1-s2.0-S0890695599001224-main

International Journal of Machine Tools & Manufacture 40 (2000) 10731098

Sensor signals for tool-wear monitoring in metal cuttingoperationsa review of methods

Dimla E. Dimla Snr. *

Department of Mechanical and Manufacturing Engineering, De Montfort University, The Gateway, LeicesterLE1 9BH, UK

Received 12 May 1999; received in revised form 12 October 1999; accepted 26 November 1999

Abstract

The state of a cutting tool is an important factor in any metal cutting process as additional costs in termsof scrapped components, machine tool breakage and unscheduled downtime result from worn tool usage.Several methods to develop monitoring devices for observing the wear levels on the cutting tool on-linewhile engaged in cutting have been attempted. This paper presents a review of some of the methods thathave been employed in tool condition monitoring. Particular attention is paid to the manner in which sensorsignals from the cutting process have been harnessed and used in the development of tool condition monitor-ing systems (TCMSs). 2000 Elsevier Science Ltd. All rights reserved.Keywords: Tool wear; Cutting forces; Vibration signals; Acoustic emission; Tool temperature; Tool condition monitor-ing systems

1. Introduction

Manufacturing industries drive for cost savings and productivity improvements have culmi-nated in the creation of minimally manned factories. In manufacturing processes that involve metalcutting operations, untended machining concentrated on tool change procedures when required aswith computer numerical control machines. The late 1980s and early 1990s have witnessed achange from the old practice of changing tools automatically, to the feasibility of instituting toolchange procedures based on monitoring the amount of wear on the cutting tool-edges through theimplementation of adaptive tool inspection mechanisms. Traditionally, a tool was changed to meetnew process requirements (e.g., differing tool geometry and material) or to suit the kind of cutting

* Corresponding author. Tel.: +44-7932-755708.E-mail address: [email protected] (D.E. Dimla Snr.).

0890-6955/00/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved.PII: S0 890- 695 5(99 )001 22- 4

1074 D.E. Dimla Snr. / International Journal of Machine Tools & Manufacture 40 (2000) 10731098

Nomenclature

d depth of cut, mmf feed-rate, mm/revFxFyFz cutting force components in the principal three axes, Ni-D ith dimensionKtd crater wear depth, mmV cutting speed, m/minVB max maximum flank wear land length, mmVB mean mean flank wear land length, mmVB min minimum flank wear land length, mm

(roughing, finishing and profiling). To change a tool when worn requires on-line wear monitoring.The reason for this is primarily because tool wear is a complex phenomenon that manifests itselfin different and varied ways.

2. Forms of tool wear occurring during metal cutting

Cutting tool wear can be classified into several types and summarised as follows:

O adhesive wear associated with shear plane deformation,O abrasive wear resulting from hard particles cutting action,O diffusion wear occurring at high temperatures, andO fracture wear such as chipping due to fatigue.

Tool wear processes generally occur in combination with the predominant wear mode, dependentupon the cutting conditions, workpiece and tooling material, and the tool insert geometry. For agiven cutting tool and workpiece material combination, the tool wear form may depend exclus-ively on the cutting conditions, principally cutting speed V and the un-deformed chip thicknesst, and a combination of the aforementioned wear mechanisms. Ranges of cutting speed whereeach type of wear is predominant can be identified by considering the product of these values asVt, which is directly proportional to the cutting speed [1]. Sometimes, the tool life can be consider-ably reduced if the area of cut, the area swept by the cutting tool, is significantly increased (i.e.by increasing the depth of cut mainly). At low cutting speeds, the tool wears predominantly bya rounding-off of the cutting point and subsequently looses sharpness. As the cutting speedincreases the wear-land pattern changes to accommodate the ensuing change with extremely highvalues leading to plastic flow at the tool point. Cratering on the other hand depends largely onthe cutting temperature than on the cutting speed. The various forms of wear-land pattern andprevailing cutting speed are shown in Fig. 1 for a turning operation. The more predominantlyoccurring forms of cutting tool wear often identified as the principal types of tool wear in metalturning using single-point tools are nose, flank, notch and crater wear. Fig. 2 shows how these

1075D.E. Dimla Snr. / International Journal of Machine Tools & Manufacture 40 (2000) 10731098

wear features can be measured in a turning process through implementation of appropriate Inter-national Standards Organisation criteria.

Nose Wear or edge rounding occurs predominantly through the abrasion wear mechanism onthe cutting tools major edges resulting in an increase in negative rake angle. Nose wear can bedependent entirely on the implemented cutting conditions with tool sharpness lost through plasticor elastic deformation. At high cutting speeds, the edge deforms plastically and may result in thelost of the entire nose, Figs 1a and 2b. Edge chipping and cracking occurs during periodic breaksof the built-up edge in interrupted cuts with brittle tool and thermal fatigue. Catastrophic failuremay also occur if the nose is considerably worn or as a result of the utilisation of inappropriatemachining conditions and brittle tools such as ceramics and cemented carbide [2].

Flank wear arises due to both adhesive and abrasive wear mechanisms from the intense rubbingaction of the two surfaces in contact, i.e. the clearance face of the cutting tool and the newlyformed surface of the workpiece. Its rate of increase at the beginning of the tool life is rapid,settling down to a steady state then accelerating rapidly again at the end of tool life. Flank wearleads to a deterioration of surface quality, increased contact area and consequently to increasedheat generation (Figs 1b and 2c).

Wear notch forms at the depth of cut line as the tool rubs against the shoulder of the workpiece(Fig. 2b and c). Wear notch can lead to abrasion setting by the surface layers accelerated byoxidation or chemical reactions, possibly leading to total tool failure.

Crater wear results from a combination of high cutting temperatures and high shear stressescreating a crater on the rake face some distance away from the tool edges, quantified by depthand cross-sectional area (Fig. 1c). Crater wear also arises due to a combination of wear mech-

Fig. 1. Cutting tool wear forms in orthogonal metal cutting.


Fig. 2. Conventional features of turning tool wear measurements.

anisms: adhesion, abrasion, diffusion or thermal softening, and plastic deformation. Severe depthsof crater may trigger catastrophic collapse of the cutting point (Fig. 1d).

3. Requirements of a tool condition monitoring system

The need for monitoring in a metal cutting process encompasses monitoring the machine andthe cutting process dynamics, cutting tools and workpiece to insure optimum performance of thesystems [3]. A tool condition monitoring system can therefore be viewed as serving the follow-ing purposes:

1. advanced fault detection system for cutting and machine tool,2. check and safeguard machining process stability,3. means by which machining tolerance is maintained on the workpiece to acceptable limits by

providing a compensatory mechanism for tool wear offsets, and4. machine tool damage avoidance system.

The lack of a TCMS may include amongst others, excessive power take-off, inaccurate tolerances,


serrations and uneven workpiece surface finish, eventually leading to machine tool and/or machineperipheral damage incurring unnecessary costs.

Much research has been carried out concerning the development of a reliable TCMS. However,none has yet found ubiquitous industrial use [4,5]. Several factors have impeded advances in thedevelopment of TCMSs including inappropriate choice of sensor signals and their utilisation. Oneof the primary reasons for the lack of industrial application of TCMSs is due to the fact thatTCMSs have been developed based mainly on mathematical models, which require huge amountsof empirical data. Another possible hindrance lies in the nature and characteristics of the utilisedsensor signals in general, which tend to be stochastic and non-stationary and therefore difficultto model. The random behaviour can be attributed to the large-scale variation and non-homogen-eities that exist in the workpiece [6]. There are difficulties involved in designing TCMSs thattake account of the noise sources. Typically, most metal cutting processes can be classified ashaving one or more of the following characteristics [7]:1. complex to chaotic behaviour due to non-homogeneities in workpiece material,2. sensitivity of the process parameters to cutting conditions, and3. non-linear relationship of the process parameters to tool wear.

Misleading information easily results as disturbances from either of the aforementioned sourcespose a practical problem, limiting precision and control of the cutting process. A high degree ofreliability for diagnosing the ensuing tool condition with a sufficient level of accuracy thereforebecomes paramount [8]. When unexpected disturbances occur, a TCMS ought to be capable ofdiagnosing and identifying the fault, and possibly isolate or respond with remedial action withina prescribed response time [7].

Interest in the development of intelligent TCMSs in specified operations such as drilling orturning in a rapidly changing and unpredictable environment has been attempted [9]. A majordrawback though to the implementation of the aforementioned system has been that developmentof an adequate sensor for on-line monitoring of the cutting process has not yet been accomplished[4,10,11]. Historically, human operators, using a combination of sight and sound have performedtool condition monitoring (TCM). It is impossible to develop sensors to mimic exactly the humanoperator, who is subjective and flexible but inaccurate. This problem has been circumvented byusing descriptive parameters of the cutting process that show sensitivity to tool wear, thus preclud-ing the need for a unique sensor. In effect, this could be viewed as an extension of the traditionalhuman operator TCM process, and its main advantages are that, emerging and established sensorsignal processing technologies can be employed on the tapped signals, making use of less compli-cated and widely available commercial sensors and instruments. Advanced sensor design permitsinformation gathering of the cutting process enabling adequate measurements. Recent surveys ofsensor application in TCM can be found in Refs [3,5,1216]. A database of TCM literature hasbeen compiled by Teti [17] containing over 500 publications from 1960 to 1995.

