procedure in a metal turning operation using vibration features.Machining of EN24was carried out using coated grooved inserts,and on-line vibration signals were obtained.The measured tool-wear forms were correlated to features in the vibration signals in the time and frequency domains.Analysis of the results sug-gested that the vibration signals’features were effective for use in cutting tool-wear monitoring and wear qualification. Keywords:Metal turning;Tool condition monitoring; Vibration signals;Wear measurements1.IntroductionDuring metal turning operations,a complex and dynamically varying system develops between the workpiece and the cutting tool.Close supervision of tool wear levels must be maintained as excessive tool wear could lead to severe deterioration of the workpiece surfacefinish,possibly incurring additional costs in material and work.In the worst-case scenario,catastrophic tool failure might lead to machine tool damage,and high replacement costs[1–3].Direct methods for determining tool wear such as those associated with measuring and correlating the volumetric loss of the cutting tool material are time consuming.A number of mechanical and physical based sensors such as touch trigger probes,and optical,radioactive,and electrical resistance measurement techniques and proximity sensors have been developed towards this purpose[3].These sensors often require the interruption of the cutting process in order to take measure-ments(post-wear monitoring),thereby increasing the lead-and downtime.Indirect methods of tool wear monitoring on the other hand, can be applied on-line(real-time),as they rely on process changes resulting from the tool wear,cutting conditions,and Correspondence and offprint requests to:Dr D. E.Dimla,Rapid Prototyping&Manufacturing Group School of Engineering and Tech-nology,De Montfort University,The Gateway,Leicester,LE19BH, UK.E-mail:ddimlaȰ variability in the workpiece material,etc.Suitable sensors have been developed for particular applications,such as static and dynamic force measurement,acoustic emission and vibration. Several methods have also been pursued in the quest to develop tool-wear monitoring systems to identify and separate tool-wear effects on the sensor signals from those emanating from cutting conditions.These methods include mathematical and adaptive models,time and frequency domain analysis,pattern recognition,artificial intelligence,and knowledge-based systems [4–6].When a cutting tool pushes through the workpiece material, the workpiece undergoes plastic deformation(dislocation movements)releasing strain energy referred to as acoustic emission(AE)as the bonds between the metal atoms are disturbed.Other sources of AE include phase transformations, friction mechanisms(tool–workpiece contact)and crack forma-tion[1,7,8].Dimla[5]has presented a brief review of AE application sensing techniques in tool-wear detection and moni-toring in metal cutting operations.There are many types of AE sensor,but only a few of these can cope with the rugged conditions that prevail in the machining environment.Most AE sensors have been designed for non-destructive testing and inspection and are therefore unsuitable for tool-wear monitoring applications as they cannot withstand the extreme conditions at the cutting point such as high temperatures,large coolant volumes,and the impact from chips[9].Many new concepts on AE sensing have been proposed for use in the metal cutting environment,e.g.Dolinsek and Kopac[9]use the coolant jet as an AE signal transmission medium between the cutting zone and a PZT transducer element.Kim et al.[10]developed an on-line cutting tool life monitoring system using an integral pre-amp AE sensor for gear shaping applications.With the AE sensor attached directly to the workpiece material,r.m.s.values of the AE signal were recorded that reportedly correlated with the pinion cutter tool wear through a series of experimental investigations.Carolan et al.