Self Improvement

Tool wear monitoring based on a non-monotonic signal feature

Tool wear monitoring based on a non-monotonic signal feature
of 8
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
  Tool wear monitoring based ona non-monotonic signal feature K Jemielniak  Faculty of Production Engineering, Warsaw University of Technology, Institute of Manufacturing Technology,Narbutta 86, Warsaw 02-524, Poland. email: The manuscript was received on 1 February 2005 and was accepted after revision for publication on 3 October 2005. DOI: 10.1243/095440506X77625  Abstract:  Most tool-condition-monitoring (TCM) strategies are founded on the assumptionthat the monitored diagnostic signal feature is increasing, the monotonic function of the tool wear, but this is not always the case. The paper presents the TCM strategy based on non-monotonic signal features. The used-up portion of the tool life ( D T  ¼ t  / T  ) was used as the toolcondition indicator. It is much more informative than the direct value of a sensor signal orany signal feature, and more practical than tool wear measures (VB and KT) which are notusually measured in factory floor conditions. Keywords:  cutting, tool wear monitoring  1 INTRODUCTION The application of tool-condition-monitoring (TCM)systems is an important factor in the improvementin product quality and reduction in productioncosts. Although a number of such systems is avail-able on the market and numerous are successfully applied in factory floor conditions, they are still notused very often. This is because they are still not reli-able enough and not sufficiently user friendly [ 1 ].The existing TCM systems, both laboratory andcommercially available, are based on measurementsof physical phenomena which are correlated withthe tool wear and thus can be exploited as the tool wear symptoms. The quantities most often usedin commercial TCM systems are cutting forcecomponents and force-related quantities such aspower, acoustic emission, and vibration [ 2 – 4 ]. A review of earlier developments can be found in refer-ence [ 5 ]. The monitoring strategies used in these sys-tems make use of a monotonic increment in somesignal feature accompanying the tool wear progress.The signal feature (SF) most often used is averagevalue of the signal, but also the maximum signalvalue and others also can be used.Figure 1 presents a typical strategy for tool wearmonitoring. The learning process of the system con-sist in machining the first workpiece with a new sharp tool. The obtained value SF 0  of the signal fea-ture (here it is the area under the signal versus timecurve) is automatically normalized, i.e. regarded as100 per cent. Then the threshold level SF T  of the Fig. 1  Tool-wear-monitoring strategy used in most TCM systems [ 6 ]: 1, area signal learned  ¼  100 per cent; 2, larger area signal (e.g. through worn tools); 3, learnt area (bar diagram); 4, pre-alarm limit (e.g. 130 per cent), wear limit (e.g.150 per cent); 6, pre-alarm (area exceeds wear limit); 7, wear alarm (area exceeds wear limit) JEM289    IMechE 2006 Proc. IMechE Vol. 220 Part B: J. Engineering Manufacture163  signal feature, or the so-called wear limit, is calcu-lated as the admissible increase in the signal featurevalue in percentage terms according to SF T  ¼  SF 0  1  þ  d  SF T 100 %    ð 1 Þ  where dSF T -limit factor, i.e. the admissible relativeincrease in the signal feature value given by  d  SF T  ¼  SF T    SF 0 SF 0 ·  100 %  ð 2 Þ  While monitoring, after each cycle, the systemdisplays the value of the selected signal feature in a digital or graphic form (Fig. 2). When the featurereaches the threshold level, the tool life is assumedto have come to its end (tool failure). Sometimestwo limits can be set: a warning limit (e.g. 130 percent) and the actual tool failure threshold (e.g.150 per cent). The admissible increase in the dSF T measure in percentage terms can be preset by themanufacturer, but its final and conclusive valuemust be determined by the operator. System tuning consists in correcting the dSF T  value. The operatorhas to make some additional computationsaccording to equations (1) and (2), which is a ratherinconvenient and unclear procedure that requiressome expertise. There are no reasons why theoperator should not be relieved of dealing withsignal values by making the system tuning easier by means of simplifying the communication betweenthe system and the operator.The necessary condition for signal feature applic-ability for TCM is a correlation of the feature withthe tool wear. Commercial systems make use of only those measures that are positive, monotonic,and increasing. This eliminates many signal featuresthat do not fulfil these conditions, although they arequite well correlated with the tool wear.The project undertaken at the Warsaw University of Technology aims at changing this. The systemshould require the user to input only simple andessential information. Instead of quoting signalvalues, the system should be able to indicate tool wear in percentage terms. The system’s learning  Fig. 2  Examples of information on signal feature values presented to the user [ 7 ] Fig. 3  Transformation of the matrix   SF Op  into the matrix   SF D T  164 K Jemielniak Proc. IMechE Vol. 220 Part B: J. Engineering Manufacture JEM289    IMechE 2006  process should be easy, requiring no direct defini-tion of threshold or factor values. The tuning of thesevalues should be equally straightforward. Moreover,the system strategy should make it possible to usethe signal features, regardless of their sign, directionof changes, and (if possible) also their monotonicity.This paper presents the basic principles of such a strategy. 