A new versatile in-process monitoring system for milling

A new versatile in-process monitoring system for milling
of 10
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
     U   N  C  O   R   R   E  C   T   E   D    P   R  O  O   F International Journal of Machine Tools & Manufacture  ]  ( ]]]] )  ]]]  –  ]]] A new versatile in-process monitoring system for milling Mathieu Ritou, Sebastien Garnier  , Benoit Furet, Jean-Yves Hascoet Institut de Recherche en Communications et Cybernetique de Nantes (IRCCyN), UMR CNRS 6597, 1 rue de la Noe, BP92101, 44321 Nantes, Cedex 03,France Received 10 October 2005; received in revised form 19 December 2005; accepted 3 January 2006 Abstract Tool condition monitoring (TCM) systems can improve productivity and ensure workpiece quality, yet, there is a lack of reliable TCMsolutions for small-batch or one-off manufacturing of industrial parts. TCM methods which include the characteristics of the cut seem tobe particularly suitable for these demanding applications. In the first section of this paper, three process-based indicators have beenretrieved from literature dealing with TCM. They are analysed using a cutting force model and experiments are carried out in industrialconditions. Specific transient cuttings encountered during the machining of the test part reveal the indicators to be unreliable.Consequently, in the second section, a versatile in-process monitoring method is suggested. Based on experiments carried out under arange of different cutting conditions, an adequate indicator is proposed: the relative radial eccentricity of the cutters is estimated at eachinstant and characterizes the tool state. It is then compared with the previous tool state in order to detect cutter breakage or chipping.Lastly, the new approach is shown to be reliable when implemented during the machining of the test part. r 2006 Published by Elsevier Ltd. Keywords:  Milling; Monitoring; Cutting force model; Cutter breakage; Small-batch manufacturing 1. Introduction Machining problems, such as cutter breakage, excessivewear, chatter and collision, impede production consistencyand quality. Loss can be significant, particularly when highadded value parts like moulds and dies or aeronauticalmotor and structure parts are machined. They aremanufactured in small batches or one-off productions.Thus, their machining should be monitored as soon as thefirst part is produced. Loss due to disturbance could beprevented, or at least limited, using an in-process toolcondition monitoring (TCM) system. An accurate andreliable TCM system could increase savings between 10%and 40% [1]. However, there is a lack of reliable TCM solutions for small-batch manufacturing of industrial partsin milling [2]; the subject of this paper. Part machining timecan last for several days, without stopping during the off-duty hours of the operators. Thus, to prevent machiningfrom being stopped too often, no false alarms can beallowed. The TCM system must, therefore, be completelyreliable [3] as soon as the first part is machined.Certain information is needed to evaluate processconditions. This is provided by one or several sensorswhich are placed in the machine tool. Various methodsthen allow analysis and decision-making, Fig. 1. The teach-in method is used for mass production and mostcommercial TCM systems are based on this principle [4].It requires the machining of a few parts (trial cuts) tomeasure a reference signal. Thresholds are then set oneither side of the signal, based on heuristic knowledge [5].As monitoring trial cuts is impossible, this is incompatiblewith small-batch manufacturing [6].It was suggested that the measured reference signal bereplaced with an estimated one, using a cutting force model[7,8]. This enabled us to monitor the machining of the firstpart. The relevancy of this method relies on the accuracy of the force model. Till now, average cutting force values perspindle revolution were estimated and the gap betweenmeasured and estimated forces was significant. Further-more, in milling, it is assumed that if a tooth is chipped orbroken, this tooth removes a smaller volume of material   ARTICLE IN PRESS 1357911131517192123252729313335373941434547495153555759616365676971737577798183 3B2v8 : 06a = w ð Dec52003 Þ : 51c þ model  MTM : 1857  Prod : Type:FTPpp : 1 2 10 ð col : fig : :4 ; 5 ; 8 ; 9 ; 10 ; 11 Þ ED:R : JothiL : PAGN:bhagya SCAN:v4soft 0890-6955/$-see front matter r 2006 Published by Elsevier Ltd.doi:10.1016/j.ijmachtools.2006.01.001  Corresponding author. Tel.: +33240376954; fax: +33240376930. E-mail address: (S. Garnier).     