The vast amount of literature in this field suggests that a variety of process parameters in themetal cutting environment can be tapped and used to predict the cutting tool-state. This paperpresents a review of some typical application scenarios along with their correlation to tool wearunder experimental conditions. It is provided to cover:

O acoustic emission,


O tool temperature,O cutting forces (static and dynamic),O vibration signature (acceleration signals), andO miscellaneous methods such as ultrasonic and optical measurements, workpiece surface finish

quality, workpiece dimensions, stress/strain analysis and spindle motor current.

4. Acoustic emission (AE)

During metal cutting the workpiece undergoes considerable plastic deformation as the toolpushes through it. Within the deformation zones (dislocation movements) strain energy is releasedas the bonds between the metal atoms are disturbed. This released energy is commonly referredto as acoustic emission. Other sources of AE include phase transformations, friction mechanisms(tool-workpiece contact) and crack formation or extension fracture.

Choi et al. [18] fused AE and cutting forces in their attempt to develop a real-time TCMS forturning operations. Two sets of experiments were conducted using tungsten carbide insert tipswith one set slotted by wire EDM to accelerate fracture while the second was brazed to theworkpiece to induce tool breakage. The recorded data was analysed through a fast block-averagingalgorithm for features and patterns indicative of tool fracture, and it showed the occurrence of alarge burst of AE at tool breakage. Similar work conducted by Jemielniak and Otman [19] useda statistical signal-processing algorithm to identify the root mean square (RMS), skew and kurtosisof the AE signal in the detection of catastrophic tool failure. Cutting force measurements recordedsimultaneously were used as reference signals to indicate when the failure actually occurred.Inspection of the test results indicated the skew and kurtosis to be better indicators of catastrophictool failure than the RMS values.

Kakade et al. [20] used AE analysis to predict tool wear and chip-form in a milling operationby selecting AE parameters (ring-down count, rise time, event duration, frequency and event rate)recorded simultaneously with the corresponding flank wear land length measured at the selectedintervals. Analysis of the results concluded that AE signals could distinguish clearly the cuttingactions of a sharp and worn or broken tool. As for continuous tool monitoring, no suggestionswere made, presumably because the method was deemed unsuitable.

Zheng et al. [21] presented an intrinsic method for AE sensing based on an optic fibre sensor.The sensor consisted of two distinct parts: the sensing element and an interferometer. The sensingelement principally was used to produce a shift of phase in the light transmitted through theoptical fibre allowing the interferometer to detect and measure the photoelastic modulations inthe light intensity. Preliminary drilling and milling operation tests were conducted and the AEsignal measured using an optic fibre sensor and a commercially available PZT AE sensor. Theobtained results were compared and these showed a reasonable degree of agreement. Howeverdetailed experimental trials were not conducted.

Konig et al. [22] performed test cuts to detect fracture and/or monitor the condition of smalldrills using AE features. They circumvented the time consuming and cost-intensive signal pro-cessing usually accompanying AE applications by employing simple process-adapted band passfilters, a rectifier and a low pass filter to convert the normally high frequency AE signals to low


frequency signals. Drilling operations were performed and a ceramic knock detector sensordesigned for industrial application was used to measure the AE signals. The recorded AE-RMSwas plotted for the number of holes that each drill performed before it failed. Inspection of theirplots showed that at the closing phase of tool life (i.e. tertiary phase of tool wear), the RMS valueincreased dramatically. The rise, one could argue, was a direct response to the fractured tool.Hence Konig et al. used this as a prescribed threshold which the RMS for normal operating drillsshould not exceed. This method however was found to be sensitive to tool chipping.

Blum and Inasaki [23] performed experimental test cuts to determine amongst other things, theinfluence of flank wear on the generation of AE signals. They were particularly interested in theuse of the AE mode, a parameter describing the whole characteristics of the cutting process asthe DC component of the measured signal. Experiments were conducted from which AE and thecutting forces were recorded simultaneously for pre-ground and sharp tool inserts. The ensuinganalysis involved studying the effects of the cutting conditions on the chosen AE features andtool flank wear. With knowledge of the former, it was possible for its influence in flank wearinterpretation not to be misconstrued. Inspection of the obtained graphs for AE-mode/cuttingforces and flank wear for various cutting speeds showed an indispensable correlation of AE-modeto flank wear. A not-so-good correlation was realised with the cutting forces, as its slopes weresignificantly smaller compared to the AE-mode ones that were almost linear. This was interpretedas the flank wear length being extremely sensitive to AE-mode. They, however, concluded thatextraction of such information from the AE signal was difficult.

Moriwaki and Tobito [24] proposed a method based on AE measurement and analysis forcoated tool life estimation. The underlying principle behind their devised method for coated toollife estimation was that, during progressive tool wear, the tool material changes from one substratelayer to another and emits AE signals that could be monitored to determine tool life. An experi-mental test rig was set up and tests were conducted on it. The AE and tool wear (flank and crater)were measured together with the surface roughness. Typically, AE RMS values for the recordedAE signal and the wear values (initial, middle and tertiary stages of tool wear) were graphed onthe same scale for comparison. Inspection of the presented plots indicated a strong correlation ofAE RMS amplitude to tool wear, increasing with wear progression. Further analyses to extractstatistical features (mean, variance and the coefficient of RMS) were performed and plots madefor the complete cutting cycle. The variance clearly was the most sensitive to tool wear, as it hadthe largest amplitude in the final phase of tool life. The recorded data was applied to a patternrecognition system and it performed reasonably well. Thus, by measuring the AE emitted fromspecially treated coatings on a cutting tool, it was possible to identify and predict the ensuingtool life. The only drawback to the application of this method, it could be argued, was its exclusiveuse of coated tool inserts.

Roget et al. [25] carried out machining tests from which the sensed AE signals from the cuttingoperation was used to predict the state of the cutting tool. They concluded that such task couldonly be successfully accomplished under specific and limited conditions. Using custom-made AEsensors, turning and milling test cuts with normal grade and alloyed steels were carried out.Several parameters of the AE were recorded (i.e. RMS, mean and peak values). Further statisticalfeatures such as the variance, kurtosis and skew were extracted from the recorded parameters. Acomparison of the recorded AE and measured flank wear curves were carried out, and showed aremarkable similarity with the characteristic three distinct phases depicted on both the wear-time


and AE-time plots. There was a corresponding initial rise in both curves due to the tool beingengaged in cutting. A slowing down of the wear rate reciprocated on the AE signal curve by amuch gentle slope followed this. As soon as the flank wear began to increase to catastrophiclevels, the AE signal reflected the increase and its undulations became more erratic. They extendedtheir method to identifying tool breakage as well but using a milling operation instead of the testbed. Their final conclusions were that AE provided sufficient warning of the ensuing changes inboth cutting conditions, tool breakage and tool wear.

During metal cutting, substantially little AE is thought to be generated compared to a largerAE accompanying tool breakage and fracture [26,27]. AE, it could be argued, is dependent onthe structure of the cutting material than on the cutting tool, with its signal reflecting the behaviourof the response from the machine tool set-up rather than the cutting tool. As the emphasis on anyTCMS would generally be on tool wear rather than tool fracture, AE is not a suitable tool wearindicator in monitoring applications, but could be used to good effect in detecting tool tip breakagein machining centres. On another note, Lister [4] was of the opinion that the most profoundlimiting factor in the application of AE to a TCMS does not lie on the sensing technology, buton the ensuing analysis. This void is due primarily to a lack of a suitable database on AE, implyingthat the user has to experiment and establish the necessary trigger responses to a variety of machin-ing conditions thereby placing considerable burden upon him/her. When compared to instanceswhere for example, the concept of lowering and increasing the force limits might be understood,spectral analysis is not so easily comprehended. Hence instead of the system being an aid to theoperator, it rather presents a real-time quandary. Furthermore, because AE might be sufficientlyavailable on the entire machining area, choosing a suitable area to place the AE sensor to trapsufficient AE signals is debatable, as an understanding of the AE path has to be established. Adrawback to the application of AE as an indicator of tool wear is the fact that its signals are moresensitive to variations in the cutting conditions and noise than of the tool condition itself. UsingAE on its own to monitor the state of a cutting tool is a difficult task [11,28]. AE in the viewof the present author is deemed only suitable as an additional sensing method for increasedreliability for TCM. Dornfeld [29,30] present compelling reviews on the application of AE sensingtechniques in manufacturing processes particularly applied in tool wear detection in machining.