2 USED-UP PORTION OF TOOL LIFE AS THEINDEX OF TOOL WEAR  Although every machine tool operator knows whatthe cutting force or the motor power is, in practicehe/she does not use these quantities and has nointuitive sense of their values and changeability. Asfar as the tool condition is concerned, the naturalcategories would rather include ‘sharp’, ‘partially  worn’, or ‘worn out’ (failed). Such tool wear mea-sures as VB or KT are seldom used in factory floorconditions. Therefore the concept of the  used-upportion of tool life  ,  D T   is introduced and defined asthe ratio of the cutting time as performed so far tothe overall tool life span:  D T  ¼ t  / T  . As already stated, the user is informed about theinitial value SF 0 , the current value SF, and thethreshold value SF T  of the signal feature (see Fig. 2, where these values are designated as A,  c &&  and Lrespectively). These values are not interesting forthe operator from the viewpoint of their inherentvalues. In comparing them, he/she tries to evaluatethe quantity which is interesting, namely the used-up portion of the tool life, which given by  D T   ¼  t T   ¼  SF    SF 0 SF T    SF 0 ð 3 Þ Thus for the data in Fig. 2, for example D T   ¼  300    250400    250  ¼  0 : 33 Of course equation (3) is in terms of an SF valuethat is linear, which is not always the case.Therefore, a special procedure must be developedto make use of non-linear and non-monotonicsignal features. 3 STRATEGYOFTHETOOL-WEAR-MONITORINGSYSTEM3.1 Learning process of the system The learning process of the TCM system begins withthe first operation (workpiece) machined with a new sharp tool. During this operation, the systemmeasures the available signals, their maximum andminimum values, the number of cuts, and theirrespective tools. In this and in the following opera-tions, the system calculates selected signal featureand stores it in a matrix   SF Op , which is the valuesof SF assigned to the operation number. Machining is carried out until, after performing a subsequentoperation, the operator recognizes the end of thetool life, by means of criteria independent of themonitoring system, used before its installation,such as the deterioration of surface quality ormachining accuracy. It should be pointed out thatthere is no need for the operator to enter any numer-ical values. It should also the noted that it is notnecessary to determine the tool wear, i.e. to measurethe direct tool wear indices such as VB and KT. Fig. 4  Evaluation of the used-up portion of the tool lifeusing the measured value of the signal feature SFand the matrix   SF D T  Tool wear monitoring based on a non-monotonic signal feature 165JEM289    IMechE 2006 Proc. IMechE Vol. 220 Part B: J. Engineering Manufacture   After completing the first tool life, the systemtransforms the matrix   SF Op  into the matrix   SF D T  (Figure 3). First the matrix   SF Op  is smoothed (low-pass filtered) to eliminate random changes in thesignal feature. The resultant matrix is storedas  SF OpFlt . Then, to each operation number, theused-up portion of the tool life is assigned according to the formula  D T   ¼  n N   ð 4 Þ  where n  ¼  number of operations  N   ¼  total number of operations to tool failureFinally, the filtered matrix  SF OpFlt  having a numberof elements equal to the number of operations istransformed into a 20-element matrix consisting of SF values corresponding to the used-up portionsof the tool life, every 5 per cent of   D T  , according tothe formula  SF D T  ½ i   ¼ ð SF Op ½ ceil ð  x  Þ SF Op ½ floor ð  x  ÞÞ½  x     floor ð  x  Þ þ SF Op ½ floor ð  x  Þð 5 Þ  where  x   ¼  iN  /20 i   ¼  index of the  SF D T   matrix corresponding to theused-up portion of the tool life  D T   ¼  5 i  ,  i   ¼  0–20floor(  x  )  ¼  largest integer 6  x  ceil(  x  )  ¼  smallest integer >  x  Then the matrix is extrapolated to D T   ¼  150 per cent,meaning that ten elements are added according tothe formula  ST D T  ½ i   ¼ ð SF D T  ½ 20   SF D T  ½ 19 Þð i     20 Þþ SF D T  ½ 20  ;  i   ¼  21  –  30  ð 6 Þ 3.2 Tool wear monitoring  During machining with the subsequent tools, theTCM system measures the signals and, after eachoperation, it calculates the signal feature value, justlike during the learning process described above.This time, however, the system evaluates the used-up portion of the tool life by searching for the valueof SF, which is closest to that obtained in the lastoperation, in the  SF D T   matrix (Fig 4(a)).It may happen that the SF value corresponds to a value of the used-up portion of the tool life lowerthan that reached