High-speed five-axis milling of hardened tool steel - Ewp.rpi.edu

International Journal of Machine Tools & Manufacture 40 (2000) 869–885

High-speed five-axis milling of hardened tool steel C.E. Becze a, P. Clayton a, L. Chen b, T.I. El-Wardany b, M.A. Elbestawi a

a,*

Intelligent Machines and Manufacturing Research Centre (IMMRC), Department of Mechanical Engineering, McMaster University, Hamilton, Ontario, Canada L8S 4L7 b United Technologies Research Center, East Hartford, CT, USA Received 18 February 1999; accepted 27 September 1999

Abstract This paper investigates critical issues related to high-speed five-axis milling of hardened D2 tool steel (hardness HRc 63). A forging die cavity was designed to represent the typical features in dies and molds and to simulate several effects resulting from complex tool path generation. Cutting tool materials used were coated carbide for the roughing and semi-finishing processes and polycrystalline cubic boron nitride (PCBN) for the finishing process. The effects of complex tool paths on several critical machining issues such as chip morphology, cutting forces, tool wear mechanisms, tool life and surface integrity were also investigated. The main tool failure mode was chipping due to the machine tool dynamics. A five-axis analytical force model that includes the cutter location (CL) data file for computing the chip load has been developed. The effect of instantaneous tilt angle variation on the forces was also included. Verification of the force model has been performed and adopted as a basis for explaining the difficulties involved with high-speed five-axis milling of D2 tool steel.  2000 Elsevier Science Ltd. All rights reserved. Keywords: High-speed milling; Dies and molds; Cubic boron nitride (CBN)

1. Introduction High-speed machining (HSM) is recognized as an enabling manufacturing technology for higher productivity and throughput, and is thus creating considerable interest for die and mold manufacture. Several investigations have been performed to study the feasibility of HSM of dies and molds in their hardened state [1–9]. These studies involved several issues related to HSM such as the chip morphology, the tool wear mechanisms, the effect of process parameters on cutting forces, * Corresponding author. Tel.: +1-905-525-9140 ext. 24991; fax: +1-905-572-7944. E-mail address: [email protected] (M.A. Elbestawi)

0890-6955/00/$ - see front matter  2000 Elsevier Science Ltd. All rights reserved. PII: S 0 8 9 0 - 6 9 5 5 ( 9 9 ) 0 0 0 9 2 - 9

870

C.E. Becze et al. / International Journal of Machine Tools & Manufacture 40 (2000) 869–885

surface finish, geometric tolerance, tool life, etc. However, the results obtained from these studies were limited to tests performed using straight cuts (i.e., production of a workpiece with flat surfaces) [1–8]. In five-axis milling, the cutting process involves continuous changes in the cutting geometry and, consequently, variations in forces generated, associated tool deflections and thermal loads [9]. Contours of small radii of curvature require constant changes in the direction of the tool path, and thus require rapid deceleration or acceleration of the machine tool axes. The flexibility inherently associated by the fourth and fifth axes of the machine/workpiece system causes significant vibrations, which in turn may lead to tool run-out problems, or have a significant impact on tool life as well as surface finish and accuracy. The dynamics of the process will affect tool performance detrimentally and shift the main mode of tool failure to brittle fracture (chipping) of the cutting edge. This paper examines the cutting performance of carbide and polycrystalline cubic boron nitride (PCBN) tools in the high-speed five-axis machining of D2 tool steel in its hardened state. Coated carbide tools were used for rough milling of the pocket, and PCBN tools were used for the finishing process. A five-axis force model was developed and adopted as a basis for explaining the difficulties involved with high-speed five-axis milling of D2 tool steel. 2. Experimental procedure 2.1. Three-dimensional (3D) solid modeling and numeric control (NC) preparation of the pocket The pocket designed for this investigation represents the features of a medium-sized die or mold. A 3D commercial CAD/CAM system was used to generate the free-form surface of the pocket depicted in Fig. 1. The surface was designed such that tool gouging was eliminated. The geometry of the surface was then transferred into the NC programming system to generate a five-axis/one-way flow line tool path. Flow line machining was utilized to provide uniform scallop heights around the surface, facilitate the editing of the post-processor program, allow use of the down milling process, and enable stopping of the machine after each 10th pass to gage

Fig. 1. Sculptured surface with typical features of dies/molds and tool path directions.