5. The tool tip/cutting edges temperature

Metal cutting generates a significant amount of heat. The resultant high temperatures aroundthe cutting tool edges has a direct controlling influence on the rate and mode of cutting tool wear,the friction between chip and cutting tool, and also that between the cutting tool and the newlyformed surface. Frictional behaviour on the tool faces is thought to affect the geometry of thecutting process by some mechanism not completely understood [31,32]. Two metallic surfaces insliding contact would normally experience dry friction commonly referred to as Coulomb friction.In metal cutting, the coefficient of friction is independent of sliding speed and area of contact.Force is therefore required for the continual shearing of the tips of the asperities or hills. Thisrequired force (or load) is proportional to the frictional force in dry sliding. The coefficient offriction between tool and chip varies considerably due to changes in cutting speed and rake angleresulting in high pressures. In the meantime, the real area of contact would approach unity thereby


giving rise to high frictional forces that eventually lead to high temperatures and render slidingat the interfaces almost impossible [32]. Removal of the generated heat is through the chip, work-piece and/or tool. As the temperature distribution is not uniform, knowing the exact amount ofheat transferred via the tool is not straightforward. It is however thought that the amount of heatremoved or conducted via the chip is as high as 90% of the overall heat generated, implyingthat less than 10% of the heat is either absorbed or dispersed through the tool and workpiecematerial [31].

Sarwar et al. [33] recently developed a thermal imaging system for metal cutting applications.This system essentially consisted of a mathematical model of the energy partition during metalcutting. Thermal imaging data obtained showed that it was practicable and modifiable to fulfilthe requirement of orthogonal cutting. Further work was reportedly underway to extend the modelto accommodate the transient nature of the temperature field during dynamic cutting conditions.

Lin [34] in his attempt to measure the cutting tool temperature on-line during a milling processdevised an inverse approach for real-time tool/workpiece interface temperature. Infrared pyrome-try was employed to measure the actual temperature on the machined surface and a least squareinverse method applied through an ellipsoidal mapping model of the heat conduction equation.Using a 1-D co-ordinate transformation of a moving heat source system, the measured temperatureand heat dissipation to the workpiece was calculated inversely by finite element analysis (FEA)to predict the tool-workpiece interface temperature considered as the heat source. The designedmodel was tested through application of a known heat flux input and the inverse output verified.Plots of the actual temperature of heat source and that estimated by the proposed method showedminute deviations. He proceeded to test the model further using flame heating to primarily verifythe uncertainties in temperature measurement of a moving heat body. The final stage of Linswork involved on-line temperature measurements of a milling process using his now tried andtested inverse estimation method. He concluded that such agreement implied that his model pro-duced accurate real-time temperature estimates and was therefore acceptable for tool-workpieceinterface temperature estimation. The main drawback of this method seem to be the influence oftemperature measurements using this model by the thermal properties of the workpiece and toolingmaterial, and the employed pyrometer had a limited measurement range of 0500 C. Detailedknowledge of the thermal properties of the workpiece material such as its density, thermal dif-fusion and thermal conductivity was deemed crucial to realising this method, effectively rendingit redundant. On another note, Radulescu and Kapoor [35] designed and tested an analytical tooltemperature fields prediction model for use during continuous and interrupted cutting. Testingshowed that the modelled tool-chip interface temperature agreed well with experimental tests.

Raman et al. [36] proposed and developed a mathematical model for cutting tool temperaturemeasurement based on the remote thermocouple sensing (RTS) principle. Differential quadraturemodelling of the forward thermal behaviour of the insulated cutting tool was pursued. The credi-bility of the method lied in the fact that the tool/chip interface temperature distribution had aunique characteristic relationship between the tool-chip temperature and the remote thermocoupletemperature. Cutting process temperature change was then determined by observing the behaviourof the rest of the tool temperature (sink response) to variations in the source of temperature (tooltip). Once the behaviour between sink and source temperatures had been established, it was poss-ible to estimate on-line, the tools interface temperature based on the remote thermocoupleresponse. The existence of large temperature gradients closer to the tools cutting edge led to an


initial modelling to achieve an accurate theoretical base for experimentation, i.e. exact locationsof the thermocouples. They concluded that the mathematical analysis performed was quite suitablefor RTS but more analysis was required to improve the technique as it was still in its early phaseof development.

Stephenson and Ali [37] performed studies on tool temperature effects on interrupted metalcutting and reported theoretical and experimental results. The experimental analysis involved usingboth infrared and tool-chip contact thermocouple temperature measurements. The measured tem-peratures for a series of machining conditions were found to be dependent upon two main factors:

O length of cutting cycles, andO length of cooling interval between cycles.

Overall, the temperature measurements were found to be lower when the cutting was occasionallyinterrupted than for continuous cutting under the same cutting conditions. They pointed out thatit was difficult to instrument a thermocouple for tool-chip or tool-workpiece interface temperaturesensing when the workpiece was not hollow. The best option was the use of a non-contactmeasurement technique such as infrared thermal imaging where taking measurements required adetailed understanding of black body radiation. This technique was only capable of temperaturemeasurements that might be considered at best averages rather than the true temperatures, andtherefore tended to be dominated by chip images.

Chow and Wright [38] devised an on-line method for tool-chip interface temperature measure-ment in a turning process using a standard thermocouple inserted at the bottom of the tool insert.Experiments were conducted from which practical cutting data were collected for comparisonwith predicted interface temperatures from a theoretical model. The test cuts involved dry machin-ing performed on plain steel tube (AISI 1020) with coated and un-coated controlled contact toolinserts. Analysis of the experimental results obtained and verified by the theoretical model showedthat an increase in the tool wear resulted in an increase in the cutting temperature. They concludedthat the temperature increases were primarily due to tool wear, which could be used to effectTCM during metal cutting.

Kitagawa et al. [39] presented what could be regarded as the most interesting experimentalapproach of cutting tool edge temperature measurement. A special purpose thermocouple wasconstructed from first principles using two carbide tip parts and a quartz glass for insulation. Bychanging the exit position of the circuitry wire involved, they claim that it was possible to measurethe interface temperature anywhere on the flank face of the tool (i.e. the cutting edges). To validatethis design, a series of test cuts were conducted using nominally sharp and artificially worn toolinserts. From the ensuing analyses and discussions, the dependency of temperature change rateson flank wear length for both interrupted and continuous cuttings were established. From theviewpoint of tool life estimation, observation of wear characteristics were in agreement with thewear rate equation:

dWstdL

Celqt (1)

where C and l are tooling and workpiece material constants; W and L are wear volume per unit


length of the tool insert face and wear distance, respectively; qt and st are the absolute temperatureand normal stress distribution respectively. However, beyond designing the thermocouple, theiraim was not to perform TCM but to establish the mechanisms principally responsible for flankwear rate.

Shaw [40] cited the complexity involved in any attempt to predict the mean tool face tempera-ture as it defies exact solutions. He proposed and evaluated an approximate solution based on theprinciple of moving heat source.

For practical applications such as on-line TCM, remote thermocouple sensing appear to be theonly worthy way to measure the workpiece-tool temperature but a direct measurement of the tool-tip or rake face temperature distribution cannot be obtained. Past attempts at measuring the cuttingedge temperature have proven exceptionally difficult due to lack of direct access to the cuttingzone. Boothroyd [31] proposed the use of thermocouple techniques in the workpiece-tool interface,and through such technique, the generated EMF at the junction is considered to be a measure ofthe mean temperature in that region. It does not however give any indication of temperaturedistribution. Most currently available remote thermocouple sensor instruments can only alloweither the cutting tool-workpiece interface or some other remote area temperature to that effectto be measured and not the tool tip temperature. This process parameter, though a suitable toolwear indicator and desirable, is extremely difficult to measure accurately for on-line applicationsas in TCM due to the inaccessibility to the cutting area rendering it in the opinion of theauthor inefficient.

6. Cutting forces (dynamic and static)

It has been widely established that variation in the cutting force can be correlated to tool wear[4,6,4154]. In practice, application and interpretation of this parameter has been diverse withmore effort concentrated on studying the dynamic characteristic of the cutting force signal andinterpreting its relation to tool wear levels. This can largely be attributed to the fact that forcebecomes important in worn tool conditions as a result of the variations produced due to frictionbetween cutting tool flank and the workpiece [43,46].

Existing force-based TCMSs typically operate independently of absolute force levels, measuringthe relative change of force that occurs as a new tool wears [55,56] or when it fractures [18].Experiments have shown that the three components of the cutting force (Fig. 3) respond differentlyto the various wear forms occurring on the tool. For example, the feed (Fx or Ff) force is insensi-tive to crater wear whereas the feed and radial (Fy or Fr) forces may be influenced more by toolwear than the main cutting (Fz or Ft) force [4,56].

