Monitoring online cutting tool wear using low-cost technique and user-friendly GUI

Wear 271 (2011) 2619–2624 Contents lists available at ScienceDirect Wear journal homepage: www.elsevier.com/locate/wear Short communication Monitoring online cutting tool wear using low-cost technique and user-friendly GUI J.A. Ghani ∗ , M. Rizal, M.Z. Nuawi, M.J. Ghazali, C.H.C. Haron Department of Mechanical and Materials Engineering, Faculty of Engineering and Built Environment, Universiti Kebangsaan Malaysia, 43600 Bangi, Malaysia a r t i c l e i n f o Article history: Received 20 August 2010 Received in revised form 18 January 2011 Accepted 18 January 2011 Keywords: Flank wear Low-cost monitoring Strain gauge I-KazTM method a b s t r a c t Cutting tool wear is known to affect tool life, surface quality and production time. Because of this an online tool wear measurement and monitoring system has been developed, using a low-cost sensor. The system is able to detect and analyze signals relating to the deflection of the tool holder from the cutting force, and the corresponding estimation of wear is displayed on the computer screen. New statistical analysis is used to identify and characterise the changes in signals from the sensors. A two-channel strain gauge is mounted at the tool holder to measure the deflection in both tangential direction and feed direction. The signal is transmitted to the signal conditioning device, then to data acquisition, and finally to the computer system. MATLAB software is used as the platform software to develop a user-friendly graphical user interface (GUI) for online monitoring purposes. Results show that this developed online monitoring system, using the strain gauge signal, is an effective method of detecting the progression of flank wear width during machining. This is an efficient and low-cost method which can be used in the real machining industry to predict the level of wear in the cutting tool. © 2011 Elsevier B.V. All rights reserved. 1. Introduction In recent years the importance of tool wear monitoring has been recognized in manufacturing industries. In practice, approximately 20% of the downtime of machine tools is reported to be due to tool failure, and the cost of cutting tools and their replacement accounts for between 3 and 12% of total production costs [1,2]. In order to improve the quality of machined parts, reduce production time and save costs, tool condition monitoring systems have been extensively studied by researchers since the late 1980s [3]. Several methods have been proposed to monitor tool wear. There are two main categories: direct methods and indirect methods. Direct methods, such as machine vision systems, use a charged-couple-device (CCD) camera or optical microscope [1–4]. Direct methods have the advantage of capturing actual geometric changes arising from the wear of the tool. However, direct measurements are very difficult to obtain due to the continuous contact between the tool and the workpiece, and are made almost impossible by the presence of coolant fluids. These difficulties severely limit the application of a direct approach [5,14]. Indirect methods correlate or match appropriate sensor signals to tool wear states. The advantages are a less complicated setup and greater suitability to practical application. In indirect methods, tool condition is not ∗ Corresponding author. Tel.: +60 3 8921 6505; fax: +60 3 8925 9659. E-mail address: jaharah@eng.ukm.my (J.A. Ghani). 0043-1648/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.wear.2011.01.038 captured directly, but estimated from the measurable signal feature. This signal feature is extracted through signal processing steps to obtain a sensitive and accurate representation of its state. Indirect methods include those based on sensing of the cutting forces [6,7], vibrations [8,9], acoustic emission [10,11], motor current [12] and other elements [13]. Cutting force is generally considered one of the most significant variables in the turning process [14]. It has been widely recognised that variation in the cutting force can be correlated to tool wear, as a result of the variations produced by friction between cutting tool flank and workpiece [15,16]. Tool dynamometers are commonly used to record