[11]with the aid of a non-contact fibre optic AE sensor,carried out face milling of steels and aluminium,recording the AE signals for tool-wear monitoring. From the ensuing analysis,a model detailing the relationship between measured tool wear and the energy in the r.m.s.value of the AE signal,was developed and validated for various workpiece materials.By using a non-contact AE sensor,uncer-706 D.E.Dimla,Snrtainties from frequency response and variation that characterise the AE signal were eliminated,thus boosting confidence in the results and the reliability of the application.The most important limiting factor in the application of AE in a tool condition monitoring system(TCMS)does not lie in the sensing technology,but on the ensuing analysis[3].This is primarily due to the lack of a suitable knowledge base on AE.Consequently,the user has to experiment and establish the necessary trigger responses to a variety of machining conditions,which involves considerable work.When compared with the concept of lowering and increasing the force limits which can be understood by an operator,spectral analysis is not so easily comprehended.Hence,instead of the system being an aid to the operator,it rather presents a real-time quandary. Choosing a suitable position in which to place the AE sensors so as to sense sufficient AE signals is contentious,as an understanding of the AE path has to be established.AE propagates and proliferates over the entire machining area.A further drawback to the application of AE as an indicator of tool wear is the fact that it is more sensitive to variations in the cutting conditions and noise than to the tool condition itself.As such,using AE on its own to monitor the state of a cutting tool is a difficult task[12].AE,it could be argued, is suitable only as an additional sensing method for increased reliability.Substantially little AE is generated during the tool-wear process compared to the large amount which accompanies tool breakage and fracture[13],with as much depending on the cutting material(workpiece)structure as on the cutting tool.As the emphasis on any tool condition monitoring system would generally be on tool wear rather than tool fracture,AE is not a suitable tool-wear indicator in monitoring applications, but could be used to good effect in detecting tool-tip breakage on machining centres.The development of chatter and its subsequent progression to vibration leads to a change in the work surface relative to the instantaneous cutting direction.Excessive vibration can cause damage to the machine tool,interfering with the feedrate, cutting speed,and depth of cut.Monitoring the relative rate at which the system is disturbed in all three planes can furnish the TCMS with adequate information about the tendency of direction of movement,and hence,can give an indication of that quantity.Vibration is produced by cyclic variations in the dynamic components of the cutting forces,resulting from periodic wave motions.The nature of vibration signals arising from the metal cutting process incorporates free,forced,periodic,and random types of vibration.Direct measurement of vibration is difficult to achieve because the vibration mode is frequency dependent. Hence,related parameters are measured such as the rate at which dynamic forces change per unit time(acceleration)and the characteristics of the vibration are derived from the patterns obtained.Several investigations have been carried out to corre-late vibration signals to tool wear[5,6].El-Wardany et al.[14]investigated the use of vibration signature characteristics in on-line drill-wear monitoring and breakage.Vibration signature features sensitive to the tool wear were identified in the time(ratio of absolute mean value and kurtosis)and frequency(power spectra and cepstra ratio) domains.Experimental results showed that the kurtosis values increased drastically with drill breakage,whereas frequency analysis revealed sharp peaks indicating drill breakage.By combining both techniques,it was possible to devise an effec-tive drill monitoring system.Yao et al.[15]investigated the detection and estimation of groove wear at the minor cutting edge of a cutting tool,by monitoring vibration signatures.A multivariate time series analysis was carried out on the recorded vibration signals using an autoregressive moving averages method.The dispersion of the signals’autocovariance was obtained,which showed that the vibration signal was sensitive to the length of groove wear on the cutting tool.Dan and Mathew[16]employed a discrete modelling method known as a data dependent system to correlate vibration signals to cutting-tool wear.Results showed a consistent variation in the vibration energy within specific frequency bands,regardless of the ensuing cutting conditions.The application of spectral analysis to the acceleration signals showed sensitivity to the cutting-tool wear.Tool wear monitoring ought to involve the complex and diverse nature of the metal cutting process,in the development of a TCMS,which should provide an indication as to when the cutting tool should be changed,without compromising the workpiece