in the previous operation. Such a system indication would be disorienting for theoperator. Therefore, it was assumed that the searchstarts from the  D T   value obtained last time, whichmeans that the used-up portion of the tool life pre-sented to the operator cannot decrease (Fig. 4(b)).Sometimes it happens that the SF value affectedby some disturbances corresponds to a very largeincrease in the tool wear. To remedy such a mistake,it was assumed that the TCM system is applicable if during the tool life at least three operations can beperformed. Therefore the search for the SF value islimited to six elements of the  SF D T   matrix, i.e. to30 per cent of the tool life (see Fig. 4(c)). This meansthat, in the case of accelerated tool wear, the systemallows three operations to be performed before itsignals tool failure. This procedure also has anotherpurpose, namely it enables signal features whichare not monotonic with respect to the used-up por-tion of the tool life to be utilized, at least to someextent, as presented in Fig. 4(d). In the exampleshown here, the signal feature values correspond to D T   ¼  70 per cent and  D T   ¼  90 per cent. Restrictionof the matrix search to 30 per cent of  D T   results indi-cates that  D T   ¼  70 per cent.It may also happen that, because of some distrac-tion, the SF value ceases its typical changes andremains in the value range denoting absence of tool wear. On the other hand, in properly selected cutting conditions, a standard deviation of the tool lifeusually does not exceed 15 per cent of the averagevalue of the tool life. Thus, to force SF to rise in any  Fig. 5  The signal feature (average value of the feed force) versus number of operations in three selected tool lives 166 K Jemielniak Proc. IMechE Vol. 220 Part B: J. Engineering Manufacture JEM289    IMechE 2006  case (the tool cannot stop undergoing wear), it wasassumed that the system indication cannot be lowerthan more than approximately two standard devia-tion of the tool life (30 per cent), meaning 70 percent of that resulting from previous experience,calculated accordingly to the formula  D T   >  70  n N B ð 7 Þ  where n  ¼  number of just finished operations  N  B  ¼  number of operations in the previous tool lifeThe value of the used-up portion of tool life calcu-lated in this way is displayed by the system after eachoperation. When  D T   exceeds 100 per cent, the sys-tem generates an alarm signal which can be usedby the machine tool control system to replace thefailed tool automatically or simply to inform theoperator about the tool failure. The operator can judge that the tool failed earlier or later thanindicated by the system by pressing a key. Whenhe/she regards the tool as being dull, making use of the system indication or not, the matrix   SF Op  istransformed into  SF D T  , like during the learning pro-cess. However, during monitoring, the new contentsof the matrix   SF D T   is calculated using just theobtained results  SF D T  current  and the previous con-tents  SF D T  prev   of the matrix according to SF D T   ¼  0 : 25 SF D T  current  þ  0 : 75 SF D T  prev   ð 8 Þ Thus the matrix   SF D T   is based on several previoustool lives (i.e. the system gathers experience), and a singular untypical signal feature course has lessinfluence on the system performance. 3 RESULTS OF EXPERIMENTS The experiments have been performed on a Venus450 turning centre equipped with an industrial cut-ting force sensor (Kistler 9601A31) installed underthe turret. The workpieces were steel C 45 bars, of 160 mm diameter, machined in subsequent cuts with depths of cut,  a p  ¼  1.5 (13 cuts) and  a p  ¼ 2 mm (nine cuts), feed  f   ¼  0.1 mm/rev, and cutting speed  v  c  ¼  150 m/min, down to 85 mm diameter.Sintered carbide tools CNMG 10408 BP30A wereused. More details of the experimental procedurehave been given in reference [ 8 ].Figure 5 presents the signal feature (the averagevalue of the feed force signal in this case) obtainedin three selected tool lives. It can be seen that thissignal feature (and many others, not presentedhere) is a non-monotonic function of the used-upportion of the tool life (tool wear). Here, this iscaused rather by the sensor installation than by thereal character of the feed force but, as may happenin factory floor conditions, it should be taken intoaccount. Tool wear estimation is based on thereverse function, where the value of the signal fea-ture determines the tool wear estimation. As thenon-monotonic functions are not reversible, allTCM systems using such strategies must fail if thesignal feature appears to be non-monotonic. Fig. 6  Transformation of the  SF Op  matrix into the  SF D T  matrix after (a) the first tool life, (b) the secondtool life, and (c) the third tool life, together with(d) the resultant  SF D T   matrix contents after thesetool lives Tool wear monitoring based on a non-monotonic signal feature 167JEM289    IMechE 2006 Proc. IMechE Vol. 220 Part B: J. Engineering Manufacture
Similar documents
View more...
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

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!