871

tool life. The instantaneous variation of the tilt angle was minimized by keeping the culler location (CL) data points very close together, nominally 0.025 mm apart for finishing and 0.254 mm for roughing and semi-finishing. Setting the CL data points much closer affected the prescribed feed rates, which were rarely achieved by the machine controller due to processing limitations, as well as excessive rotations inherent in five-axis machining. The pre-specified feed was incorporated in the post processor to yield inverse time feed commands. This was necessary because the machine controller cannot interpret conventional (mm/min) feed commands in rotary motion even though the physical interpolation between consecutive points is linear. 2.2. Experimental set-up Dry milling tests were performed using ball nose end mills. The workpiece material was D2 (63 HRc) tool steel prepared in the form of 50 mm×100 mm×300 mm blocks. Coated carbide inserts were used for roughing and semi-finishing of the pocket. The cutter diameter was 12.7 mm and the rake angle was 0°. For the finishing process, 0.91 volume fraction PCBN with cobalt binder and sharp edge preparation tools were used. The cutter diameter was 38.1 mm and rake angle was ⫺10°. All tools were balanced for high spindle speeds. The optimum cutting conditions for both roughing and finishing processes were obtained from previous investigations [1,10] conducted for straight milling tests. The spindle rotational speed was kept constant (6000 rev/min for roughing and 10,000 rev/min for finishing) for all tests. The cutting conditions are listed in Table 1. The cutting force components (Fx, Fy, Fz) were measured using a Kistler type 9255b table force dynamometer. The dynamometer was secured to the rotary table using the recommended eightbolt pattern, and provided a flat response up to 2.4 kHz. At the end of each 10th pass the tool was examined for chipping, breakage, notch wear or material sticking on the tool rake face using a toolmaker’s microscope of 20× magnification. The parameters measured to represent the progress of the wear were average and maximum tool wear, VBb and Vbmax, respectively. The position of the maximum tool wear on the cutting edge was recorded. At the end of tool life, the tools were examined by scanning electron microscopy (SEM) to study the different mechanisms of tool wear. Chips were collected and mounted in an epoxy metallurgical mount, polished, etched and examined by optical microscopy at high magnifications. The surface roughness was measured after each 10th pass.

Table 1 Cutting conditions for roughing and finishing Speed, V (m/min)

Roughing Semi-finishing Finishing

Minimum

Maximum

101 101 207

238 170 1197

Feed, f (mm/tooth)

Axial depth of cut (mm)

Radial width of cut (mm)

0.1016 0.1016 0.0508

1.905 1.25 0.635

0.635 0.635 0.254

872


3. Experimental results and discussions The variation of the instantaneous tilt angles (one of the main features in five-axis machining), as well as geometric influences of the die surface topography, cause a non-uniform chip load during milling of the cavity. Also, variable axial depths of cut occur due to the cornering of the tool path, combined with the various rotations of the rotary axis. These phenomena affect the pattern of chip morphology, tool wear mechanisms, cutting forces and tool life. 3.1. Chip formation mechanism in roughing The chip formation process in five-axis roughing was similar to linear cut test data [10]. All roughing conditions produced saw-toothed chips with a white layer formed on the secondary shear zone. However, the size of the chips produced during the five-axis roughing process varied substantially, indicating a variation in the chip load. Fig. 2 shows chip cross-sections produced by sharp and worn tools. Shear lines in the white layer were observed irrespective of the tool condition. In the white layer, the material flowed around the large M23C6 alloy carbide particles. The smaller spherical carbide particles residing on or in the shear bands can affect the forces due to the Orowan stress created by a particle in the path of a mobile dislocation. An increase in the forces was observed, and consequently the width of the white layer also increased. 3.2. Chip formation mechanism in finishing Non-uniform saw-toothed chips were formed during five-axis finishing of D2 tool steel with PCBN tools. The chips produced were blue at the thickest portion, whereas the thin sections were straw-colored. During milling of the wall (attributed to the maximum cutting speeds of 1197 m/min), the chips produced were white hot and, as tool wear progressed, the chips produced appeared oxidized, characterized by a dull finish. Excessive chip thinning was also observed at high cutting speeds as seen in Fig. 3. The maximum chip thickness was smaller than the mean grain size of the workpiece material because of the low feeds. Several chips were also observed where melted chip segments adhered to the free surface of the chip in the form of recrystallized

Fig. 2. Chip cross-section produced during roughing.


Fig. 3.