surfacefinish,the machine integrity,and the manufactured component tolerances.Sensor signals for TCMS have to be robust,reliable,and applicable.The vibration signatures satisfy these conditions reasonably well,in addition to the fact that vibration signatures require fewer peripheral instruments than AE for instance.Furthermore,vibration signals have the quick response time needed to indicate changes for on-line monitoring[1].An analytical on-line system for TCM based on vibration signature features,is reported in this paper.The main objective of the study was to develop a TCM system based on a turning operation for determining characteristic vibration signal features which are sensitive to tool wear.Analytical modelling of the experimentally acquired on-line acceleration data in the three principal axes was carried out,and their suitability for tool wear monitoring by time and frequency analyses was deter-mined.The effect of tool wear on the sensitivity and repeat-ability of the vibration signals was investigated.A reasonable correlation was observed between the spectral energy at certain frequencies and the condition of the measured cutting tool. 2.Experimental procedureThe machine tool used was a centre lathe onto which was attached a Kistler triaxial(x-,y-,and z-planes)accelerometer (type8730A).A power supply/coupler(Kistler type5134)was used to amplify and decouple the acceleration(vibration) signals which were sampled at over30kHZ using an Amplicon PC-30data acquisition card.A schematic diagram of the instrumentation set-up is presented in Fig.1,and details of the workpiece and tooling material shown in Table1.The tool inserts were coated by chemical vapour deposition with P15having a thick wear resistant coating on a hard resistant substrate.Its inner coating was8␮m thick comprising of6␮m of aluminium oxide(Al2O3)and a thinner outer layerThe Correlation of Vibration Signal Features707Fig.1.Schematic diagram of the experimental test rig.Table1.Test material type and tool details.Workpiece Work material EN24BS970material Hardness Brinell255Composition0.4%C,0.28%Si,0.27%Mo,1.18%Cr,0.5%Mn,1.4%Ni Tooling materialTool holder Sandvik SSBCR2020K12Tool type SCMT120408UMTool material Sandvik Coromant P254025andP154015Overhang/mm45Cuttingfluid Noneusedof titanium nitride.P25had a thick(Ϸ10␮m)layer of Al2O3 over a medium size layer of titanium carbon nitride[17]. The cutting conditions were selected based on the toughness of the workpiece within the tool manufacturer’s recommended range for the workpiece–tooling geometry configuration.The tests were conducted at the following cutting conditions: Cutting speeds:275,300,and350m min−1.Feedrates:0.1,0.2,and0.3mm rev−1.Fixed depth of cut:2mmInterrupted machining was conducted using a fresh tool until the tool failed or the cutting edges were severely worn,when continued cutting risked catastrophic failure.The duration of each test allowed complete stabilisation of the cutting process to be attained while allowing for significant tool wear.Measure-ments offlank,nose,and notch wear lengths were made after the vibration signals had been recorded using a toolmaker’s microscope and the observations had been supplemented by scanning electron microscope profiling.The chip type,the presence of a built-up edge,and tool chipping or fracture were noted.3.P25Insert ResultsInspection of the results showed significant changes in the acceleration signals as the wear length on the cuttingfaceFig.2.Wear vs.time(V=300m min−1,f=0.1mm rev−1).Fig.3.Wear vs.time(V=300m min−1,f=0.3mm rev−1). increased.The results also showed that increasing the feedrate at constant speed decreased tool life,by increasing the burnout activity.The same was true for cutting speed changes at a constant feedrate,though this was more significant in terms of active tool life.3.1Time Domain AnalysisThe wear-time diagrams(Figs2–5)show that it is possible to detect the three phases of tool wear(primary,secondary,and Fig.4.Wear vs.time(V=275m min−1,f=0.1mm rev−1).708 D.E.Dimla,SnrFig.5.Wear vs.time (V =275m min −1,f =0.3mm rev −1).tertiary)but only at low feedrates,regardless of cutting speed.The tool lasted longer at low cutting speeds and the wear phases were more visibly pronounced on the Figures.The most noticeable indicator of imminent tool failure was the sudden escalation of nose wear to failure.Typically,when nose wear approached 0.2mm,there was a risk of catastrophic failure.This level of wear was also found to be an applicable safety threshold for flank wear length.Notch wear,on the other