these cutting forces [13]. However, dynamometers are not suitable instruments for shop floor use due to their high cost, negative impact on machining system rigidity, geometric limitations and lack of overload protection [17]. Therefore, a low-cost system for measuring cutting forces is necessary. In the past, Scheffer and Heyns [18] used a simple sensor-integrated tool holder using strain gauges, and Audy [19] also presents an overview of techniques and equipment used for measuring cutting forces using a strain gauge-based system. Many approaches to online monitoring have been explored. Zhang and Chen [20] developed software using VBA and SoftWIRE to monitor tool condition in a CNC end-milling machining process, based on the vibration signal collected through a microcontrollerbased data acquisition system. They used the Fast Fourier Transform (FFT) function, and its graphic display was integrated into the software program developed by SoftWIRE. Choudhury and 2620 J.A. Ghani et al. / Wear 271 (2011) 2619–2624 Nomenclature ch Dc Fz Fx f FCD K Mr N s VB Vc Z2∞  channel depth of cut, mm tangential force, N feed force, N feed rate, mm/rev Ferum Casting Ductile kurtosis the rth order of moment number of data points standard deviation flank wear width, mm cutting speed, m/min I-KazTM 2D coefficient variance of data Srivinas [21] developed a mathematical model (regression model) to predict the flank wear on a cutting tool. The wear was predicted taking into account the cutting conditions. Parameters for the model were estimated from experimental data. They concluded that the predicted results correlated well with measured values of flank wear. Chen and Li [7] developed a tool wear observer model to monitor flank wear when machining nickel-based alloys. The model can be implemented in an online tool wear monitoring system which predicts the actual state of tool wear in real time by measuring the cutting force variations. The correlation between cutting force components and flank wear width has been established through experimental studies. This paper will describe the application of a low-cost sensor for monitoring online cutting tool wear with a Graphical User Interface (GUI), by measuring cutting tool deflection and analysing its signal using a new statistical-based method called Integrated Kurtosisbased Algorithm for Z-filter Technique (I-KazTM ), pioneered by Nuawi [22]. Machining experiments were carried out, using CNC machines, to collect data which related the signal from cutting tool deflection to gradual flank wear. Several equations of flank wear and signal characteristic correlation were then installed on a Matlab GUI program, in order to monitor and predict online flank wear progression in the turning process. 2. Methodology There are two stages involved in this study. Stage 1: machining experiments for data collection, and database development relating to flank wear and its signal characteristic. Stage 2: developing a user-friendly GUI using Matlab GUI programming for online tool wear monitoring and prediction of flank wear in the turning process. 2.1. Design and procedure of the experiment In this study, FCD700 (JIS) grade ductile cast iron with spherical graphite and ferrite was chosen as the workpiece material, with a composition by wt% of: 3.32% C, 2.68% Si, 0.46% Mn, 0.028% P, 0.018% S, 0.85% Cu and 0.09% Mg. The Brinell hardness and tensile strength were in the range of 241 HB and 845 MPa respectively with elongation of 6%. The machining tests were carried out on a CNC Colchester Master Tornado T4 lathe machine, under dry cutting conditions. The tool used was a coated carbide insert (Toolmex CNMG120404-MB) with NC30P grade 5. The tool had a rhombic 80◦ shape, 4.76 mm thickness and 0.4 mm nose radius. This grade of carbide tool is suitable for heavy duty cutting of cast iron and steel. Two strain gauges were mounted on a simple tool holder and used Table 1 Experimental design. Experiment no. Cutting speed, Vc (m/min) Feed rate, f (mm/rev) Depth of, Dc , cut (mm) 1 2 3 4 5 6 7 8 9 180 180 180 230 230 230 270 270 270 0.15 0.20 0.25 0.15 0.20 0.25 0.15 0.20 0.25 0.5 1 1.5 1 1.5 0.5 1.5 0.5 1 to detect the dynamic changes in both Fz (tangential direction) and Fx (feed direction) in cutting tool deflection. The direction of radial force, Fy , was approximately zero and was ignored [19]. Different combinations of machining parameters were selected using the Taguchi method. The Taguchi method is a cost-effective and timesaving method of exploring inter-relationships between various parameters [23]. It has been extensively used to determine optimum parameters. An L9 (33̂) orthogonal array was chosen because of its minimum number of experimental trials. The three main machining parameters, namely cutting speed, feed rate and depth of cut, were set prior to the machining run. Three levels of cutting speed (Vc = 180, 230 and 270 m/min), three levels of feed rate (f = 0.15, 0.20 and 0.25 mm/rev) and three levels of depth of cut (Dc = 0.5, 1 and 1.5 mm) were used. Therefore nine combinations of those cutting parameters were taken into consideration as shown in Table 1. The ranges for the cutting parameters were selected based on the real turning operation of FCD 700 carried out in one of the local automotive industries for producing camshafts and crankshafts. During the turning operation, the insert was periodically removed from the tool holder, and the flank wear on the flank face was measured using a Mitutoyo toolmaker’s microscope equipped with a graduated scale in mm. The measured parameter to represent the progress of wear was flank wear width, VB. The turning operation was stopped and the insert was discarded when VB reached 0.3 mm. This is a standard recommended value in defining a tool life end-point criterion based on ISO 3685:1993 [24]. In order to detect the deflection effect, the distance between the strain gauge and the tip of the tool insert was 45 mm. The sensors were connected to a signal conditioning device using two wires sheathed in cable to avoid noise. The transfer cable was then covered with aluminum foil to protect it from damage caused by chips. A beam shaped tool holder with sectional size 20 × 20 mm was used in the study. The sensor configuration used a half bridge strain gauge set in the signal conditioning device. The sampling rate was set at 1 kHz and zero calibration was carried out before the experimental machining. The signals were captured in two channels in the time domain. Every step in the machining process provided one set of cutting tool deflection signals. These signals were analyzed using I-KazTM to estimate the characteristic due to flank wear. The correlations between signals characteristic and flank wear data were compiled and installed in the software program to provide a database for prediction purposes. The flow chart of the experimental process is shown in Fig. 1. 2.2. Signal analysis method The signal analysis method used was a statistical signal processing based on Kurtosis I-KazTM method. The I-KazTM method was pioneered by Nuawi [22], who studied random or nondeterministic signal characteristics. In order to classify random J.A. Ghani et al. / Wear 271 (2011) 2619–2624 2621 signals, the rth order of moment (Mr ) is frequently used. The Mr for the discrete signal in the frequency band can be written as: n Mr = 1 (xi − x̄)r N (1) i=1 where N is the number of data points, xi is the data value at the instantaneous point and x̄ is the mean. Eq. (1) provides the basis for calculating the kurtosis value. Kurtosis, which is the signal’s 4th statistical moment, is a global signal statistic which is highly sensitive to the spikiness of the data. For discrete data sets the kurtosis, K, is defined as: n K= 1  (xi − x̄)4 N 4 (2) i=1 Fig. 1. Procedure of experimental process. where N is the number of data points,  is the variance, xi is the data value at the instantaneous point and x̄ is the mean of the data. The kurtosis value is approximately 3.0 for a Gaussian distribution. Higher kurtosis values indicate the presence of more extreme values than should be found in a Gaussian distribution. Kurtosis is Fig. 2. Variation of cutting tool deflection signal and actual tool wear measurement: (a) when flank wear land VB = 0.017 mm, (b) when flank wear land VB = 0.188 mm and (c) when flank wear land VB = 0.319 mm. 2622 J.A. Ghani et al. / Wear 271 (2011) 2619–2624 used in engineering for detection of fault symptoms because of its sensitivity to high amplitude events. Based on kurtosis, the method provides a graphical representation of the measured signal distribution. Specifically, the time domain signal was decomposed into two signals channels. In order to measure the degree of scattering in the data distribution, the IKazTM coefficient calculates the distance of each data point from the signal’s centroid. I-KazTM coefficient was defined as: I-kaz 2D coef =  1  I  1  II  M4 + M4 N N (3) where N is the number of data points and M4I , M4II are the 4th order of moment in channel I and channel II respectively. The I-KazTM coefficient was substituted in Eq. (4) and the symbol of Z2∞ was used to represent the I-KazTM coefficient. Z2∞ 1 = N  KI sI4 + KII sII4 (4) where N is the number of data points, KI and KII are the kurtosis of signal in ch-I and ch-II, and sI and sII are the standard deviation of signal in ch-I and ch-II respectively. 3. Results and Discussion Nine experimental machining were performed to collect cutting tool deflection signals. The signals from two strain gauge sensors were used to measure deflection on the tool holder relating to tangential direction (Fz ) and feed direction (Fx ) due to cutting force action. For every set of experiments, two sets of data were obtained: cutting tool deflection signals and tool wear value from actual flank wear width, measured with a microscope. The signals and tool wear data were recorded from the first cutting until the cutting tool wear reached 0.3 mm. As an example, the raw signals and actual tool wear of the tool insert from the end of the first cutting, mid cutting and at the end of the last cutting are shown in Fig. 2(a), (b) and (c) respectively. These figures show the signal obtained when cutting FCD700 FCD ductile cast iron at a cutting speed of 180 m/min, feed rate of 0.25 mm/rev and depth of cut of 1.5 mm. During turning of the FCD700 ductile cast iron under dry conditions, the amplitude of cutting signals increased with the width of flank wear width. The amplitude of cutting signals increased by 40% from the beginning of the cut (sharp tool) until the tool was worn out. Therefore, the influence of flank wear in machining force was very significant. Fig. 2(a) shows the third of fifteen steps during the cutting operation, and wear stage is still in its initial ‘break-in’ period, with a flank wear width of VB = 0.017 mm. This period lasts only a few minutes, after which wear occurs at fairly uniform rate as shown in Fig. 2(b). At a flank wear width of VB = 0.188 mm, an increase in the cutting tool deflection signals was observed and recorded. Fig. 2(c) shows the cutting tool deflection signals at a flank wear width of VB = 0.319 mm. The machining test aims to collect the cutting signals data in order to build a database that can be used for online monitoring of cutting tool wear. Based on the I-KazTM method, every machining signal has its own characteristic feature due to the effect of increasing flank wear. Eq. (4) is used to calculate the characteristic of raw signals. In this method, components of the machining force, Fz , and Fx are converted into channel I and channel II respectively. Therefore, the calculation of the value of the signal characteristic is called the I-KazTM 2D coefficient, indicated as Z2∞ . For example, Fig. 3 shows the I-KazTM 2D coefficient versus flank wear width at various cutting speeds and feed rates while depth of cut is kept Fig. 3. Correlation between the I-KazTM 2D coefficient and flank wear at various cutting parameters. constant. It shows that trend correlation of the curve indicates a decrease in the coefficient of Z2∞ when flank wear increases. This decline in the coefficient of Z2∞ is in accordance with the trend of power-law curve fit. Fig. 3 shows that experimental data based on cutting tool deflection signals can be used as an input parameter for flank wear prediction. The R-square value of the regression coefficients at Vc = 180 m/min and f = 0.20 mm/rev is 0.991; when Vc = 230 m/min and f = 0.25 mm/rev it is 0.938, and when Vc = 270 m/min and f = 0.15 mm/rev it is 0.941. Comparing these R-square values, the best regression is indicated at Vc = 180 m/min and f = 0.20 mm/rev. Based on the results of curve fitting a new equation is derived, which can be used to detect the flank wear. The equation based on power-law curve fitting is shown as Eq. (5) below. y = ax−n (5) I-KazTM 2D coefficient, a and n are conwhere y is the value of stants which depend on the cutting parameter, and x is the value of flank wear, VB. Therefore Eq. (5) can be written as: Z2∞ = a(VB)−n Z2∞ , (6) In this case, when cutting at Vc = 180 m/min and f = 0.20 mm/rev, the values of a and n were 5 × 10−10 and −0.4692 respectively. When cutting at Vc = 230 m/min and f = 0.25 mm/rev, the value of a and n were 5 × 10−10 and −0.4793 respectively. Then, cutting at Vc = 270 m/min and f = 0.15 mm/rev, the value of a and n were 5 × 10−10 and −0.4738 respectively. Eq. (6) is the main formula for estimating flank wear in the turning process. The graphical user interface displays the flank wear in real time by capturing and analyzing the cutting force signal. The GUI was tested with various cutting parameters to obtain a real time prediction of the flank wear width. The aim of the testing was to compare the performance of flank wear prediction using the developed low-cost monitoring technique, with actual flank wear measurement. In order to compare the performance of the two tool wear prediction processes, some of the cutting parameters were selected to represent the testing, as shown in Fig. 4. The input parameter was the cutting tool deflection signal that was calculated using the mathematical model in Eq. (6). The equation was programmed into the database of the online tool wear monitoring system. Fig. 4 shows the comparison of predicted and actual tool wear magnitude measured with different sets of cutting parameters: Fig. 4(a) when cutting condition Vc = 180 m/min and feed rate = 0.2 mm/rev, (b) Vc = 230 m/min and feed rate = 0.25 mm/rev, (c) Vc = 270 m/min and feed rate = 0.15 mm/rev, and all with a depth of cut of 1 mm. A plot was drawn with the actual measured flank wear values on the x-axis and the predicted values on the y-axis. J.A. Ghani et al. / Wear 271 (2011) 2619–2624 2623 Fig. 4. Comparison of predicted and actual flank wear magnitude measured. It can be observed from Fig. 4 that there is a strong correlation between the predicted and the actual measured values of tool wear, since all the points fall close to a line with a slope equal to 1. The actual correlation coefficient between the experimental and the predicted values in case (a) was found to be around 0.931; in case (b) it was around 0.959 and in case (c) around 0.915. In order to evaluate the error in the predictions, a root-mean-squared (RMS) error between the actual measured magnitude and predicted readings was computed using the equation: %RMSerror =   n  1  VBmeasured − VBpredicted n 1 VBpredicted 2 × 100 (7) The root-mean-square error between the actual measured and the predicted value of the flank wear was found to be between 5.91% and 16.03%. Therefore, the monitoring system is reliable in detecting and estimating the actual flank wear width during the machining process. In addition, this system is able to give an early warning regarding the cutting tool wear condition. This will ensure the part produced in the machining process is of an acceptable quality. The system is being developed at the advanced manufacturing laboratory and is still in the experimental stage. A patent application for the system has been filed and is currently pending. 4. Conclusions This project aimed to apply a new technique to monitor and predict flank wear in CNC turning processes, using a low-cost sensor. A user-friendly GUI was developed, using MATLAB as the software platform, to display the status of the cutting tool condition. A new correlation has been developed between the I-KazTM coefficient of the raw signal and flank wear data. The regression trend of its correlation shows a power-law curve with the R-square values of the regression coefficient between 0.938 and 0.991. This regression was the main formula for flank wear prediction which was installed in the software. The significant features of the user-friendly GUI are its ability to graphically display the results of the cutting force signal, and its statistical characteristic value which is influenced by flank wear. Therefore, this system is able to give an early warning of flank wear in the cutting tool, and assists the smooth operation of the machining process in order to produce a part of acceptable quality. Acknowledgements The authors would like to thank the Government of Malaysia and Universiti Kebangsaan Malaysia for their financial support under a 03-01-02-SF0647 Grant. References [1] S. Kurada, C. Bradley, A review of machine vision sensors for tool condition monitoring, Comput. Ind. 34 (1997) 55–72. [2] M. Castejon, E. Alegre, J. Barreiro, L.K. Hernandez, On-line tool wear monitoring using geometric descriptors from digital images, Int. J. Mach. Tools Manuf. 