Dimla [41,57] undertook an extensive and elaborate experimental investigation into the devel-opment of an on-line tool wear monitoring system for metal turning operations using cutting forcemeasurements fused with vibration signatures. An experimental test-bed consisting of a centre-lathe with a toolpost dynamometer was used to generate cutting force data. Interrupted test cutswere conducted using double-coated carbide-grade tool inserts (with integral groove chip-breakergeometry) of the P15 type (titanium nitride outer coating and aluminium oxide inner coating) andP25 (thick layer of aluminium oxide on top of a medium-size titanium carbide nitride) were usedto machine EN24 alloy steel. Measurements of flank, nose and notch wear lengths were made


Fig. 3. Cutting force components on a single point tool during turning.

immediately proceeding recording the on-line data. Static and dynamic entities of the sampledcutting force were extracted as the mean and oscillatory components respectively (see Fig. 4) andanalysed in time and frequency domains from which features sensitive to tool wear were identified.Time domain established the nature and level of static force magnitude change while frequencyanalysis demonstrated the dynamic force signatures response to cutting conditions as well asaccrued wear levels. The more succinct elements of tool wear as it gradually wore to catastrophicfailure were observed better in the frequency domain with certain frequencies correlating excep-tionally well to the dynamic force changes. Overall, flank and nose wear were established asbetter indicators of tool wear than notch wear.

Bayramoglu and Dungel [42] present a systematic investigation on the use of cutting forceratios in TCM for turning operations. The initial phase of the work concentrated on using extensiveexperimental data obtained under a wide variety of cutting conditions to investigate the effect

Fig. 4. Force amplitude vs time (fixed cutting conditions).


each parameter had on the force ratios (F1 to F4) using statistical methods. Having establishedand understood the effects the parameters had on the force ratios, they proceeded to conduct testsusing worn tools in order to examine how the force ratios could be used to monitor tool wear.Results showed that two of the force ratios to be particularly sensitive to the accrued flank wear,thus demonstrating potential applicability in TCM operations.

Lister [4] analysed the power spectra of dynamic cutting forces and reported an increase inpower level frequency bands as the tool wore. The results bore a strong indication of the depen-dency of dynamic force on cutting tool wear though not a general trend. As with many non-linearsystems, modelling of the process is very often used in order to capture and gain an understandingof process behaviour.

Ravindra et al. [6] developed a mathematical model for tool wear estimation that involvedcarrying out turning experiments from which the wear progression was studied and the cuttingforces modelled by a multiple regression analysis method. The experiments indicated that thewear trend propagated with increases in cutting speed, and the occurrence of inflections on thewear display curves indicated a predominant thermally controlled mechanism. An increase in themagnitude of the components of the triaxial cutting force was evident as the wear on the usedinserts increased. The wear-time and wear-force plots seemed to support their propositions, fromwhich they concluded that the experiments had provided vital evidence of a good correlationbetween flank wear and feed and radial forces.

Lee et al. [47] in their quest for an on-line TCMS developed a personal computer based fastFourier transform software to track the dynamic cutting force signal. Intermittent test cuts wereperformed on a Colchester mascot lathe at 100 mm intervals using two workpieces and a single-tool insert type (P30). The cutting forces and wear levels were measured and recorded. Subsequentanalyses of the obtained data showed that the feed and tangential dynamic force components borea good relationship to flank wear trend.

Marques and Mesquita [48] investigated the relationship between wear of sintered high-speedsteel cutting tools and the associated cutting forces. A wear-force model equation was establishedand experimentally verified. The theoretical models used considered the independent influence ofthe flank and crater wear, and also that of the indentation force (the blunt tool effect). Experimentaltests were conducted from which the forces were measured. The test types conducted consistedof short duration cuts to establish force-wear relationship, and longer duration cuts to observe theprogressive influence of wear on the forces. They reported a good correlation between experi-mental and theoretical results.

Kim and Lee [49] performed the modelling of dynamic cutting forces and experimental testdata collected and compared with the theoretical model predictions. Good agreement between thetheoretical limits of stability and experimental data was reported. However, in the low cuttingspeed range, deterioration between the model results and the experimental was observed, a fact,which they attributed to the effect of built-up edge. Grabec [58] and Khraisheh et al. [59] haveperformed similar modelling of the dynamic cutting force for chatter prediction.

Oraby and Hayhurst [50] developed a model for tool wear analysis in a turning operation byforce characteristics within the different phases of tool wear. Quantification of the developedmodel was performed by measurement of the variation of radial/vertical force component ratios,accomplished through force/wear inter-relationship formulated on simple 2-D plots. The experi-mental data for these models were obtained from machining test cuts of an alloy steel using a


Colchester mascot lathe and triple-coated carbide tool inserts. These tests were carried out follow-ing a central composite design strategy comprising of 24 tests at variable cutting speeds, feed-rate and depth of cut. For each test, the wear values (nose, flank and notch) were measured andrecorded. Time displays of the forces and wear types were deduced. The wear displays confirmedtheoretical predictions: a slow rise, then almost constant, and a fast climb to failure. The forcedisplays indicated a similar trend to the wear plots but exhibited a higher rate of change in thetertiary phase of the wear. The force displays were also reported to provide evidence of the natureof the relationship between force and wear but was affected by cutting conditions. The radialforce component was reported to be the most sensitive force to nose wear, with the feed forceand the radial force components affected by flank wear. To establish a universal rather than acase-based example of individual characteristics, they proposed a mathematical method to quanti-tatively formulate the wear-cutting force relation to the cutting speed (V), feed-rate (f) and depthof cut (d) to achieve repeatability and reliability for practical applications. The equation belowwas used and a statistical method adopted in order to obtain the constants ao, a, b and g.

T5a0Vafbd g (2)

Following the mathematical analysis, they concluded that there existed a strong correlationbetween force variation and wear progression. However their model was only applicable whenthe tool insert was in the primary or secondary phase of wear.

Yao et al. [51] and, Yao and Fang [52] investigated what they described as a comprehensiveTCMS which included the measurement of major and minor flank, crater, and nose wear basedon the analysis of dynamic cutting forces. Tool wear experiments were carried out on a Colchesterlathe at varying cutting conditions using only one tool insert and workpiece type and the threeorthogonal pre-processed force components recorded. Eight parameters describing the desiredwear features were selected. A dispersion analysis based on auto-regressive moving averages wasperformed. Two distinct frequency bands were obtained in all three axes associated with a wearrate mechanism of some sort: a low frequency band 0.51 Hz and a higher band 2.63.5 kHz.These trends were in agreement with the recorded wear values, thus, indispensable as a wearmonitoring system.

Shi and Ramalingam [53] conducted machining tests to investigate the feasibility of usingdifferent force components for on-line TCM. Through their investigation, they were able to corre-late the feed and cutting force components to flank wear length though it also became evidentthat these parameters were sensitive to changes in the cutting conditions. Experimental observationof the effect feed force to cutting force ratio showed sensitivity to flank wear but was insensitiveto process changes (cutting speed and depth of cut). By combining a prior fast tool fracturedetection scheme with their developed flank wear sensing method, it was possible to develop anew TCM strategy.

Lee et al. [27] pointed out that significant variation existed in the findings of various researcherswho attempted to correlate static cutting forces to tool wear. Some limitations of the static forceapproach such as disturbances caused by variation in workpiece material, depth of cut and tooledge geometry were put forward as evidence that approaches based upon static forces had notmade any significant gains. They proceeded to examine the nature and principal source of thedynamic force frequency and its correlation to flank wear. Experiments were carried out from


which both the static and dynamic forces, and the flank wear length were measured. A timedisplay of the forces and flank wear showed the three features to vary differently. The staticforces showed an initial rise as wear increased then a consistent fluctuation whereas the dynamiccutting forces exhibited a monotonic increase followed by a relatively sharp fall towards the endof tool life. Flank wear increased gradually in the earlier phase of cutting and before the toolfailed, its value monotonically increased. The drop in dynamic force magnitude in the latter stateof tool wear was attributed to blunt cutting edges and their subsequent deformation. A kind ofdamping effect results, because the edges are collapsing under plastic deformation thereby causinggreater damping of the tool oscillation, resulting in a decrease in the amplitude of the dynamicforce. The rapid increase in crater wear and a corresponding rise in the working rake angle accountfor the drop in dynamic force.

An inter-relationship between the tangential, feed and normal components of the dynamic cut-ting forces have been established in Dan and Mathew [5]. In their review, the dynamic forcesare reported to fluctuate with excursions to zero and then to higher magnitudes during cutting. Aconsequence of these excursions has been the onset of tool holder vibration whereby chatteringat high magnitudes often results.