hand,increased steadily,and then remained practically the same for the remainder of tool life.Wear monitoring through the application of the sum total power (STP)or energy contained in the vibration spectra (Figs 6and 7)showed that increases in flank wear length led to an increase in the STP value.As the wear increased,the STP increased gradually,peaking at a flank wear length of 0.2mm before falling.This trend was most noticeable when cutting at high speeds and feedrates.3.2Frequency Domain AnalysisThe noisy nature of the sensor signals required an active selection or mapping of useful features from the sampled sensor signals,to yield features that might be sensitive to tool wear but not sensitive to other parameters.MATLAB codes were developed and used for all the signal processing.TheFig. 6.STP vibration spectra vs.flank wear (V =300m min −1,f =0.2mm rev −1).Fig.7.STP vibration spectra vs.flank wear (V =275m min −1,f =0.3mm rev −1).signal-processing paradigm involved obtaining the fast Fourier transform of the acceleration,passed through a Hanning win-dow for repeatability of the record length throughout.The FFT data obtained was then smoothed using a seventh-order factor to enhance the quality of any visual plots (spectra and contour plots).With the spectra plots,it was possible to visualise the amplitude variations and their corresponding frequencies,whereas the overall frequency bands were more visible on contour diagrams.3.2.1Spectra AnalysisThe vibration spectra (Figs 8–10)had two main peak frequency bands at 2–4kHz (the fundamental resonant frequency of the tool holder)and at 9–11kHz (chip lamination frequency).Detailed inspection of the spectra plots showed that the z -direction was the most sensitive to tool wear.The 2–4kHz band increased steadily,peaking before falling,just before tool failure.The 9–11kHz peak increased as the cutting tool wore,and showed the largest increase.Increasing the feedrate led to a proportional increase in the magnitude of both peaks.How-ever,at high feedrates,the spectra were noisier over wider frequencybands.Fig.8.Z -vibration spectra vs.flank wear (V =300m min −1,f =0.1mm rev −1).The Correlation of Vibration Signal Features709Fig.9.Z -vibration spectra vs.flank wear (V =275m min −1,f =0.1mm rev −1).Fig.10.X -vibration spectra vs.flank wear (V =275m min −1,f =0.3mm rev −1).3.2.2.Contour AnalysisContour diagrams (Figs 11and 12)clari fied observations made from the spectra plots (i.e.the exact location and variation in the principal frequency peaks).The contour diagrams showed a low energy concentration band at 2–4kHz and a larger band at 9–11kHz.No effects of tool wear werevisible.Fig.11.Z -vibration contour (V =300m min −1,f =0.2mm rev −1).Fig.12.Z -Vibration contour (V =275m min −1,f =0.3mm rev −1).3.3Summary of ObservationThe sensitivity of the vibration components to tool wear was obtained by building a sensitivity index table.The table was constructed using the following indices:3,most sensitive;2,moderately sensitive;1,least sensitive;0,not sensitive at all.The sensitivity matrix shown in Table 2was obtained.Table 2shows that nose and flank wear gave a coherent indication of tool wear.All the components of the vibration signal were sensitive to nose and flank wear forms.The y -and z -direction components were the most sensitive,regardless of the cutting conditions.4.P15Insert Results4.1Time Domain AnalysisInspection of the wear diagrams (Figs 13–15)reveals that the low feedrate wear –time plot gave a better indication of the tool life.After 100s of active tool cutting,a very sharp rise in wear was followed by two further rises.Flank and nose wear,as observed with the P25insert,are better indicators of tool wear/life than the notch wear.The STP vibration diagrams (Figs 16and 17)show the z -direction to be the most sensitive to both wear forms.The x -and y -curves remain constant when feedrate was increased from 0.1mm rev −1to 0.3mm rev −1,but rose slightly when the wear length exceeded 0.3mm.At this stage,the z -value was almost 300%higher compared to the x -and y -values,and hadTable 2.Sensitivity matrix for P25tool insert.Vibration componentsCriteriax -direction y -direction z -direction Sensitivity to flank wear 223Sensitivity to nose wear 233Total response456710 D.E.Dimla,SnrFig.13.Wear vs.time (V =300m min −1,f =0.1mm rev −1).Fig.14.Wear vs.time (V =300m min −1,f =0.3mm rev −1).Fig.15.Wear vs.time (V =275m min −1,f =0.3mm rev −1).a downward trend.At low cutting speeds,the STP vibration decreased (in the z -and y -directions)until the wear