873

Chip cross-section produced during finishing.

spheres. This indicates that extremely high temperatures were generated in the primary and secondary shear zones, greater than the melting temperature of the tool steel. Thus, significant thermal energy has been produced by the shearing process. Compared to roughing with carbide tools, little or no white layer is evident for the chips produced by the PCBN tool, indicating that the white layer is dependent on the tribological characteristics of the cutting tool material. The chips produced from the wall, on the other hand, clearly revealed a white layer in the secondary shear zone (due to the high temperatures generated). 3.3. Tool wear and tool life Irrespective of the tool wear mechanisms, or tool failure mode, uniform wear patterns were observed in straight milling. This is due to uniform contact of the cutter in the same area throughout the cut. For instance, when milling linear cuts, the depth of cut notch was predominant at the highest cutting speed zone [10]. However, when five-axis milling was performed, variable tool wear patterns were always observed due to the variation of the true chip load and in-cut geometry. In addition, the tilt angle was constantly changing, causing shifting of the contact area of the maximum engagement point. 3.3.1. Modes of tool failure in five-axis roughing of D2 tool steel Flank wear was the main wear mechanism in roughing. However, the tool failure mode was largely chipping at the lowest cutting speed zone (101 m/min). Fig. 4 illustrates the tool failure mode for five-axis roughing. As the tool wear progressed to approximately 0.3 mm flank wear, sparks were produced from the worn tool and the cutter was much more susceptible to chattering as the tool exited a corner into a straight cut due to higher chip loads. Chatter was detected through surface roughness observation as well as characteristic chatter noise. 3.3.2. Modes of tool failure in five-axis semi-finishing of D2 tool steel Fig. 5 shows an SEM image of the carbide tool wear mechanisms. Flank and crater wear as well as material adhesion on the cutting edge were observed. Owing to the excessive vibration observed during the milling of D2 tool steel, the tools failed prematurely and repeatedly by extensive chipping in the middle of the length of contact. Flank wear never progressed to end of tool life as in linear cutting tests.

874


Fig. 4.

Tool wear mechanisms during roughing.

Fig. 5. Tool wear mechanisms during semi-finishing.

3.3.3. Modes of tool failure in five-axis finishing of D2 tool steel Nose wear was the main mechanism of tool wear during the finishing process with PCBN. Plastic deformation was observed on the cutting edge as shown in Fig. 6. During the milling of the wall, the tool nose wear progressed to 0.08 mm, which was attributed to the maximum cutting speeds of 1197 m/min. Catastrophic failure, characterized by microchipping of the cutting edge, occurred without prior indication of imminent failure. Edge chipping occurred in the cutting speed region of approximately 700–1000 m/min and can be correlated to the vibration of the machine/workpiece system. Fig. 7 shows a comparison of the tool life obtained during the straight and five-axis milling of D2 tool steel with the same cutting conditions. Comparable tool life was achieved in both roughing and semi-finishing processes. However, tool life was 1.5 times longer in five-axis finishing of D2 tool steel. During straight milling, uniform contact of the cutter in the same area throughout the


Fig. 6.

875

Tool wear mechanisms during finishing process.

Fig. 7. Tool life in machining D2 tool steel.

cut occurs. Hence, depth of cut notch wear was formed in the highest speed zone and became predominant. During five-axis finishing, the tilt angle constantly varies, changing the contact area of the maximum point of engagement. Therefore, depth of cut notching was not as predominant even during milling of the back wall of the pocket, where severe cornering and generally larger axial immersions were required. Premature failure of the PCBN tools was mainly due to the machine dynamics, specifically the rigidity of the fourth axis (the tilting axis about the x-direction). A free-vibration trace of the test set-up (workpiece, fixture, dynamometer and rotary table) [Fig. 9(b)] shows the natural frequency to be 675 Hz.

876


3.4. Surface roughness Surface roughness was tracked in several areas, for both the sharp and worn tools (0.08 mm flank wear), and prior to chipping. The distribution of the measurements on the milled surface is presented in Fig. 8. An average surface finish of 0.3 µm was achieved, increasing to 1 µm at the wall, even with the effect of vibrations present due to the machine dynamics. 3.5. Cutting forces Fig. 9(a) shows the cutting forces measured for cutting speeds of 10,000 rev/min and 10° tilt angle. In general, high dynamic force components were observed during milling of D2 tool steel. This phenomenon was not recognized during milling of H13 tool steel [10]. The high-frequency dynamic force component present during milling of D2 tool steel was caused by the high density of the carbide particles in the material matrix and also could be related to the dynamics of the cutting tool. The magnitude of the dynamic component of the cutting forces was almost of the same order as that of the steady-state components. In general, a stable cutting process was not achieved even after employing a wide range of cutting conditions. This dynamic instability will facilitate the probability of tool chipping and premature catastrophic failure. Another problem encountered during the milling process was the inability of the milling

Fig. 8. Surface finish produced in machining D2 tool steel.