U   N  C  O   R   R   E  C   T   E   D    P   R  O  O   F than before breakage and the following one a largervolume [9]. Therefore, the cutting force per tooth period should be considered, rather than per spindle revolution.This method is consequently not suitable for cutterbreakage detection.The milling forces waveform has led various authors tofeature extraction methods from the force signals of anincident. These methods are generic and applicable fromthe production of the first part. Many studies have beencarried out on artificial intelligence (AI), e.g. neuralnetworks, fuzzy logic. Neural networks or hybrid AIsystems are viable for TCM [10]. Networks are trained using trial cuts. This then leads to the generalizationproblem: under other cutting conditions, the neuralnetwork may be unreliable [11]. Users may have to trainthe network again in order to monitor the machining of anew part [12].Other authors suggested specific feature extraction of anincident using digital signal processing methods, e.g.autoregressive filter [13,14], synchronized averaging [15,16], wavelet transform [17,18]. However, if basic process knowledge is ignored, it is harder to differentiatetool breakage from the effects of tool runout or transientcutting. Indeed, adequate force models have been devel-oped [6] and geometric, kinematic and mechanisticcharacteristics of the cutting process are, or couldpotentially be, controlled during milling operations [11].They can be used to improve or simplify the signal-processing method and increase its reliability.Little literature has been published on process-basedsignal processing, where characteristics of the cut wereincluded (Table 1). This method is particularly suitable forthe monitoring of small-batch manufacturing of industrialparts.At each spindle revolution, a force value is extracted foreach tooth, before calculating the indicator value (Fig. 2).Altintas and Yellowley [6] used the first and the seconddifferences of mean forces, between adjacent teeth. It wasshown that, it was impossible to distinguish tool breakagefrom cutter runout [19]. Lee et al. [19] added a new indicator to the first-order autoregressive filter proposed byAltintas [14]: the relative variation of average tooth force, between two consecutive revolutions. They introduced theidea that each tooth can be monitored individually. But thecriteria [6,19] were affected by changes in cutting condi-tions and therefore tool breakage detection was unreliable[20]. Kim and Chu [20] proposed the tool failure index (TFI), which is the ratio of peak-to-valley cutting forcesbetween adjacent teeth, divided by its own past averageratio. The average ratio is intended to prevent the TFIfrom cutting conditions changes. In this paper, the TFI isretained for further experiments to examine whether it isunaffected by changes in cutting conditions. Lastly,Deyuan et al. [21] proposed two indicators: the peak rateKm is the ratio of the difference to the sum between peakforces of adjacent teeth. The relative eccentricity rate Bm issimilar to the ratio of tooth eccentricity to maximum chipthickness. The authors specified that the indicators wereindependent of cutting conditions. As for TFI, the twoindicators have been retained.Generally, few experiments are carried out to evaluatethe relevancy of the criteria. Nevertheless, machining doesnot comply with high-speed machining cutting conditionsand trajectories used for industrial manufacturing. Theyconsist of a simple straight path, conducted under an over-limited range of cutting conditions. The latter are generallylow, e.g. cutting speeds of less than 40m/min whilemachining carbon steel or aluminium alloy[8,15,17–19,22,23]. In this way, significant and suddenchanges are encountered neither in cutting conditions nor