reached 0.1mm,and then remained constant.The x -axis did not change until a flank wear length of 0.25mm,after which,the value fellsuddenly.Fig.16.STP vibration spectra vs.flank wear (V =300m min −1,f =0.1mm rev −1).Fig.17.STP vibration spectra vs.flank wear (V =300m min −1,f =0.3mm rev −1).4.2Frequency Domain AnalysisThe vibration spectra (Figs 18–20)generally were noisier than those from machining with the P25insert.Three main peaks were identi fied in all three axes at the 1–2,3–5,and 10kHz bands,though with signi ficant variations.In the x -direction,the 1–2kHz peak had a large bandwidth that decreased as the wear length increased.No signi ficant changes occurred ontheFig.18.X -vibration spectra vs.flank wear (V =300m min −1,f =0.1mm rev −1).The Correlation of Vibration Signal Features711Fig.19.Y -vibration spectra vs.flank wear (V =300m min −1,f =0.3mm rev −1).Fig.20.Z -vibration spectra vs.flank wear (V =275m min −1,f =0.3mm rev −1).y -axis spectra.On the z -axis,the most signi ficant change was at the 10kHz peak.This increased considerably as the wear length increased and fell as the tool entered its tertiary wear phase.Increases in the feedrate led to a widening of the bandwidth,which became wider as the wear length increased.Cutting speed changes did not appear to have any signi ficant effect on the spectra plots.The contour diagram shows that except for the z -axis plot (Fig.23),the x -and y -plots (Figs 21and 22)were scattered at low and high feedrates,covering a wide bandwidth.The z -axis had its clearest and sharpest concentration at 10kHz.No other signi ficant features were immediately apparent.4.3Summary of ObservationA sensitivity matrix was constructed as before with numbers allocated (3,most sensitive;2,good indicator;1,least sensitive)and is shown in Table 3.The STP vibration components were more sensitive to tool wear than the spectra displays (considering the total response).In terms of the individual axes,the z -axis cutting direction was the most sensitive for both the STP and spectra displays.5.Discussion of ResultsAs the vibration signals consist mainly of a combination of stochastic,periodic,and superimposed harmonic signals,aFig.21.X -vibration contour (V =300m min −1,f =0.3mm rev −1).Fig.22.Y -vibration contour (V =300m min −1,f =0.1mm rev −1).method of observation that combines both aspects of signal and process behaviour is required.The computation of the sum total power contained in each spectrum ensured that the most signi ficant properties affected by cutting tool wear on the tool edges were included in the analysed signal.A possible shortcoming in these investigations was the rela-tive dif ficulty of accurately measuring the flank,nose,and notch wear lengths,especially when they were irregular.Quoted wear values would be expected to deviate from the true or ideal values because of the element of subjectivity.The maximum flank wear length (V BMax )was used as the standard measure of flank wear length.In almost all the tests,the end of tool life was indicated by the sudden escalation of both the flank and nose wear land lengths.Increased wear levels led to an increase in the contact area owing to crumbled cutting edges.The transition of friction from static to sliding owing to an increase in the contact area generates high signal pulses,resulting in an increase in the magnitude of the fundamental resonant peak.Beyond a certain flank wear value,the fundamental frequency peak begins to reduce as the contact area increases monotonically,partly712 D.E.Dimla,SnrFig.23.Z -vibration contour (V =300m min −1,f =0.3mm rev −1).Table 3.Sensitivity matrix for P15tool insert.Vibration componentsx -direction y -direction z -direction Total signalresponseSTP vibration signal 2237Vibration spectra 1135Component total 336responsebecause of plastic deformation and the acceleration of nose wear.The chips formed indicate that the high burst was due to an increase in the delamination frequency,arising from the change in deformation behaviour of the material in the shear zone.A noticeable difference in the two insert types was the occurrence of built-up edges (BUE).Although considerable BUE occurred in the P25insert,little or no BUE occurred when machining with the P15insert.Nose wear was found to rise at a faster rate than flank wear land length for both insert types.This seems to suggest that catastrophic tool failure occurred largely at the tool nose,and that nose wear was perhaps a better indicator of the end of tool life.Under such circumstances,it would be possible to use nose