The degree of variability apparent in the cutting forces in metal cutting processes require thatsome estimation of the static and dynamic cutting forces be known. The static cutting forces arenecessary to enable the surface cutting speed on the workpiece material to be kept within availablepower of the machine tool. Consequences of overloading the machine tool could lead to instancesof excessive power take-off causing the machine to stall or unacceptable deflections of workpieceand machine tool. The measurement of the static cutting forces or the fluctuation of its componentswould provide valuable information on the static behaviour of the cutting process. The nature ofthe cutting process is such that it can not be regarded as without deflections and instability. Thejoints and couplings of the machine tool and minute changes in the cutting conditions lead tofluctuations in the static force components. The cyclic variations of the static forces if not limitedleads to dimensional inaccuracy of the cutting operation as chatter results. It is difficult to predictthe conditions under which it occurs or select cutting conditions necessary to correct this phenom-enon. Therefore, to get an indication of the system fluctuations, the dynamic forces need to beknown and estimates obtained directly or calculated as the frequency dependent component ofthe static cutting forces as shown on Fig. 4. In the opinion of the author, static and dynamiccutting forces are vital in the development of a TCMS [60].

7. Vibration signatures (acceleration signals)

Vibrations are produced by cyclic variations in the dynamic components of the cutting forces.Usually, these vibrational motions start as small chatter responsible for the serrations on thefinished surface and chip thickness irregularities, and progress to what has come to be commonlytermed vibration. Mechanical vibrations generally result from periodic wave motions. The natureof the vibration signal arising from the metal cutting process is such that it incorporates facetsof free, forced, periodic and random types of vibration. Direct measurement of vibration is difficultto achieve because its determining characteristic feature, the vibration mode is frequency depen-dent. Hence, related parameters such as the rate at which dynamic forces change per unit time


(acceleration) are measured and the characteristics of the vibration derived from the patternsobtained.

Dimla [60] presents a detailed investigation of progressive tool wear results obtained during ametal turning operation. Dry cutting was conducted using P15 and P25 tool inserts from whichthe three principal components of the vibration signal was obtained. A tool life picture was con-structed in both time and frequency domains for each vibration component using three tool wearform measurements (flank, nose and notch). The accumulative sum total power of the spectrasignal was used to interpret the time domain characteristics while spectra and contour plotsdescribed the frequency characteristics. The accumulative power features were ineffective in grad-ual tool wear monitoring but reasonable at chipping/fracture detection. Spectra plots for the toolcharacteristics at first contact (i.e. when tool was new and therefore sharp) to those describingthe severely worn tool characteristics (or the occasional chipping/failure modes) were produced.Observation of the waterfall spectral plots showed the cutting direction (z-axis) components ofthe vibration signals to be the most sensitive and the x-axis generally the least sensitive to wearaccrued. Contour plots clarified observations from the spectra plots i.e. pinpointed the exact princi-pal frequency peaks. The two types of tool inserts used showed slight differences in their activetool lives as a result of the hot hardness effect of the coatings, but the investigation indicatedthat the amount of wear and its form was reliant more on the prevailing cutting conditions.

El-Wardany et al. [61] investigated the use of vibration signature characteristics in on-line drillwear monitoring and breakage. Vibration signature features sensitive to the tool wear were ident-ified in time (ratio of absolute mean value and kurtosis) and frequency (power spectra and cepstraratio) domains. Experimental results showed that the kurtosis values increased drastically withdrill breakage while frequency analysis revealed sharp peaks indicating drill breakage. By combin-ing both techniques, it was possible to devise an effective drill monitoring system.

Yao et al. [62] investigated detection and estimation of groove wear at the minor cutting edgeof the tool by monitoring vibration signatures. A high precision lathe on to which was attacheda miniature 3-D accelerometer, a single tool geometry and material combination with five cuttingconditions were utilized. Each cut was interrupted to take measurements of the wear mark values.A multivariate time series analysis was carried out on the recorded vibration signals using acombination of autoregressive moving averages and some explicit functions to obtain a dispersionof the signals auto-covariance, decomposed into the various eigenvalues and normalised to withina range of 21 and 1. The dispersion analysis showed that the thrust cutting force and vibrationwere sensitive to the length of groove wear with two peaks: one at a very low frequency ,200Hz and the other at a high frequency $10 kHz.

Dan and Mathew [5] employed a discrete modelling method called data dependent system tocorrelate vibration signals to cutting tool wear. The implementation of this method involved theisolation of the vibration signal deemed to be most sensitive to tool wear. Obtained results showedsome variation in the amount of vibration energy within a specific frequency band that was con-sistently observed regardless of cutting parameter. The application of spectral analysis to theacceleration signals revealed a linear relationship between the cutting speed and tool wear showingthat vibration signals were sensitive to tool wear.

Rotberg et al. [63,64] were interested in mechanical signature analysis (vibration) for tool stateprediction during interrupted cutting. In Rotberg et al. [63] the emphasis was on the milling toolentry and exit conditions. Face milling experiments were conducted and the ensuing flank and


crater wear measured. Detailed signal processing of the recorded signals was carried out. Wearcurves, average envelop at three points of tool life (sharp, part worn, worn) and spectral descrip-tions of the three wear phases were established. Inspection of the plots indicated that the vibrationsignal was a suitable indicator of tool wear as it demonstrated considerable change during toollife. Rotberg et al. [64] focused on tool wear monitoring using vibration as the principal signal.The analysis was performed using two basic models characterised by their signal features: lowfrequency and high frequency. Experiments were conducted to validate these models utilisingrecorded acceleration signal features. They concluded that their analysis showed certain peculiarand universally occurring features from the vibration signals that could be harnessed in developinga TCMS as it correlated well with tool wear.

Jiang et al. [65] using a purposely-built test-rig was able to investigate the effects of vibrationin the cutting and feed directions on the cutting tool. Using fixed cutting conditions, tests wereconducted utilising plane-faced P10 tool inserts with each cut lasting 10 minutes in a single pass.During the time of active tool engagement, the wear on the tool gradually increased until itcatastrophically failed. The recorded signals were post-processed and power spectra density (PSD)of the vibration signals produced. Three distinctive regions identified on the PSD could be dividedinto the frequency ranges: up to 100 Hz, 117510 Hz and 5101 000 Hz, from which they perfor-med an in-process method of tool wear monitoring based on frequency band energy analysis.They concluded that their experimental investigations provided sufficient evidence that vibrationsignals were sensitive to tool wear states.

The inter-relationship between vibration signals and the cutting forces determines the dynamicnature of the cutting process, making the utilisation of these process parameters attractive in thedevelopment of TCMSs [60]. The static behaviour is governed solely by the cutting forces andmomentum (or torsion of the tool holder). The dynamic behaviour on the other hand embodiesvibration and certain aspects of the dynamic cutting force. The combination of elements of thecutting forces and the vibration signals fused in developing a multiple sensor-based TCMS wouldprove indispensable in the shop floor.

8. Miscellaneous sensors and methods

Generally speaking, other varieties of sensors have been employed in various attempts at toolwear prediction, monitoring or process parameter measurements in a metal cutting process. Thesemethods fall principally into the following categories:

O optical methods,O stress/strain measurement,O methods based on measuring the workpiece dimension,O spindle motor current/torque/power,O surface finish quality measurement, andO ultrasonic methods


8.1. Optical methods

Martin et al. [16] summarised in a review that, amongst the optical methods to have beenproposed for tool wear measurements were the following:

O laser measurement of the surface finish,O photodiodes used in reflected light measurement from the cutting edge, andO fibre optic photocell for reflectance measurement of worn and unworn areas of the tool flank.

Clearly, none of these methods are applicable in a TCMS scenario as in-process tool wearmeasurement is more difficult to achieve. Other attempts based on optical techniques for toolwear sensing are listed in Ref. [66].

Determining tool wear from processed images of the cutting tool has been pursued for overthree decades now. Essentially, an image of the tool is captured and displayed on a TV screen,and then analysed to provide information on wear pattern or quantity of wear accrued. Kuradaand Bradley [13] present a review of the basic principle, instrumentation and various image-processing schemes involved in the development of a vision based TCMS. Oguamanam et al. [67],Du et al. [68] and Cuppini et al. [69] carried out implementation of this method. In Oguamanam etal. the vision machine system is used for monitoring purposes. For any given image, five featuresof the image were extracted and used in classifying the state of the tool (good, worn or broken).Extensive computer image processing algorithms were developed to capture the image (off-line),segment it and then identify the tool state from the digitised image. Du et al. on the other handattempted to devise an on-line optical method utilising an off-the-shelf optical tool conditionsensor. The sensor was reputed for capturing the image of the tool edges and converting theimage into pixels for computer processing. The obtained tool profile could then be filtered andrepresented by an array of numbers for comparison to that of the master tool profile (i.e. capturedwhen new and sharp). Experimental verifications showed that repeatability was a major problemto the success of the method and at best, only average values of the measurements were obtainable.The aforementioned techniques are expensive, inflexible and cannot truly be applied on-line.