ARTICLE IN PRESS MTM : 1857 13579111315171921232527293133353739414345474951535557596163656769717375777981838587899193959799101103105107109111113 Industrial & academic approachesto TCM in millingPart specific approachesGeneric approachesTeach-inmethodPredictivemethodSimple signalprocessingProcess-basedsignal processingArtificialIntelligence Fig. 1. TCM classification.Table 1Summary of properties of process-based TCM criteriaProcess-basedTCM criteriaCutter breakageor chippingdetectionUnaffected bycuttereccentricityUnaffected bychanges incuttingconditionsAltintasYellowley [6] |    Lee et al. [19]  | |   Kim Chu [20]  | |  ??Deyuan et al.[21] | |  ??   ϕ s N   a   e f  z F t tooth  j   ε  j -1 tooth  j  -1h c ( ϕ )F x F r F y ϕ   m  e  a  n   f  o  r  c  e  p  e  a   k  -   t  o  -  v  a   l   l  e  y  p  e  a   k 1 spindlerevolution   r  e  s  u   l   t  a  n   t   f  o  r  c  e time Fig. 2. Chip thickness and forces in milling. M. Ritou et al. / International Journal of Machine Tools & Manufacture  ]  ( ]]]] )  ]]]  –  ]]] 2     U   N  C  O   R   R   E  C   T   E   D    P   R  O  O   F in cutting forces, respectively. Thus, there is a lower riskthat cutting forces transients would be misinterpreted andgenerate false alarms. So, there is generally a lack of experiments under industrial cutting conditions andtrajectories. This is also the case with the retained criteria.In the first section of this paper, we will present a studyof three process-based TCM indicators, extracted fromliterature. It is verified whether they are unaffected bychanges in cutting conditions, so as to evaluate theirrelevancy for the monitoring of small-batch manufacturingof industrial parts. Experiments were carried out undervarious real industrial machining settings. The criteria werefound to be unreliable, due to misinterpretation of suddenchanges in cutting conditions. Therefore, in the secondsection, a versatile in-process monitoring system issuggested, to tackle the problem of reliability. A newapproach is proposed based on our experiments of millingforces under a range of cutting conditions. Relative radialeccentricity is estimated and this characterizes instant toolstate. Unlike the criteria extracted from literature, this newapproach was successfully implemented using the sameexperiments. 2. Criteria extracted from literature  2.1. Definition  ½ 20 ; 21  Peak  F   j   and peak-to-valley PV  j   values are extracted frommilling resultant forces, for each tooth  j   and for eachspindle revolution (Fig. 2). PV  j   represents the mean of PV  j  over the last 10 spindle revolutions, whereas  F   is the meanof the peak forces during the current spindle revolution.  Z  is the tooth number. Then, the criteria for each tooth andeach spindle revolution are calculated as follows:TFI  j   ¼ PV  j  PV  j   1 PV  j  PV  j   1 ;  Km  j   ¼  F   j     F   j   1 F   j   þ  F   j   1 ,Bm  j   ¼ X Z k  ¼ 2 F  k  F    1   .  ð 1 Þ The TFI takes into account the last 10 spindle turns toprevent it from tool runout and changes in cuttingconditions. So, whether the tool is new or worn, TFI  ¼  1during steady cuts. It focuses on sudden force transients todetect cutter breakage. After an event, it returns to 1. Thus,it is important to distinguish transient cut and problemcorrectly when forces vary. Bm and Km characterize theprocess at any given spindle revolution and their values arecompared to fixed thresholds [21,24].  2.2. Analytic study In order to determine what the criteria depend on,Sabberwal [25] force models are used, where  k  t  and  k  r  areconstants,  h c ð j Þ ¼  f  z  sin ð j Þ  the instant chip thickness [26], a p  the depth of cut,  f  z  the feed per tooth. An overview andparticulars of force models were published [27,28]. How- ever, for an early study, Sabberwal force models were used. F  t  ¼  k  t h c ð j Þ a p ; F  r  ¼  k  r F  t : (  (2)The term  e  j   defines tooth radial eccentricity, the influence of the cutter shape, tool and spindle runout and the amountof cutter chipping. Relative radial eccentricity  D e  j   isintroduced in the expression of instant chip thicknessremoved by tooth #  j   [21,29]: h c  j  ð j Þ ¼  f  z  sin j  þ  e  j     e  j   1  ¼  h c ð j Þ þ  D e  j  . (3)Using the hypothesis that only one tooth participates in thecut at the same time ( h c  is the maximum chip thickness and K   a specific cutting coefficient),PV  j   ¼  F   j   ¼  Ka p ð h c  þ D e  j  Þ . (4)An equivalent formulation of criteria is obtained, depend-ing on cutting conditions:TFI  j   ¼  h c  þ  D e  j  h c  þ D e  j   1 ,  h c  þ  D e  j  h c  þ  D e  j   1 ,Km  j   ¼  D e  j    D e  j   1 2 h c  þ D e  j   þ D e  j   1 ;  Bm  j   ¼  e  j  h c .  