47 (2007) 1847–1853. [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 the CIRP 44 (1995) 541–567. [4] H.H. Shahabi, M.M. Ratnam, In-cycle monitoring of tool nose wear and surface roughness of turned parts using machine vision, Int. J. Adv. Manuf. Tech. 40 (2009) 1148–1157. [5] J.A. Ghani, M. Rizal, M.Z. Nuawi, C.H.C. Haron, R. Ramli, Statistical analysis for detection of cutting tool wear based on regression model, in: Proceedings of the International Multi-Conference of Engineers and Computer Scientists 2010, vol. III, 2010, pp. 1784–1788. [6] P. Bhattacharyya, D. Sengupta, S. Mukhopadhyay, Cutting force-based real-time estimation of tool wear in face milling using a combination of signal processing techniques, Mech. Syst. Signal Process. 21 (2007) 2665–2683. [7] X.Q. Chen, H.Z. Li, Development of a tool wear observer model for online tool condition monitoring and control in machining nickel-based alloys, Int. J. Adv. Manuf. Tech. 45 (2009) 786–800. [8] G.H. Lim, Tool wear monitoring in machine turning, J. Mater. Process. Technol. 51 (1995) 25–36. [9] C. Scheffer, P.S. Heyns, Wear monitoring in turning operations using vibration and strain measurements, Mech. Syst. Signal Process. 15 (6) (2001) 1185–1202. [10] K. Jemielniak, O. Otman, Catastrophic tool failure detection based on acoustic emission signal analysis, CIRP Ann.—Manuf. Technol. 47 (1) (1998) 31–34. [11] X. Li, A brief review: acoustic emission method for tool wear monitoring during turning, Int. J. Mach. Tools Manuf. 42 (2002) 157–165. [12] X. Li, A. Djordjevich, P.K. Venuvinod, Current-sensor-based feed cutting force intelligent estimation and tool wear condition monitoring, IEEE Trans. Ind. Elect. 47 (3) (2000) 697–702. [13] A.G. Rehorn, J. Jiang, E. Orban, State-of-the-art methods and results in tool condition monitoring: a review, Int. J. Adv. Manuf. Tech. 26 (2005) 693–710. [14] B. Sick, On-line and indirect tool wear monitoring in turning with artificial neural networks: a review of more than a decade of research, Mech. Syst. Signal Process. 16 (4) (2002) 487–546. [15] S.K. Choudhury, K.K. Kishore, Tool wear measurement in turning using force ratio, Int. J. Mach. Tools Manuf. 40 (2000) 899–909. [16] E.D.Snr. Dimla, Sensor signals for tool wear monitoring in metal cutting operations – a review of methods, Int. J. Mach. Tools Manuf. 40 (2000) 1073–1098. [17] G. O’Donnell, P. Young, K. Kelly, G. Byrne, Towards the improvement of tool condition monitoring systems in the manufacturing environment, J. Mater. Process. Technol. 119 (1-3) (2001) 133–139. [18] C. Scheffer, P.S. Heyns, An industrial tool wear monitoring system for interrupted turning, Mech. Syst. Signal Process. 18 (2004) 1219–1242. 2624 J.A. Ghani et al. / Wear 271 (2011) 2619–2624 [19] J. Audy, An appraisal of techniques and equipment for cutting force measurement, J. Zhejiang Univ. Sci. A 7 (11) (2006) 1781–1789. [20] J.C. Zhang, J.C. Chen, Tool condition monitoring in an end-milling operation based on the vibration signal collected through a microcontrollerbased data acquisition system, Int. J. Adv. Manuf. Tech. 39 (2008) 118–128. [21] S.K. Choudhury, P. Srinivas, Tool wear prediction in turning, J. Mater. Process. Technol. 153–154 (2004) 276–280. [22] M.Z. Nuawi, F. Lamin, M.J.M. Nor, N. Jamaluddin, S. Abdullah, C.K.E. Nizwan, Development of integrated kurtosis-based algorithm for z-filter technique, J. Appl. Sci. 8 (8) (2008) 1541–1547. [23] J. Ross, Taguchi Techniques for Quality Engineering, McGraw-Hill, Inc., New York, 1996. [24] ISO (International Organization for Standardization) (Ed.), Tool-life Testing with Single-Point Turning Tools (ISO 3685), 2nd edition, Reference Number ISO 3685: (1993)(E).