Wong et al. [70] devised a vision-based TCMS using laser scatter pattern of reflected laser rayin the roughing to near-finishing range. The system consisted of a laser source focused on thefinished workpiece such that its reflected ray is captured through a digital camera. The recordedimage was processed and characterised using the mean and standard deviation of the scatter patternand the intensity of the optical parameters, and their distribution correlated to the surface rough-ness. The deduced surface roughness in turn was then related to the ensuing state of the cutting-tool wear. They concluded that it was very difficult to determine tool wear by observing machinedsurface roughness. Nevertheless, the study showed a good correlation between tool wear andintensity of the scattered light pattern.

Shiraishi [15] through many years of research and experience cited that optical based methodsfor TCM are riddled with high inaccuracies and are therefore unreliable.

8.2. Stress/strain measurements

Noori-Khajavi and Komanduri [71] used amongst other sensors, strain sensors in their studyof the correlation of process parameters to drill wear. The recorded signals were analysed in both


time and frequency domains but meaningful correlation of the drill wear could only be achievedin the frequency domain. The area under the x-axis PSD for the strain sensor was found tocorrelate well to drill wear and they proceeded to base their study on this assumption.

Zhou et al. [72] proposed to monitor the stresses acting in a cutting edge during a machiningprocess in order to predict tool spontaneous failure. However, because of the impracticalities ofin-process measurement of the cutting tool stress, an on-line stress estimation based on in-processcutting forces, load functions and the cutting conditions were used instead. They designed andimplemented a real-time TCMS based on VME computer system and real time kernel. This incor-porated a fast data acquisition unit that analysed the stresses from force measurements. By moni-toring the risk factor defined as a ratio of the instantaneous stresses, they reported that it waspossible to predict spontaneous failures.

Lee et al. [54] proposed another method based on stress analysis of three-dimensional loading.They combined FEA and detailed stress analysis of the cutting edges and tips of sharp and worntools from which they concluded that it was possible to predict the mode and location of tool fail-ure.

8.3. Workpiece dimension

El-Gomayel and Bregger [73] proposed a method for tool wear monitoring based on measure-ments of workpiece deviation. They employed two electromagnetic probes on opposite sides ofthe workpiece such that electromagnetic waves could flow from the probe to the metal allowingaccurate measurements of the workpiece diameter. Increases in the workpiece diameter were usedas a tool wear criterion requiring elaborate calibration procedure. An experimental verification ofthe method suggested a relationship between the accumulated flank wear length (manuallymeasured) and accumulative voltage differences from the two differential probes. The resultshowever were affected by several factors such as vibrations, deflections and misalignment. Theyconcluded that though their model could measure minute tool wear, it could not quantify it (i.e.distinguish nose wear from flank wear).

8.4. Electric motor current/power measurement

Zhang et al. [74] used a Hall effect sensor to measure the current supplied to the spindle motordrive of a vertical NC miller together with the cutting forces. The relationship between the meas-ured motor current and the milling torque was developed by modelling the acquired motor currentand torque. Computer algorithms were developed to track the waveforms, rate of peak changeand the relative eccentricities of the modelled relation. They concluded that cutter breakages werereliably diagnosed with motor current measurements than force/torque measurements.

Constantinides and Bennett [75] obtained the spindle motor power from a vertical millingmachine and attempted to analyse the PSD, the moving average, running mean and the accumulat-ive sum power to estimate tool wear. They concluded that spectral energy fluctuations of thespindle motor power were linearly related to the tool wear rate but was also affected by thecutting conditions and tool geometry.

Shaft power, cutting forces, torque and motor current are all related to each other, originatingfrom, and depending entirely on one another. It suffices therefore to measure just one of these


parameters as demonstrated by Rangwala and Dornfeld [76,77] who dropped the electric currentin preference to the cutting forces.

8.5. Magnetism

Jetley and Gollajesse [78] proposed the magnetisation of tool inserts and then, monitoring themagnetic field flux reduction as the tool wore. A preliminary investigative study involving theuse of magnetised drills was devised and implemented in order to validate their methodology.They concluded that it was possible to accurately predict the end of tool life or fracture on-lineby observing the magnetic flux, and the system was cost effective with potential for implemen-tation in most metal cutting environments.

8.6. Ultrasonic methods

Abu-Zahra and Nayfeh [79] developed a normalised ultrasonic signal based method for in-process tool wear monitoring for turning operations. A purposely-designed tool-holder accommo-dating the ultrasonic transducer was used to measure the gradual flank and nose wear forms.Analysis involved comparing recorded waveforms for fresh (no wear) and used tool inserts withaccrued wear. Inspection of results showed a good correlation between the measured wear valuesand the absolute change in the ultrasonic waves. As the system developed required a specialisedtool-holder, they conceded that commercialisation would require a more versatile tool-transducercoupling to accommodate the various tool-holder and insert geometry.

9. TCMS developmenta perspective on sensor signals

The accuracy and tolerance on a finished workpiece can be determined by deviations at thecutting point from the required working movements between the tool and workpiece. These devi-ations arise primarily from static and dynamic force deformations in addition to geometric andkinematic errors of machine frames, beds, spindles and slides. Present day instrumentation havemade it possible for the behaviour of these process parameters to be monitored. In as much asforce types are concerned, they can only be viewed as an approximation because there is a lackof knowledge as to the damping and stiffness conditions of workpiece-tool area, joints and coup-lings [80].

Process parameter selection for the development of a TCMS has to take into consideration, therobustness, reliability and applicability of the sensor signals but conform to the following criteria:

O easy to measure,O have a high signal to noise ratio, though could be improved through application of RC filters,O consistent in wear sensitivity, andO require minimal peripheral instruments for harnessing.

Different machining operations require varying cutting speeds, while cutting may be continuous(several hours) or abrupt (few seconds). Poor cutting conditions give rise to increased forces,


energies, tool temperature thus resulting in higher tool wear rate [31]. Cutting at high speeds andfeed-rates causes an increase in temperature on the tool face resulting in increased crater wear.At low speeds, built up edges are formed. The effects of the cutting conditions on the static anddynamic forces, and vibration signatures have a wide margin of variability. It is difficult to predictexactly what the magnitude of the cutting forces would be at specific cutting speeds. Therefore,it is plausible to utilise the cutting conditions as significant variations in the sensor signals arisefrom changes in these parameters.

10. Sensor fusionsynergy of signal integration

Sensors for engineering process monitoring are generally designed to measure a desired para-meter (e.g., accelerometers) and then correlated to the process of interest (e.g., vibrationsignatures). During measurement, the measurand is such that the determining principal signalcomponent has distortion of some kind (noise). If the signal to noise ratio is high then the measur-and yields meaningful correlation to the desired quantity, but if it is low the converse is true. Apromising approach in engineering applications has been the utilisation of more than one sensorsignal from different sources to detect the same parameter commonly termed sensor fusion. Sensorfusion combines the noise dominant probably mediocre facets (similar or contradictory) of theindividual signals to yield a better outcome [81].

The choice of a suitable sensor type as well as the point of application (sensor location) areinextricably linked with a suitable location being where the specified signal has the highest con-centration and best reproducibility. However, certain limitations do exist due to the structure ofthe machine (i.e. bearings, lubrication and linkages). Information from a variety of different sen-sors therefore has to be collected and these signals of varying reliability integrated. Sensor fusionserves the following purposes:

O enhances the richness of the underlying wear level information contained in each signal, andO increases the reliability of the monitoring process as loss of sensitivity in one signal could be

offset by that from another.

The use of a single sensor signal in the development of a TCMS fails to recognise the complexand diverse nature of the cutting process and such TCM models are often less robust, unreliableand generally incapable of total TCMs ability to recognise incipient, partial, complete or cata-strophic tool failure. The adoption of feature space dimensioning as a means of increasing thenumber of signals by identifying and extracting more than one wear sensitive feature from asingle signal is debatable. Primarily, multi-feature extraction from a single signal does not consti-tute multi-sensing. In summary the following observations has been made on feature space dimen-sioning [82]:

O representing statistically independent probability distributions in just a single signal, andO minimal mutual information acquired compared to features extracted from more than one signal.

When sensors are fused, the task requirements would normally determine the structure of the


ensuing fused data. Tanner and Loh [83] take the view that three possible outcomes could resultfrom data fusion: competitive, complementary, and uniquely independent outputs. Competitivefusion is when each fused sensor data increases in value or repudiates with all other sensor data.The competitive method of data fusion would be beneficial in a multivariate TCMS because whenthe various sensor signals are fused, they either reinforce or contradict with each individual sensorsignal. For example, if all the sensor signals are geared towards measuring the same parametersuch as tool wear, then if several report a similar behaviour pattern as the tool wear level increasesthen an increase in confidence in the parameter of interest (wear) is developed. If the sensorsreport dissimilar values, a reason for the discrepancy has to be established. Independent datafusion is simply when one sensor is used to sense a particular feature. Clearly this represents asituation where the data is not fused at all, hence a degenerate method. Complementary sensorfusion extracts relative differences from sensors to enhance the advantages, while covering thedisadvantages of the individual sensors.

The decision as to which sensor signal and how many to use in a TCMS is indeed a difficultone to make. One consideration is the cost of successfully tapping any chosen sensor signal.Another is how practical it would be to obtain the chosen signal. If these two questions can beanswered, then choosing the sensor signals is made much easier. However, care and considerationmust be taken to ensure that the measuring apparatus for the chosen sensor signal does not disturbor interfere with the machining process, and that the employed instrumentation is suitable for usein such an environment, i.e. adequate shielding. In case of interference with the machining pro-cesses, for example an exaggerated tool overhang, effects of such an exaggeration on the dynamicsignals ought to be known.

11. Conclusions

The evaluation of the suitability and sensitivity of the most widely used process parameters totool wear, and their potential applicability for successful on-line TCM, based on ease of harnessingand reliability has been conducted. Clearly, cutting forces (static and dynamic) and vibration(acceleration) are considered to be the most widely applicable parameters. Advances and increasedsophistication in instrumentation technology employed for measuring these parameters make themviable, practical, cost effective, robust, easy to mount and have the quick response time neededto indicate changes for on-line monitoring.

The survey also suggests that most developed TCMSs have not been successful implementedprimarily because inadequate sensor information and machining process models have been util-ised, which did not satisfactorily reflect the process complexity. It is sufficient to develop a systemthat can be employed in the systematic monitoring and diagnosis of cutting tool conditions suchas wear levels, chipping and/or fracture, by learning and identifying tool states from experience,thus recognise and localise tool failure scenarios. A better TCMS needs to accommodate thecomplex and diverse nature of the metal cutting operation.


Acknowledgements

The author would like to thank the University of Wolverhampton for providing the funds andfacilities to carry out this study. Dr Paul Lister and Dr Nigel Leighton are thanked for theirguidance and support. Dr Charles Elad is thanked for commenting on an earlier draft of this paperand Maria Giovanna Lo Porto-Dimla for assistance in the preparation of the manuscripts.

References

[1] M.C. Shaw, Metal Cutting Principles, Clarendon Press, Oxford, 1984 Chapter 3.[2] J.A. Schey, Introduction to Manufacturing Processes (second ed.), McGrawHill Book Company, London, 1987

Chapter 8.[3] G. Byrne, D. Dornfeld, I. Inasaki, G. Ketteler, W. Konig, R. Teti, Tool condition monitoring (TCM)the status

of research and industrial application, Annals of CIRP 44 (2) (1995) 541567.[4] P.M. Lister. On-line measurement of tool wear. Ph.D. thesis, Manufacturing and Machine Tools Division, Depart-

ment of Mechanical Engineering, UMIST, Manchester, UK, 1993.[5] L. Dan, J. Mathew, Tool wear and failure monitoring techniques for turninga review, International Journal of

Machine Tools and Manufacture 30 (4) (1990) 579598.[6] H.V. Ravindra, Y.G. Srinivasa, R. Krishnamurthy, Modelling of tool wear based on cutting forces in turning,

Wear 169 (1993) 2532.[7] G. Warnecke, A. Jenewein, A. Reinfelder, Tool monitoring based on process identification, monitoring and control

for manufacturing processes, in: Winter Annual Meeting of the ASME, Dallas, TX, USA, 1990, pp. 4355.[8] A. Novak, H. Wiklund, Reliability of tool-wear monitoring, Annals of the CIRP 42 (1) (1992) 6366.[9] G.S. Choi, Z.X. Wang, D.A. Dornfeld, K. Tsujino, Development of an intelligent on-line tool wear monitoring

system for turning operations, in: Proc. USAJapan Symposium on Flexible Automation, A Pacific Rim Confer-ence ISCIE Kyoto, 1990, pp. 683690.

[10] L.I. Burke, Automated identification of tool wear states in machining processes: an application of self-organisingneural networks. Ph.D. thesis, Department of Industrial Engineering and Operations Research, UC at Berkeley,USA, 1989.

[11] S. Rangwala, Machining process characterisation and intelligent tool condition monitoring using acoustic emissionsignal analysis. PhD thesis, Mechanical Engineering Department, UC at Berkeley, USA, 1988.

[12] P.W. Prickett, C. Johns, An overview of approaches to end milling tool monitoring, International Journal ofMachine Tools and Manufacture 39 (1) (1999) 105122.

[13] S. Kurada, C. Bradley, A review of machine vision sensors for tool condition monitoring, Computers in Industry34 (1) (1997) 5572.

[14] K.F. Martin, A review by discussion of condition monitoring and fault diagnosis in machine tools, InternationalJournal of Machine Tools and Manufacture 34 (4) (1994) 527551.

[15] M. Shiraishi, Scope of in-process measurement, monitoring and control techniques in machining processesPart1: In-process techniques for tools, Precision Engineering 10 (4) (1988) 179189.

[16] K.F. Martin, J.A. Brandon, R.I. Grosvenor, A. Owen, A comparison of in-process tool wear measurement methodsin turning, in: Proceedings International 26th MATADOR Conference, Manchester, UK, 1986, pp. 289296.

[17] R. Teti, A review of tool condition monitoring literature database, Annals of the CIRP 44 (2) (1995) 659666.[18] D. Choi, W.T. Kwon, C.N. Chu, Real-time monitoring of tool fracture in turning using sensor fusion, International

Journal of Advanced Manufacturing Technology 15 (5) (1999) 305310.[19] K. Jemielniak, O. Otman, Tool failure detection based on analysis of accoustic emission signals, Journal of

Material Processing Technology 76 (1998) 192197.[20] S. Kakade, L. Vijayaraghavan, R. Krishnamurthy, In-process tool wear and chip-form monitoring in face milling

operation using acoustic emission, Journal of Material Processing Technology 44 (1994) 207214.[21] S.X. Zheng, R. McBride, J.S. Barton, J.D.C. Jones, K.F. Hale, B.E. Jones, Intrinsic optical fibre sensor for monitor-

ing acoustic emission, Sensors and Actuators, A: Physical 31 (1992) 110114.


[22] W. Konig, K. Kutzner, U. Schehl, Tool monitoring of small drills with acoustic emission, International Journalof Machine Tools and Manufacture 32 (4) (1992) 487493.

[23] T. Blum, I. Inasaki, A study on acoustic emission from the orthogonal cutting process, ASME Trans. Journal ofEngineering for Industry 112 (3) (1990) 203211.

[24] T. Moriwaki, M. Tobito, A new approach to automatic detection of life of coated tool based on acoustic emissionmeasurement, ASME Trans. Journal of Engineering for Industry 112 (3) (1990) 212218.

[25] J. Roget, P. Souquet, N. Gsib, Application of acoustic emission to the automatic monitoring of tool conditionduring machining, ASNDT Materials Evaluation 46 (1988) 225229.

[26] I.N. Tansel, C. McLaughlin, On-line monitoring of tool breakage with unsupervised neural network, Trans.NAMRI/SME (1991) 364370.

[27] L.C. Lee, K.S. Lee, C.S. Gan, On the correlation between dynamic cutting force and tool wear, InternationalJournal of Machine Tools and Manufacture 29 (3) (1989) 295303.

[28] D. Yan, T.I. El-Wardany, M.A. Elbestawi, A multi-sensor strategy for tool failure detection in milling, Inter-national Journal of Machine Tools and Manufacture 35 (3) (1995) 383398.

[29] D.A. Dornfeld, Application of acoustic emission techniques in manufacturing, NDT and E International 25 (6)(1992) 259269.

[30] D.A. Dornfeld, In process recognition of cutting states, JSME International Journal, Series C: Dynamics Control37 (4) (1994) 638650.

[31] G. Boothroyd, Fundamentals of Metal Machining and Machine Tools, McGraw-Hill, New York, 1975.[32] E.M. Trent, Metal Cutting, Butterworths, London, 1977.[33] M. Sarwar, J. Li, R. Penlington, W. Ahmed, Development of thermal imaging systems for metal cutting appli-

cations, in: Proceedings Advanced Manufacturing Processes Systems and Technologies Conference (AMPST 96),Bradford, UK, 1996, pp. 361369.

[34] J. Lin, Inverse estimation of the tool-work interface temperature in end milling, International Journal of MachineTools and Manufacture 35 (5) (1995) 751760.

[35] R. Radulescu, S.G. Kapoor, An analytical model for prediction of tool temperature fields during continuous andinterrupted cutting, ASME Trans. Journal of Engineering for Industry 116 (2) (1994) 135143.

[36] S. Raman, A. Shaikh, P.H. Cohen, A mathematical model for tool temperature sensing, ASME ComputationalMethods in Materials Processing, PED 61 (1992) 181193.

[37] D.A. Stephenson, A. Ali, Tool temperatures in interrupted metal cutting, Winter Annual Meeting of the ASME,PED 43 (1990) 261281.

[38] J.G. Chow, P.K. Wright, On-line estimation of tool/chip interface temperatures for a turning operation, ASMETrans. Journal of Engineering for Industry 110 (1988) 5664.

[39] T. Kitagawa, K. Maekawa, T. Shirakashi, E. Usui, Analytical prediction of flank wear of carbide tools in turningplain carbon steels (Part 1), Bull. Japan Soc. Precision Engineering 22 (4) (1988) 263269.

[40] M.C. Shaw, Temperature in cutting, ASME Proceedings Symposium on Thermal Aspects in Manufacturing 30(1988) 133143.

[41] D.E. Dimla Snr, Tool wear monitoring using cutting force measurements, in: 15th NCMR: Advances in Manufac-turing Technology XIII, University of Bath, 68 September, 1999, pp. 3337.

[42] M. Bayramoglu, U . Dungel, A systematic investigation on the use of force ratios in tool condition monitoringfor turning operations, Trans. Institute of Measurement and Control 20 (2) (1998) 9297.

[43] T.J. Ko, D.W. Cho, Cutting state monitoring in milling by a neural network, International Journal of MachineTools and Manufacture 34 (5) (1994) 659676.

[44] S. Purushothaman, Y.G. Srinivasa, A back-propagation algorithm applied to tool wear monitoring, InternationalJournal of Machine Tools and Manufacture 34 (5) (1994) 625631.

[45] Y.S. Tarng, Y.W. Hseih, S.T. Hwang, Sensing tool breakage in face milling with a neural network, InternationalJournal of Machine Tools and Manufacture 34 (3) (1994) 341350.

[46] D.A. Dornfeld, Neural network sensor fusion for tool condition monitoring, Annals of the CIRP 39 (1) (1990)101105.

[47] K.S. Lee, L.C. Lee, S.C. Teo, On-line tool wear monitoring using a PC, Journal of Material Processing Technology29 (1992) 313.


[48] M.J.M. Marques, R.M.D. Mesquita, Monitoring the wear of sintered high-speed steel tools, Journal of MaterialProcessing Technology 25 (1991) 195213.

[49] J.S. Kim, B.H. Lee, An analytical model of dynamic cutting forces in chatter vibration, International Journal ofMachine Tools and Manufacture 31 (3) (1991) 371381.

[50] S.E. Oraby, D.R. Hayhurst, Development of models for tool wear force relationships in metal cutting, InternationalJournal of Machine Tools and Manufacture, 33 (2) (1991) 125138.

[51] Y. Yao, X.D. Fang, G. Arndt, Comprehensive tool wear estimation in finish-machining via multivariate time-series analysis of 3-D cutting forces, Annals of the CIRP 39 (1) (1990) 5760.

[52] Y. Yao, X.D. Fang, Modelling of multivariate time series for tool wear estimation in finish-turning, InternationalJournal of Machine Tools and Manufacture 32 (4) (1992) 495508.

[53] T. Shi, S. Ramalingam, Real-time flank wear sensing, Winter Annual Meeting of the ASME, PED 43 (1990)157170.

[54] L.C. Lee, K.Y. Lam, X.D. Liu, Characterisation of tool wear and failure, Journal of Materials Processing Tech-nology 40 (1994) 143153.

[55] S.V.T. Elanayar, Y.C. Shin, S. Kumara, Machining condition monitoring for automation using neural networks,Monitoring and Control of Manufacturing Processes, PED, 44 (1990) 85100 Winter Annual Meeting of theASME.

[56] L. Gould, Sensing tool and drive element conditions in machine tools, Sensor (1988) 513.[57] D.E. Dimla Snr, P.M. Lister, On-line metal cutting tool condition monitoringI: Force and vibration analyses,

International Journal of Machine Tools and Manufacture 40 (5) (2000) 739768.[58] I. Grabec, Chaotic dynamics of the cutting process, International Journal of Machine Tools and Manufacture 28

(1) (1988) 1932.[59] M.K. Khraisheh, C. Pezeshki, A.E. Bayoumi, Time series based analysis for primary chatter in metal cutting,

Journal of Sound and Vibration 180 (1) (1995) 6787.[60] D.E. Dimla Snr., Multivariate tool condition monitoring in a metal cutting operation using neural networks. Ph.D.

thesis, School of Engineering and the Built Environment, The University of Wolverhampton, UK, 1998.[61] T.I. El-Wardany, D. Gao, M.A. Elbestawi, Tool condition monitoring in drilling using vibration signature analysis,

International Journal of Machine Tools and Manufacture 36 (6) (1996) 687711.[62] Y. Yao, X.D. Fang, G. Arndt, On-line estimation of groove wear in the minor cutting edge for finish machining,

Annals of the CIRP 40 (1) (1991) 4144.[63] J. Rotberg, S. Braun, E. Lenz, Mechanical signature analysis in interrupted cutting, Annals of the CIRP 36 (1)

(1987) 249252.[64] J. Rotberg, S. Braun, E. Lenz, Vibration generation models for cutting tool monitoring, Diagnostics, Vehicle

Dynamics and Special TopicsASME Design Engineering Division 18 (5) (1989) 16.[65] C.Y. Jiang, Y.Z. Zhang, H.J. Xu, In-process monitoring of tool wear stage by the frequency band-energy method,

Annals of the CIRP 36 (1) (1987) 4548.[66] P.M. Lister, G. Barrow, Tool condition monitoring systems, in: Proceedings International 26th MATADOR Con-

ference, Manchester, UK, 1986, pp. 271288.[67] D.C.D. Oguamanam, H. Raafat, S.M. Taboun, A machine vision system for wear monitoring and breakage detec-

tion of single-point cutting tools, Computers in Industrial Engineering 26 (3) (1994) 575598.[68] R. Du, B. Zhang, W. Hungerford, T. Pryor, Tool condition monitoring and compensation in finish turning using

optical sensor, ASME Symposium Mechatronics, PED 63 (1993) 245251.[69] D. Cuppini, G. DEnrrico, G. Rutelli, Tool image processing with application to unmanned metal-cutting, a com-

puter vision system for wear sensing and failure detection, SPIE 701 (1986) 416422.[70] Y.S. Wong, A.Y.C. Nee, X.Q. Li, C. Reisdorf, Tool condition monitoring using laser scatter pattern, Journal of

Materials Processing Technology 63 (1997) 205210.[71] A. Noori-Khajavi, R. Komanduri, Frequency and time domain analyses of sensor signals in drillingI. Correlation

with drill wear, International Journal of Machine Tools and Manufacture 35 (6) (1995) 775793.[72] J.M. Zhou, M. Andersson, J.E. Stahl, A system for monitoring cutting tool spontaneous failure based on stress

estimation, Journal of Materials Processing Technology 48 (1995) 231.[73] J.I. El Gomayel, K.D. Bregger, On-line tool wear sensing for turning operations, ASME Trans. Journal of Engin-

eering for Industry 108 (1986) 4447.


[74] D. Zhang, S. Dai, Y. Han, D. Chen, On-line monitoring of tool breakage using spindle current in milling, in: 1stAsiaPacific and 2nd JapanChina International Conference Progress of Cutting and Grinding, Shanghai, China,September, 1994, pp. 270276.

[75] N. Constantinides, S. Bennett, An investigation of methods for on-line estimation of tool wear, InternationalJournal of Machine Tools and Manufacture 27 (2) (1987) 225237.

[76] S. Rangwala, D. Dornfeld, Integration of sensors via neural networks for detection of tool wear states, WinterAnnual Meetings of the ASME, PED 25 (1987) 109120.

[77] S. Rangwala, D. Dornfeld, Sensor integration using neural networks for intelligent tool condition monitoring,ASME Trans. Journal of Engineering for Industry 112 (3) (1990) 219228.

[78] S.K. Jetley, A. Gollajesse, Measuring tool wear using magnetism, in: Proceedings: JapanUSA Symposium Flex-ible AutomationA Pacific Rim Conference, July 1118, Kobe, Japan, 1994, pp. 345347.

[79] N.H. Abu-Zahra, T.H. Nayfeh, Calibrated method for ultrasonic on-line monitoring of gradual wear during turningoperations, International Journal of Machine Tools and Manufacture 37 (10) (1997) 14751484.

[80] M. Weck, Metrological Analysis and Performance Tests, in: Handbook of Machine Tools (fourth ed.), Vol. 4,John Wiley and Sons, Chichester, 1984.

[81] P. Holmbom, O. Pedersen, B. Sandell, A. Lauber, Fusing sensor systems: promises and problems, Sensor Reviews9 (3) (1989) 143152.

[82] L. Parra, G. Deco, S. Miesbach, Statistical independence and novelty detection with information preserving non-linear maps, Neural Computation 8 (1996) 260269.

[83] R. Tanner, N.K. Loh, A taxonomy of multi-sensor fusion, Journal of Manufacturing Systems 11 (5) (1994)314325.

Documents

1-s2.0-S0890695599001224-main