877

Fig. 9. (a) Unfiltered force trace showing cutting dynamics for finishing. (b) Free-vibration force trace of rotary table/workpiece/dynamometer set-up.

machine to provide the desired feed rate. The feed rate was not maintained when the die topography required large angular rotations of the fourth and fifth axes. The maximum angular velocity of the rotary table was less than that required for maintaining the desired feed rate. This is a potential problem with all milling operations utilizing more than three axes. The reduction of the feed rate at the corner is expected to reduce the actual chip load and consequently increase the tool life as shown in Fig. 7. For a better understanding of the change in force pattern generated during high-speed five-axis machining and consequently the change in the instantaneous stresses applied on the tool, the five-axis milling process was simulated and the actual cutting forces were predicted and verified experimentally. 3.6. Force model development A mechanistic force model able to predict cutting forces for any five-axis configuration was developed. The cutting conditions and tool paths specified by the five-axis CL data file, in addition to the equation describing the cutting edge, are the main requirements in predicting the instantaneous cutting forces. Furthermore, the calibration process is extremely simple, requiring only a force trace from a single cutting pass. A back-propagation neural network program then determines the optimum calibration constants for the cutter/workpiece combination. The force model presented does not take into account the dynamics of the machining process. Developing a dynamic five-axis force model was not the main focus of this paper. For a fiveaxis dynamic milling model to be accurate, the dynamics of the tool and spindle, as well as the flexibility of the rotary table, must be taken into account. The force system considered at any point along the cutting edge was modeled as an oblique equivalent of an orthogonal machining process. The magnitudes of the tangential (Ft) and radial (Fr) forces are modeled as being proportional to the area of the material being removed (A). The force relations are: Ft⫽Kt·b·t⫽Kt·A

and Fr⫽Kr·b·t⫽Kr·A,

(1)

where Kt and Kr are empirical tagential and radial cutting force constants, t is the instantaneous chip thickness and br is the radial width of cut. The area A is then projected back on to the primary axes and hence the principal forces are determined, which is fairly straightforward in three-axis milling [11–14]. However, this becomes considerably more difficult with five-axis mill-

878


ing since the chip load changes continuously as the tilt angle changes with respect to the surface geometry. To obtain a uniform division of the chip load, it is convenient to use spherical coordinates: q, f and R (the cutter radius). This provides a more uniform discretization of the cutting edge and increases the accuracy in calculating the cutting forces. Each point along the cutting edge is determined for a given angle q and f, as illustrated in Fig. 10. Once the points have been calculated with respect to the tool coordinate system (TCS), they are then transformed into the workpiece coordinate system (WCS). This is where the spindle rotation (⌽), A-axis rotation (a) and B-axis rotation (b) are included. Each of the points along the new surface is then tested to determine if it is in the cut or not. The matrices defining the transformation of the edge points to the workpiece coordinate system are as follows:

冤

1 0

冥冤

cos b 0 sin b

0

a⫽ 0 cos a −sin a , b⫽ 0 0 sin a cos a

1 0

冥

−sin b 0 cos b

冤

cos ⌽ −sin ⌽ 0

冥

and ⌽⫽ sin ⌽ cos ⌽ 0 . 0

0

(2)

1

The actual matrix operation appears as

冤冥 x y z

WCS

冤冥 x

⫽[b][a][⌽] y z

.

(3)

TCS

Fig. 10. (a) Ball-end mill showing discretized cutting edge. (b) Relative position of angles q and f with respect to the cutting edge as well as the angles of rotation about the primary axes.


879

The most difficult task in modeling the five-axis force is the determination of the equations defining the chip volume removed by each cutting edge per revolution [14]. A new chip volume is bounded by three distinct geometries. The top surface of the chip is defined by the previously milled surface. For the case where the cutter radius is much greater than the radial width of cut, br, the top surface can be approximated as a plane parallel to the xy-plane. In Fig. 11, the surface formed by the line AB is the previous milling pass and can be approximated as a cylinder. This pass is composed of many tricoidal cutter rotations which on a larger scale approximate a cylinder, see curve DE in Fig. 14. The arc BC defines the surface formed by the previous cutting edge. The points defining the edges of the chip volume change as the depth of cut changes for a given cutter. Defining the width of cut, br, and feed per tooth, f, as constants, a cross-section of chip area can be exposed in the xy-plane at a depth of z. This cross-section of chip area changes in the axial direction and is dependent only on the working radius, Rw=√R2−z2. The points defining the chip volume boundary where the depth brⱕ2·Rw are A, B and C, and can be presented as follows: A: B:

C:

冑 (f−冑2·R

( 2·Rw·br−br, Rw−br) ·br−br, Rw−br)

w

.