ð 5 Þ In the case of breakage or chipping of teeth #  j  ,  e  j   decreases.So, criteria formulation allows them to be detected. Theinfluence of   h c  can be seen for each indicator. Therefore, if  ARTICLE IN PRESS MTM : 1857 13579111315171921232527293133353739414345474951535557596163656769717375777981838587899193959799101103105107109111113 Fig. 3. Test part and toolpath (on the left). Experimental set-up (on the right). M. Ritou et al. / International Journal of Machine Tools & Manufacture  ]  ( ]]]] )  ]]]  –  ]]]  3     U   N  C  O   R   R   E  C   T   E   D    P   R  O  O   F the feedrate or the width of cut varies during the machiningof a part, the criteria should vary and this could affect theirreliability.This explains why such experiments have been carriedout, where feedrate and width of cut vary duringmachining. A corresponding test part was designed. Apocketing operation was chosen with a zigzag strategy,allowing up milling and down milling. This comprises bothsimple and sharp turns (Fig. 3), where the feedrate shoulddrop, due to the limited acceleration available on themachine axes [30]. A contouring path finishes the pocket.  2.3. Experiments Cutting force signals were measured using a 9257AKistler quartz three-component dynamometer, sampling at64kHz. The dynamometer was mounted between theworkpiece and the table of a Sabre Cincinnati machiningcentre. The  X  - and  Y  -axis position encoders were measuredusing a sample frequency set at 500Hz [31]. The workpiecewas made of 7075 aluminium alloy. The parameters werethe feed per tooth (0.08, 0.12, 0.16 and 0.2mm/rev/tooth),the width of cut (15%, 40%, 65%, 90% of tool diameter)and the tool (a 32mm diameter with two inserts and a20mm diameter endmill with three flutes). Different sets of these parameters were tested at every level of the work-piece. Depth of cut was 2.5mm. Since cutting speed was650m/min, spindle speeds were 6500 and 10000RPM, andfeedrates ranged from 1 to 6m/min. So, unlike manystudies, the experiments were carried out under realindustrial cutting conditions.The  X   and  Y   force components were low-filtered at twicethe tooth passing frequency before calculating the resultantforce. Then, force minimums and maximums were calcu-lated for each tooth and for each spindle revolution, toevaluate  F   j   and PV  j  . Based on axes encoder measurements,the instant feedrate Vf was calculated as well as the instantwidth of cut  a e . The edges of the workpiece were discretizedevery 0.05mm. Then, for each new tool position, intersec-tions with the swept volume of the tool led to the entry andexit angles of the teeth and then the instant width of cut  a e was obtained. The  Z  -axis of  Fig. 4 represents the instantcutting conditions (Vf and  a e ) during the machining of thepocket. It reveals that sudden changes occur duringturning.  2.4. Implementation of criteria The criteria were applied to the force signals measuredduring the machining of the pocket. The instant feedrateand instant width of cut allowed a better understanding of the behaviour of these criteria (cf. two first graphs in Fig.5). The third graph represents the resultant cutting forces(blue curve). Peak-to-valley values were extracted from thelatter for each spindle revolution and each tooth (red, greenand black curves). Then, the TFI was calculated from thesevalues (fourth graph). In this way, peak values per toothwere extracted (fifth graph) and Deyuan et al.’s criteriawere calculated.During steady cuts, TFI  ¼  1. Bm and Km take a set of values for each tooth. If a problem occurs, force signals aremodified and the indicator values change, allowing incidentdetection [20,21]. Note that Bm always goes beyond