Robust Feedrate Selection for

3-Axis NC Machining Using

B. K. Fussell Discrete Models
R. B. Jerard
This research effort is focused on improving the efficiency of CNC machining by auto-
J. G. Hemmett matic computer selection of feedrate for 3-axis sculptured surface machining. A feedrate
process planner for complex sculptured end milling cuts is developed from mechanistic
Department of Mechanical Engineering, and geometric end milling models. The selection program uses tool deflection, surface

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

University of New Hampshire, finish, tool failure and machine power to set constraints on the cutting force and the
Durham, NH 03824 feed-per-tooth for rough, semi-finish, and finish passes. A NC part program is processed
one tool move at a time by the planner. For each tool move a geometric model calculates
the cut geometry, and an inverse mechanistic model uses this information along with the
constraint force to calculate a desired feedrate. The feedrate is written into the part
program resulting in a file that contains a feedrate for each tool move. Experimental
results for a sculptured surface show the accuracy of the algorithms in maintaining a
desired force. 关DOI: 10.1115/1.1365398兴

1 Introduction removal rate under the constraints of form error and surface
roughness. The instantaneous geometry of the cut for complex
The focus of this research is the integration of geometric and
surfaces is not considered.
mechanistic end milling models for automatically setting feedrate Spence and Altintas 关12兴 as well as Bailey et al. 关13兴 developed
for 3-Axis CNC sculptured surface machining that result in safe, a 2 1/2 axis process simulation and planning system that utilize
accurate, and efficient tool movements 关1兴. Sculptured surface ma- solid modelers for the workpiece geometry description. This pro-
chining feedrate is difficult to determine accurately because of the vides a method for determining the volumetric intersection of the
continuously changing geometry of the cut. For example, standard tool with the workpiece. Feedrates are scheduled through use of
practice for machining automobile stamping dies is to set a con- the tool/workpiece intersection data provided by the solid modeler
stant feedrate for a number of cutter paths, based on the worst and a ball end mill mechanistic model. Mounayri, Spence, and
case cut geometry. This provides a conservative cut, but does so at Elbestawi 关14兴 extended the force and torque simulation capability
the expense of efficiency. Variable feedrate that maintains safe of a solid modeler system to 3-axis milling of complex parts. They
cutting could provide substantial time savings to manufacturers. use a cubic Bezier representation of the cutting edges, and inter-
Research work on process planning by Sungurtekin and Voel- sect this with the swept tool volume. Experimental verification of
cker 关2兴 as well as Vann and Cutkosky 关3兴 indicate the importance the system is shown for 2 1/2 axis semifinish ball end milling of a
of using the desired cutting force in determining the machining die.
feedrate. Some of the first work on feedrate planning was by This paper describes a feedrate planning system that selects
Wang 关4兴, where he used a Z-buffer representation of the work- feedrate for 3-axis sculptured milling by integrating a Z-buffer
piece and a simple volumetric model to relate cutting force to the geometric model with a discrete mechanistic model of the cutter
metal removal rate 共MRR兲. However, the MRR model is limited 关15兴. The geometric model describes the contour surface of the
since the force direction cannot be determined. During semi-finish workpiece and provides the framework for calculation of the tool
and finish cutting it is important to know the magnitude and di- and workpiece intersections. The intersection information is used
rection of the cutting force. Takata et al. 关5兴 utilized a Z-buffer to find the contact area of the tool with the workpiece. The dis-
approach for the workpiece/cutter geometry description and com- crete mechanistic model can then be used to calculate a feedrate
bined this with the mechanistic model of Kline et al. 关6兴 to inves- for a desired cutting force vector for each tool movement. A can-
tigate cutting forces and tool deflection. Further work by Takata tilever beam model of the tool is used to relate the vector force to
关7兴 resulted in a process planner for 2 1/2 axis flat end milling, allowable tool deflection or tool bending stress. Other process
using a solid modeler for intersection calculations. constraints such as tooth failure, machine power, and surface fin-
Lim and Menq 关8兴 created an integrated planner for machining ish can also be used to limit the feedrate.
complex surfaces that optimized the cutting path and feedrate. The next two sections of this paper describe the discrete mecha-
They used control points and a desired cut geometry rather than nistic and geometric models of end milling. Model integration and
feedrate calculation are then described in Section 4. Finally, ex-
calculating the instantaneous geometry. Lazoglu and Liang 关9兴
perimental results are presented and discussed in Section 5.
proposed a feedrate optimization scheme for complex surfaces
that uses a frequency domain mechanistic model. They maximize
the MRR under the constraint of a maximum resultant force. Al- 2 Discrete Mechanistic Modeling
tan et al. 关10兴 have developed a process planner that keeps the
chip load constant during sculptured surface finish machining by The discrete mechanistic end milling model can be used to
adjusting both the feedrate and spindle speed. Stori, Wright, and estimate instantaneous force magnitude and direction. It can there-
King 关11兴 developed a planning system that integrates analytical fore be used to estimate tooth stresses, cutter shank stresses and
and simulation models of milling in order to maximize the metal cutting tool deflection. A discrete flat end mill model was devel-
oped by Kline and DeVor 关16兴 by slicing the cutter into a series of
Contributed by the Manufacturing Engineering Division for publication in the
axial discs. The vector force contribution from each disc flute
JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING. Manuscript received segment that is in contact with the workpiece is summed to find
Jan. 1999; revised June 2000. Associate Editor: M. A. Elbestawi. the cutting force as a function of cutter rotation angle. This allows

214 Õ Vol. 123, MAY 2001 Copyright © 2001 by ASME Transactions of the ASME
the addition of runout, cutter tilt and variable cutting constants in where ␤ describes the angle between the feedrate vector f D and
calculating the force vectors and torque on the tool. The contact the tool coordinate x axis. Note that Eq. 共2兲 is only meaningful for
area can be different for each disc and is calculated from the positive values of h( ␾ ,z).
geometric model of the cutter and the workpiece 关17兴. The mechanistic model is separated into spherical and cylindri-
A discrete ball end milling model, based on orthogonal cutting, cal sections for accurate prediction of the x and y forces acting on
was proposed by Yang and Park 关18兴. They determined that the the cutter. The approach to force calculation is the same for each
cutting model constants vary with the rake angle, cutting velocity, section; the tooth force during cutting is considered proportional
and the feedrate. Yucesan and Altintas 关19兴 also developed a dis- to the chip thickness. The predicted force of a differential tooth
crete mechanistic ball end milling model. For a known tool geom- element has both tangential and radial components, located in the
etry, they were able to accurately predict all three force compo- x-y plane of Fig. 1, that contain the pressure and friction constants
nents for a variety of cuts. Four constants are required for the of Yucesan and Altintas 关19兴 as well as the geometry of the tooth
model, based on pressure and friction loads on the rake and flank element, e.g. rake angle and relief angle. The force per unit flute
faces. length is given by:
A brief description of our method for implementing the mecha-
⌬F t ⫽K t h 共 ␾ ,z 兲

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

nistic model for ball end milling is given below. A more complete
description for a flat end mill model may be found in Choi and ⌬F r ⫽K r ⌬F t (3)
Jerard 共关20兴 pp. 292–301兲. Two concepts are very important in the
model development and implementation: ‘‘chip thinning,’’ which with the coefficients for the cylindrical section described by:
is a method of determining the equivalent chip thickness any- K t ⫽K tc 共 h a v g 兲 ⫺ p 1c
where on the cutting tool, and ‘‘contact areas’’ which are essential
in determining which portions of the cutter teeth are engaged in K r ⫽K rc 共 h a v g 兲 ⫺ p 2c (4)
cutting at a tool angular position.
where K tc , K rc , p 1c and p 2c are constants that depend on the
The left side of Fig. 1 shows the xyz local coordinate system
workpiece material properties, tooth geometry 共rake angle兲, tool
defined relative to the cutting tool. The milling forces and cut
wear and material temperature, and h a v g is the average chip thick-
geometry are defined relative to this system, which is local to the
ness over the cutter-workpiece engagement area 共note that the
cutting tool and varies with the direction of travel, denoted by unit
engagement area is defined by the contact angles for each disc兲.
vector D. The local x coordinate axis is normal to the axial direc-
Similarly, the spherical section constants are described by:
tion z, with D lying in the local x-z plane. On the right in Fig. 1,
a ball-end cutter is shown making a linear cut of axial depth a and K t ⫽K ts 共 h a v g 兲 ⫺ p 1s
radial depth b, with a feedrate vector f D. The angle ␪ 2 describes
the axial location of the current disc element on the ball. The K r ⫽K rs 共 h a v g 兲 ⫺ p 2s (5)
value of ␪ 2 is considered as zero on the cylindrical portion of the where K ts , K rs , p 1s and p 2s are again dependent on geometry,
cutter, and it increases to a maximum of 90 deg on the bottom of material properties and the average chip thickness.
the ball along the ⫺z axis. All eight constants are determined from cutting tests using the
The chip thickness h( ␾ ,z) at any location on the cutter is: same tool and material that is to be used in machining the actual
f D"Ns 共 ␾ ,z 兲 part. Any number of cutting tests can be used to determine these
h 共 ␾ ,z 兲 ⫽ (1) constants; however we determined that a minimum of eight cuts
n t⍀ are needed to obtain accurate constants. Four tests are made at ball
where n t is the number of teeth, ⍀ is spindle speed, f is the depth with a radial depth of 1.4r 共r is the tool radius兲. Each test is
feedrate, Ns ( ␾ ,z) is the cutter surface unit normal at tooth locat- at a different feedrate so the feed per tooth varies over a range of
ing angle ␾ and axial location z. expected values, e.g. 0.00635 to 0.0254 mm/tooth. An additional
A typical 3-axis example is the cutter moving in the positive x four tests are run at twice the ball depth with 1.4r radial depth.
direction and positive 共or negative兲 z direction. The chip thickness Once again the feedrate is varied for each cut. The ball depth cuts
at any particular tooth location is then: are used to determine the spherical coefficients, and the twice ball
depth cuts are used for the cylindrical terms. The procedure for
f 关 cos ␪ 2 sin ␾ cos ␤ ⫺sin ␪ 2 sin ␤ 兴 determination is similar to that presented by Kline and DeVor
h 共 ␾ ,z 兲 ⫽ (2)
n t⍀ 关16兴.