冢冪冣

(4)

f2 R2w− 4

f , 2

As the depth increases the working radius decreases until br=2·Rw and the points A and B become coincident (Fig. 12):

A, B:

冢

冪

f ,− 2

冣

f2 R2w− 4

.

C:

(5)

冢冪冣 f , 2

f2 R2w− 4

Fig. 11. Chip shape when Rw⬎1/2br.

880


Fig. 12. Chip shape when Rw⬍1/2br.

The angle of engagement becomes more obtuse as the working radius decreases to that of half the feed per tooth, Rwⱕ1/2f, at which time all three points become coincident (Fig. 13). In threeaxis milling this chip volume is deformed due to the plowing effect of the tool tip. In four- and five-axis milling the tilt angle inclines the tip out of the cut and no rubbing occurs: A, B, C: (Rw, 0).

(6)

The total instantaneous chip area is the sum of the individual areas at a given cutter rotation angle, ⌽. The individual areas are calculated as the difference between the triangles composed of the outside (P0, P2, P3) and inside surfaces (P0, P1, P4) of the newly formed chip as illustrated in Fig. 14. This area is then projected back on to the primary axes using Eqs. (7) and (8). The equations representing the instanteous chip load area have been greatly simplified by defining point P0 as the origin of the both WCS and TCS. 1 1 AX⫽ ·(y1·z2⫹y4·z3⫺y2·z1⫺y3·z4) AY⫽ ·(z1·x2⫹z4·x3⫺z2·x1⫺z3·x4) 2 2 1 AZ⫽ ·(x1·y2⫹x4·y3⫺x2·y1⫺x3·y4) 2 and

Fig. 13. Chip shape when Rw⬍1/2f.


881

Fig. 14. Illustration of a sector of the instantaneous chip load area.

冤冥冤 FX

冥冤冥

K11 K12 K13 AX

FY ⫽ K21 K22 K23 AY . FZ

(8)

K31 K32 K33 AZ

Substituting AX, AY and AZ in the following equation produces the instantaneous cutting force in the X-, Y- and Z-directions, respectively. Kij represent the force constants determined experimentally, which are dependent on process parameters such as cutting speed, feed, effective cutting angles, tool geometry, instantaneous tool angle, etc. 3.6.1. Calibration methodology An efficient calibration methodology has been developed in this investigation. As mentioned earlier, the force coefficients Kij can be represented as an empirical equation, Kij ⫽ea0·tac 1·bar 2·aan3…

(9)

Fig. 15 illustrates the technique used for calibrating the force model. The calibration process is based on a back-propagation neural network algorithm. The neural network is composed of a single-layer, fully connected, linear neural network, having both three input and three output variables. For each tool/workpiece material combination, in a specified speed range, a force trace is required from a single linear cutting pass to determine the optimum calibration constants. In the present work the values for the K matrix were found to be

882


Fig. 15. Flow chart of calibration strategy and inputs.

冤

397

−170 −1052

KRoughing⫽ −1415 665 −1062 776

冥

−2032 N/mm2. 16162

(10)

Figs. 16 and 17 show the predicted and measured cutting forces for roughing and finishing operations, generated during milling of the adopted surface. The model is able to predict the static cutting forces well but does not consider the dynamic component. (This is best illustrated in the z-axis of the force traces.) As a result there is some deviation of the predicted cutting force from the actual forces. The rotary table used to provide the fourth and fifth axes of motion was unable to provide the desired feed rate during milling at the high angular velocities required by the die topology. As a result, a lower-than-desired feed rate was achieved over portions of the die. Fig. 18 illustrates the discrepancy between the predicted and measured forces due to the inability of the rotary table to provide the required angular velocity while milling tight radii. The result is lower forces and longer machining times.


Fig. 16.

Fig. 17.

883

Roughing pass cutting forces: DOC=1.9 mm, WOC=0.64 mm, f=0.1 mm, N=6000 rev/min, a=25°.