the threshold proposed by Deyuan et al. during finish millingbecause the chip thickness is too inferior to the cuttereccentricity [24]. There would be a permanent false alarm,in this situation.During simple turns (e.g. zone 1 Fig. 3 or first turn Fig. 5), the feedrate of the axes of the machine slows downwhilst turning because the acceleration of each axis islimited [30]. Therefore, the feedrate decreases. It was foundthat  a e  also varies. Towards the end of a straight pathpreceding a turn, the width of cut increases due to thematerial left behind during the previous path, as betweenpositions  a  and  b  in Fig. 6. Turning begins at position  b .Since the acceleration available on machine tool axes islimited, feedrate slows down and the controller adds aportion of circle in the corner, to allow turning with a lowerbut acceptable feedrate [30]. Disregarding spindle rotation,the tool turns around an axis located on its left. Its rightside moves into the material, increasing the exit angle  j s and leading to down milling. On the left-hand side,material has already been partially removed, so the entryangle  j e  decreases a little and  a e  reaches a peak at position c . During the second half of the turn,  a e  decreases because j e  decreases further and  j s  has reached its maximum. After d  , the link path begins, the tool penetrates the material andthe width of cut can reach a full diameter immersion.Under these moderate changes in cutting conditions,criteria variations are negligible (Fig. 5) and tool breakagecan be detected regardless [24]. ARTICLE IN PRESS MTM : 1857 13579111315171921232527293133353739414345474951535557596163656769717375777981838587899193959799101103105107109111113 Fig. 4. Instant feedrate and width of cut during pocket machining (CAM settings: Vf   ¼  3 : 6m = min,  a e  ¼  65%, tool + 20mm). M. Ritou et al. / International Journal of Machine Tools & Manufacture  ]  ( ]]]] )  ]]]  –  ]]] 4     U   N  C  O   R   R   E  C   T   E   D    P   R  O  O   F On the contrary, during sharp turns (second turn in Fig.5), the changes in cutting conditions are more significant.The drop of   a e  is substantial because, in the second half of the turn, most of the material on the left of the tool hasalready been removed. However, the influence of   a e  isnegligible: beyond  a e  ¼  50% of tool diameter,max f sin j g ¼  1. So, maximum chip thickness  ð h c  ¼  f  z   max f sin j gÞ  and peak forces are theoretically unaffected. Inthe  a e  graph in Fig. 5, a minimum of 35% is reached. Thiscorresponds to  j  ¼  75 1  and max f sin j g ¼  0 : 95, i.e. 95% of its previous value. As the  a e  minimum is reached when thefeedrate returns to a medium value, the influence of   a e  isnegligible in this case, unlike during entry and exittransients and finish paths.During sharp turns, the fall in feedrate is significant and,due to the relative cutter eccentricity  D e  j  , some of the teethremove hardly any material. That only a few teeth of thetool participate in the cut is quite usual: this happensduring entry and exit transients, sharp turns and finishpaths (according to cutting conditions). Fig. 7 reveals that,in these cases, false alarms would have been sounded. Thiscan be explained by the equivalent formulation of thecriteria (Eq. (5)) [24]. Consequently, the criteria areunreliable during significant changes in cutting conditions.This contradicts what the developers of the criteria [20,21]suggested.  2.5. Conclusions of the studied criteria It was shown that the three process-based TCM criteriaretrieved from literature are unreliable. Although, duringsteady cut or moderate changes in cutting conditions, ARTICLE IN PRESS MTM : 1857 13579111315171921232527293133353739414345474951535557596163656769717375777981838587899193959799101103105107109111113 Fig. 5. Behaviour of TFI, Km and Bm criteria during rough milling of pockets, calculated from peak-to-valley PV  j   and peak  F   j   values of resultant cuttingforce (CAM settings: Vf   ¼  3 : 6m = min,  a e  ¼  65%, tool + 20mm). M. Ritou et al. / International Journal of Machine Tools & Manufacture  ]  ( ]]]] )  ]]]  –  ]]]  5
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks