Documente Academic
Documente Profesional
Documente Cultură
Summary
Computer-numerical-control (CNC) machines used in automated manufacturing are mechatronic systems. In this chapter, a fault detection method is developed based on a state observer model for a milling machine in a CNC machining center. The CNC machining center is treated as an uncertain linear system. To obtain more information, a robust observer is designed based on the uncertain linear model. Subsequently, this model is used as a state (tool wear) estimator, and fault diagnosis is carried out using twovariable information. The approach can be used for the detection of faults arising from the malfunction of a sensor or an actuator.
25.1 Introduction
Computer-numerical-control (CNC) machines are commonly used in automated factories for producing machined parts. A CNC machine, which consists of mechanical components, actuators, sensors, controllers, and interface hardware and software, is a mechatronic system. In a metal-cutting process using a CNC milling machine, it is possible that a fault occurs during operation even though the process parameters have been set properly. Faults in the cutting tools, which are frequently caused by tool wear, can potentially damage the workpiece. Prevention of faults is therefore important to minimize possible loss in manufacturing. Before a fault occurs, parameters of the cutting process change beyond their normal values. By detecting the unusual change of the parameters, it is possible to anticipate faults and take preventive or corrective action. This possibility of preventing faults has made continuous monitoring of metal-cutting processes and detection of the changes of parameters an important topic in the area of manufacturing automation
25-1
25-2
and mechatronics. Successful monitoring systems can properly maintain the machine tools and delay the occurrence of tool wear. Various methods have been developed to detect the tool wear state. The application of statistical algorithms to associate patterns in measurable signals with wear states is found in Reference [1]. Weck [2] and Byrne et al. [3] use the signal of cutting force to monitor tool wear. By using inexpensive current sensors, several intelligent tool-wear-monitoring systems have been developed [46]. Coker and Shin [7] developed an in-process monitoring and control system for surface roughness during machining via ultrasonic sensing. In Reference [8] an acoustic emission sensor and accelerometers are used to monitor progressive stages of ank wear on carbide tool tips. An alternative approach to tool-wear monitoring is to apply system-theoretic ideas to estimate the wear states during the cutting process. In Reference [9] a linear model is built to detect the tool wear and breakage in the drilling process. In References [10,11] another linearization model is used to design an adaptive observer for online tool-wear estimation in a turning operation. The aim of tool-wear detection is to nd the loss of the original functioning or capability of the tool to detect an abnormal state. The methods of tool-wear diagnosis have focused on the development of signal processing techniques on the measurements such as cutting force, vibration, and spindle motor current. However, a tool signal from a single measurement may make a misjudgment because of the complicated dynamic characteristics of the cutting process and sensor noise. To prevent this, a multisensor approach has been presented in References [12,13]. This requires a higher hardware device supplied for simultaneously treating increased amounts of information. This chapter presents a method of model-based process supervision with fault detection and diagnosis. The model is built based on the data collection from a practical manufacturing plant. Unlike the results of References [911], the method developed in this chapter is focused on a milling machine (see Section 25.2). Note that in general the dynamic model of a milling operation is different from drilling or turning operations, as used in References [911]. The method is also different from what is presented in References [12,13], in that the sufcient observer information (software) is used to make decisions so as to enhance the reliability of tool wear, as opposed to using a multisensor (hardware) technique. The approach presented in this chapter can be summarized in the following steps. First, multiple linear models are identied based on different working conditions, and a dominant model is obtained from the models. The used model is the dominant model plus an uncertainty model with bounded signals. Second, an observer is built based on the identied model. Third, tool-wear signatures are detected by using two signal processing methods: the estimated wear rate based on the observer and the error between the observer and actual cutting force.
(25.1)
25-3
Cutting tool
Ball screw
where F is the cutting force and is the output of the system, f s is the feed rate, and the parameters , n , and K s depend on the depth of cut d, spindle speed v, and feed rate f s . This equation can be rewritten in the statespace form as x = Ax + bu, y = CT x , with x = [F F ]T , u = f s , and 0 A= 2 n 0 1 , b = , C = [1 n Ks (25.2)
0]T
(25.3)
The model is weakly nonlinear and has signicant process parameter variations [14]. Now, consider an uncertain linear model given by x = ( A + A)x + (b + b)u + d, y = CT x , (25.4)
where A and b are the nominal matrices (system matrix and the input gain matrix, respectively) of the system, and 0 A = a1(t ) 0 1 , , b = a2 (t ) b1(t)
(25.5)
2 where a1(t ), a2 (t ), and b1(t ) are the perturbation parameters of n , n , and K s , respectively, and d is the bounded disturbance. It is seen that this model can include more classes than that of Reference [14]. In tool-wear detection, it is well known that the tool life can be divided into three phases characterized by three different wear processes: (1) break-in, (2) normal wear, and (3) abnormal or catastrophic wear. The present objective is to detect the rise in the tool wear and to diagnose the fault types so that a tool replacement decision could be made. Because fault accommodation is not addressed in this chapter, we can make the standard assumption that the control u and the state vector x remain bounded prior to and after the occurrence of a fault.