wear to signal the end of tool life and its imminent replacement.A practical approach would be to track both flank and nose wear levels to establish exactly which wear regime the tool was in (i.e.its wear phase).Therefore,a rise in nose wear would signal the approach of the catastrophic failure region and would,therefore,be an ideal time to replace the tool insert.6.ConclusionsThe analysis and correlation of vibration signal features to cutting tool wear have been carried out.Time domain features were deemed to be more sensitive to cutting condition than tool wear,whereas frequency based features correlated well with the tool wear.Based on the measurements of cutting tool wear form and the analysis of the vibration signatures,it has been possible to identify the trend of the sensor signals as the insert wear length increased.The characteristics showed that the measured wear values correlated well with certain resonant peak frequencies.The sensor signals were also affected by the different wear modes.It has been possible to establish distinct differences in signal characteristics when cutting was performed with a new insert from that occurring with a detrimental level of wear.AcknowledgementsThe author sincerely thanks The University of Wolverhampton Research Support Unit for its financial assistance.The valuable advice and guidance of Dr P.Lister and Dr N.Leighton is gratefully acknowledged.Special thanks to Maria Giovanna Lo Porto for painstakingly producing the figures.The Correlation of Vibration Signal Features713References1.D.E.Dimla Sr,“Multivariate tool condition monitoring in a metalcutting operation using neural networks”,PhD thesis,School of Engineering and Built Environment,The University of Wolver-hampton,UK,1998.2.D. E.Dimla Sr and P.M.Lister,“On-line metal cutting toolcondition monitoring–Part I:Force and vibration analyses”.International Journal of Machine Tools and Manufacture,40(8), pp.736–768,2000.3.P.M.Lister,“On-line measurement of tool wear”,PhD thesis,Manufacturing and Machine Tools Division,Mechanical Engineer-ing,UMIST,UK,1993.4.D.E.Dimla Jr,P.M.Lister and N.J.Leighton,“Neural networksolutions to the tool condition monitoring problem–a critical review of methods”,International Journal of Machine Tools and Manufacture,37(9),pp.1219–1242,1997.5.D.E.Dimla Sr“Sensor signals for tool wear monitoring in metalcutting operations–a review of methods”,International Journal of Machine Tools and Manufacture,40(8),pp.1073–1098,2000.6.G.Byrne,D.Dornfeld,I.Inasaki,G.Ketteler,W.Ko¨nig and R.Teti,“Tool condition monitoring(TCM)–the status of research and industrial application”,Annals CIRP,44(2),pp.541–567, 1995.7.D.P.Saini and Y.J.A.Park,“Quantitative model of acousticemissions in orthogonal cutting operations”,Journal of Materials Processing Technology,58,pp.343–350,1996.8.A.L.Quadro and J.R.T.Branco,“Analysis of the acousticemission during drilling test”,Surface and Coatings Technology, 93–94(1–3),pp.691–695,1997.9.S.Dolinsek and J.Kopac,“Acoustic emission signals for toolwear identification”,Wear,225(1),pp.295–303,1999.10.J.-S.Kim,M.-C.Kang,B.-J.Ryu and Y.-K.Ji,“Development ofan on-line tool life monitoring system using acoustic emission signals in gear shaping”,International Journal of Machine Tools and Manufacture,39(11),pp.1761–1777,1999.11.T.A.Carolan,S.R.Kidd,D.P.Hand,S.J.Wilcox,P.Wilkinson,J.S.Barton,J.D.C.Jones and R.L.Reuben,“Acoustic emission monitoring of tool wear during the face milling of steels and aluminium alloys using afibre optic sensor.Part1:energy analy-sis”,Proceedings Institute of Mechanical Engineers,Part B,Journal of Engineering Manufacture,211(B4),pp.299–309,1997.12.D.Yan and T.I.El-Wardany and M.A.Elbestawi,“A multi-sensor strategy for tool failure detection in milling”,International Journal of Machine Tools and Manufacture,35(3),pp.383–398, 1995.13.I.N.Tansel and C.McLaughlin,“On-line monitoring of toolbreakage with unsupervised neural network”,Transactions NAMRI/SME,pp.364–370,1991.14.T.I.El-Wardany,D.Gao and M.A.Elbestawi,“Tool conditionmonitoring in drilling using vibration signature analysis”,Inter-national Journal of Machine Tools and Manufacture,36(6), pp.687–711,1996.15.Y.Yao,X.D.Fang and G.Arndt,“On-line estimation of groovewear in the minor cutting edge forfinish machining”,Annals CIRP,40(1),pp.41–44,1991.16.L.Dan and J.Mathew,“Tool wear and failure monitoring tech-niques for turning–a review”,International Journal of Machine Tools and Manufacture,30(4),pp.579–598,1990.17.AB Sandvik Coromant Metal Working Products Catalogue–Turning Tools,1995.。