Finishing pass cutting forces: DOC=0.64 mm, WOC=0.25 mm, f=0.05 mm, N=10,000 rev/min, a=10°.

884


Fig. 18. Roughing pass cutting forces where feed rate is limited by the angular velocity of the rotary table: DOC=1.9 mm, WOC=0.64 mm, f=0.1 mm, N=6000 rev/min, a=25°.

4. Conclusions Several important issues regarding the comparison of high-speed five-axis milling with linear cutting tests have been reported. In five-axis milling, the tool life would be expected to increase due to the variation of the contact area. This extended tool life was not achieved for roughing or semi-finishing operations owing to vibrations caused by the flexibility of the machine tool structure. An increase in tool life was achieved for the finishing process. The tool life for PCBN at cutting speeds in excess of 1100 m/min was also found to be acceptable. The main tool failure mode was catastrophic chipping of the cutting edge for all cutting operations. The wear mechanisms observed were flank wear, crater wear, plastic deformation and adhesive wear. The chip formation process was not affected by five-axis milling. Excellent surface finish has also been achieved, even with the presence of the vibrations. The chip morphology varied in relation to the geometry of the cut resulting from the surface topology. The five-axis mechanistic force model presented correlates well with the measured forces in both roughing and finishing. It is only during the areas of high angular velocity where the feed rate is reduced, that the force signals diverge. The model employs an efficient calibration technique utilizing a back-propagation neural network. Acknowledgements The authors wish to acknowledge the help of Dr H.A. Kishawy with the experimental analysis.


885

References [1] M.A. Elbestawi, L. Chen, C.E. Becze, T.I. El-Wardany, High-speed milling of dies and molds in their hardened state, Annals of the CIRP 46 (1) (1997) 57–62. [2] T. Ikeda, I. Takahashi, T. Matsuoka, T. Nakagawa, Ultra high speed milling of die steel with ball nose end mill, in: Proceedings of 2nd International Conference on Die and Mold Technology, Singapore, September, 1992, pp. 48–56. [3] J. Muller, High speed milling in hardened tool steel using common CNC-machines, in: 1st French and German Conference on High Speed Machining, University of Metz, France, 1997, pp. 403–406. [4] H. Schulz, S. Hock, High-speed milling of dies and molds — cutting conditions and technology, Annals of the CIRP 44 (1) (1995) 35–38. [5] C. Le Calvez, F. Chenou, S. Gunnarsson, High speed milling of hardened tool steel with CBN tool, in: 1st French and German Conference on High Speed Machining, University of Metz, France, 1997, pp. 377–384. [6] S. Hock, High speed cutting (HSC) in die and mould manufacture, in: 1st French and German Conference on High Speed Machining, University of Metz, France, 1997, pp. 274–283. [7] M. Dumitrescu, T.I. El-Wardany, M.A. Elbestawi, L. Chen, Critical assessment of carbide and PCBN tool performance in high speed milling of dies and molds, Trans. NAMRI–SME XXVI (1998) 183–188. [8] H. Schulz, T. Moriwaki, High-speed machining, Annals of the CIRP 41 (2) (1992) 637–643. [9] T. Altan, B.W. Lilly, J.P. Kruth, W. Konig, H.K. Tonshoff, C.A. Van Luttervelt, A.B. Khairy, Advanced techniques for die and mold manufacturing, Annals of CIRP 42 (2) (1993) 707. [10] J. Boehner, M. Dumitrescu, T.I. El-Wardany, M.A. Elbestawi, L. Chen, Influence of material microstructure on tool performance in high speed machining, Trans. NAMRI–SME XXVII (1999) 129–134. [11] F. Abrari, M.A. Elbestawi, Closed form formulation of cutting forces for ball and flat end mills, Int. J. Mach. Tools Manufact. 37 (1) (1997) 17–27. [12] M. Yang, H. Park, The prediction of cutting force in ball-end milling, Int. J. Mach. Tools Manufact. 31 (1) (1990) 25–54. [13] Z. Yazar, K. Koch, T. Merrick, T. Altan, Feed rate optimization based on cutting force calculations in 3-axis milling on dies and molds with sculpted surfaces, Int. J. Mach. Tools Manufact. 34 (3) (1994) 365–377. [14] K. Weinert, A. Enselmann, J. Friedhoff, Milling simulation for process optimization in the field of die and mould manufacturing, Annals of the CIRP 46 (1) (1997) 325–328.