25-4
Assumption 1 There exist compact sets x R 2 , u R, such that x x and u for all t 0.
(25.6)
where x denotes the estimate of the state x and K = [k1 k2 kn ]T is the observer gain vector. Only the output y is assumed to be measurable. Dene the state and output estimate errors as x = x x and y = y y , respectively. Thus, the error dynamics is given by x = ( A KCT )x + Ax + bu + d, y = C T x. Theorem 1 Consider the nonlinear system described by (25.4) and the observer by (25.6). If Assumption 1 holds, x 0 x. Then, all of the signals are bounded and the state estimate x still remains in the compact ||| x x || BQ ,x x } . In addition, a small error of ||x|| may be achieved by selecting gain K . set x = {x (25.7) (25.8)
= 0 + t,
(25.9)
where w is the tool wear level, w0 is the initial tool wear level, is the wear rate, and t is the cutting time. In the normal phase, the wear rate is constant. However, a sudden rise in the wear rate can be observed in an abnormal phase. Our objective is to monitor the rise in the wear rate to give a warning to the operators so that they can determine whether to replace the tool or to take some other action. It is observed that the tool wear is related to the cutting force [5]. One may represent this by F = F0 + L, (25.10)
25-5
where F0 is the cutting force arising under identical cutting conditions, but with an unworn cutting tool, and L is a parameter dependent on the cutting speed, feed rate, and depth of cut. Substituting (25.9) into (25.10) yields F = F0 + L 0 + L t. The following wear rate is derived in Reference [5] (25.11)
F , L t
(25.12)
where F and t are the differences of F and t, respectively. Unfortunately, the measured cutting force is noisy and this causes difculty in the calculation of (25.12) in reality. On the other hand, from (25.10) it follows that F = L . (25.13)
This implies that F can be used to estimate the wear rate. Although F cannot be computed due to noise, its observer x 2 is available without the need of differentiation. When the observer is designed to satisfy the stability requirement, one can use x 2 in place of F. In order to monitor the tool wear, a time interval is dened as [t 0 ,t f ]. This interval can be computed by t f t0 = lf fr / 60
(25.14)
where l f is the reference distance that is determined by the user, fr is the feed rate (given in unit length per minutes, which is a machining parameter), and 60 is one minute in seconds. The sampling points t t can be calculated by N = f T 0 , where T is the sampling time. Thus, an estimate of x 2 during the interval is given by
i =1
|x 2 (i)|
(25.15)
FT = C1 x 2 , where C1 is a constant that is determined by experiments. Another variable to monitor is the error between the cutting force and its estimated value: e = y C T x. Similarly, the threshold value is given by
(25.16)
(25.17)
ET
=C
2
i =1
| e(i) | . (25.18)
25-6
The fault detection problem consists of checking whether the following conditions hold: |e| > ET , |x 2 | > FT . The fault detection based on multiple variables strengthens the reliability of the method. (25.19)
G1(s) =
for test 1,
(25.20)
G2 (s) =
for test 2,
(25.21)
G3 (s) =
for test 3,
(25.22)
G4 (s) =
for test 4,
(25.23)
(25.24)
(25.25)
The observer gain K is chosen as [100 10]T so that A- KC is stable (where the eigenvalues are 100.4987 and 2155.7). By choosing Q = 2I , the Lyapunov equation can be computed, providing an error estimation. The cutting processes according to the aforementioned four tests were used to compare the estimated results and the actual measurement. Figure 25.3 presents the comparisons. The estimated error was found to be within 12 m, which validates the accuracy of the proposed method. It follows that the established observer is suitable for use as a monitoring method to detect faults.
25-7
250 Cutting force 200 150 100 Cutting force 0 20 Time (a) 200 Cutting force Cutting force 180 160 140 120 100 0 5 10 Time (c) 15 20 40 60
140 120 100 80 60 40 0 20 Time (b) 180 160 140 120 100 80 0 5 Time (d) 10 15 40 60
FIGURE 25.2 Cutting force obtained from CNC milling center: (a) SS = 800 rpm, fr = 150 mm/min, depth of cut = 1 mm; (b) SS = 1000 rpm, fr = 100 mm/min, depth of cut = 1 mm; (c) SS = 1000 rpm, fr = 200 mm/min, depth of cut = 1 mm; (d) SS = 1200 rpm, and fr = 200 mm/min, depth of cut = 1 mm.
FIGURE 25.3 Comparison of actual and estimated cutting forces: (a) SS = 800 rpm, fr = 150 mm/min, depth of cut = 1 mm; (b) SS = 1000 rpm, fr = 100 mm/min, depth of cut = 1 mm; (c) SS = 1000 rpm, fr = 200 mm/min, depth of cut = 1 mm; (d) SS = 1200 rpm, and fr = 200 mm/min, depth of cut = 1 mm.
25-8
25.5 Conclusion
CNC machines used in automated manufacturing are mechatronic systems. In this chapter, a fault detection method was developed based on a state-observer model for a milling machine in a CNC machining center. Specically, a state-observer model of the cutting force was used to detect tool wear in milling operations, with satisfactory results. In this method, only the cutting force was used for monitoring the automated machining. The inexpensive technique based on the observer model was applied to a CNC milling center. Experimental results showed that the proposed method would provide robust performance and could be easily used to monitor tool wear.
Acknowledgment
The authors thank Dr. Wang Wenhui for his helpful suggestions and the effort in data collection.
References
1. Rangwala, S., Liang, S., and Dornfeld, D., Pattern recognition of acoustic emission signals during punch stretching, Mechanical Systems and Signal Processing, Vol. 1, 321332, 1987. 2. Weck, M., Machine diagnostics in automated production, Journal of Manufacturing Systems, Vol. 2, 101106, 1983. 3. Byrne, G., Dorned, D., Inasaki, I., Ketteler, G., Konig, W., and Teti, R., Tool condition monitoring (TCM)the statue of research and industrial application, Annals of the CIRP, Vol. 44, 541567, 1995. 4. Altintas, Y., Prediction of cutting forces and tool breakage in milling from feed drive current measurements, ASME Journal of Engineering for Industry, Vol. 114, 386392, 1992. 5. Li, X., Djordjevich, A., and Venuvinod, P.K., Current-sensor-based feed cutting force intelligent estimation and tool wear condition monitoring, IEEE Transactions on Industrial Electronics, Vol. 47, 697702, 2000. 6. Li, X., Li, H.X., Guan, X.P., and Du, R., Fuzzy estimation of feed-cutting force from current measurementa case study on intelligent tool wear condition monitoring, IEEE Transactions on Systems, Man, and Cybernetics, Part C, Vol. 34, 506512, 2004. 7. Coker, S.A. and Shin, Y.C., In-process control of surface roughness with tool wear via ultrasonic sensing, Proceedings of American Control Conference, Seattle, WA, 1995, pp. 17171721. 8. Prateepasen, A., Au, Y.H.J., and Jones, B.E., Acoustic emission and vibration for tool wear monitoring in single-point machining using belief network, Proceedings of IEEE Instrumentation and Measurement Technology Conference, Budapest, Hungary, 2001, pp. 15411546. 9. Isermann, R., Ayoubi, M., Konrad, H., and Rei, T., Model based detection of tool wear and breakage for machine tools, International Conference on Systems, Man and Cybernetics, Le Touquet, France, Vol. 3, 1993, pp. 7277. 10. Danai, K. and Ulsoy, A.G., An adaptive observer for on-line tool wear estimation in turning, Part I: Theory, Mechanical Systems and Signal Processing, Vol. 1, 211225, 1987. 11. Danai, K. and Ulsoy, A.G., An adaptive observer for on-line tool wear estimation in turning, Part II: Results, Mechanical Systems and Signal Processing, Vol. 1, 227240, 1987. 12. Luo, R.C. and Kay, M.G., Multi-sensor integration and fusion in intelligent systems, IEEE Transactions on Systems Man and Cybernetics, Vol. 19, 901931, 1989. 13. Noori-Khajavi, A. and Komanduri, R., On multisensor approach to drill wear monitoring, Annals of the CIRP, Vol. 42, 7174, 1993. 14. Lauderbaugh, L.K. and Ulsoy, A.G., Model reference adaptive force control in milling, ASME Journal of Engineering for Industry, Vol. 3, 1321, 1989. 15. Park, J. and Ulsoy, A., On-line tool wear estimation using force measurement and a nonlinear observer, ASME Journal of Dynamic Systems, Measurement, and Control, Vol. 14, 666672, 1992. 16. Kim, T.Y. and Kim, J., Adaptive cutting force control for a machining center by using indirect cutting force measurements, International Journal of Machine Tools and Manufacture, Vol. 36, 925937, 1996.