Fig. 1 Coordinate definition and cut geometry for ball end milling during as-
cending cut

Journal of Manufacturing Science and Engineering MAY 2001, Vol. 123 Õ 215
 The total cutting force can be determined for a given tool po-
sition angle ␪ by:
nz nt

F x共 ␪ 兲 ⫽ 兺 兺 兵 ⫺⌬F cos ␾ ⫺⌬F

i⫽1 j⫽1
t r sin ␾ cos ␪ 2 其 dS

ns nt

F y共 ␪ 兲⫽ 兺 兺 兵 ⌬F sin ␾ ⫺⌬F
i⫽1 j⫽1
t r cos ␾ ␪ 2 其 dS (6)

where ⌬F t and ⌬F r are calculated using Eq. 共3兲, n z is the total

number of cylindrical and spherical disk elements, n t is the num- Fig. 2 Geometric model with toolpath envelope
ber of teeth, and dS is the length of the cutting edge. For the
cylinder portion, dS is equal to dL/ cos(␺), with dL as the uniform
axial slicing in the z direction, and for the sphere, dS is equal to For 3-axis cutting, the ZDVs are always parallel to the tool

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

r d ␪ 2 , the uniform slicing in the ␪ 2 angle 共note that the disk z-axis 共Fig. 2兲 and intersections only have to be considered with
elements along the sphere are not uniform in z thickness兲. Both ␾ the ball end of the tool. This requires finding the intersection of
and ␪ 2 will vary as a function of the axial slice 共i兲, the tooth lines 共ZDVs兲 with the swept volume of the ball end, represented
number 共 j兲 and the tool rotation angle 共␪兲. Model accuracy is by two spheres and a cylinder. The two spheres represent the
demonstrated in the results section of the paper. beginning and end of a tool move, while the cylinder, generated
by a swept circle whose normal lies in the direction of travel,
represents the volume swept by the ball as it moves from begin-
ning to end.
3 Geometric Modeling
The toolpath envelope is used to define a region of possible
3.1 Background. The geometric model performs two im- intersections with the ZDVs. This region is established by project-
portant functions in the feedrate selection process. It is used to ing the tool move initial and final positions into the X and Y
determine the location and size of the contact area between the 共global coordinate system兲 plane, and creating a rectangular re-
cutting tool and the workpiece. It also serves as a dynamic geo- gion that encloses all possible tool positions during the move. If
metric record of the in-process workpiece. An accurate model of an intersection occurs, the radial depth of cut resulting from that
the in-process machining is important to ensure that correct cut ZDV is determined along with the axial discs affected by the
geometry parameters are calculated as the simulation progresses ZDV. The contact regions of those discs are then updated if the
and tool moves occur in regions previously milled. new radial depth of cut enlarges the contact area. The specific
The earliest attempts at geometric modeling of NC machining steps required for each ZDV are listed in Appendix A.
were based on solid modeling 关21,22兴. For the prismatic shapes After all ZDVs have been processed, the angular limits of con-
found in 2 1/2 D machining this was a successful approach, but tact for each disc can be calculated from the maximum and mini-
for sculptured surface machining an approximate method based on mum radial depths as illustrated in Fig. 3. Note that Fig. 3 repre-
‘‘z-buffers’’ was introduced by several researchers 关23兴, and also sents a disc for the spherical section of the cutter, where cutting
successfully implemented in most of the commercially available can occur in all four quadrants of the disc. Also note that the
NC simulators 共e.g. Vericut from CGTech兲. The z-buffer method radius is a function of the disc and diminishes to zero at the
requires finding the intersection of lines with the swept surfaces of bottom of the sphere.
the toolpath envelope while the solid modeling approach requires The radial depths of the first quadrant, y 1 min and y 1 max , are used
surface-surface intersections. While line-surface intersections are to determine the entrance and exit angles of that disc quadrant,
inherently more computationally efficient than surface-surface in- ␾ ent and ␾ ext :
tersections, it is also true that the latter approach is inherently ␾ ent ⫽ ␲ /2⫺sin⫺1 共 y 1 max /r 兲
more accurate due to the exact geometric representation. Im-
proved accuracy can be achieved in the z-buffer approach by re- ␾ ext ⫽ ␲ /2⫺sin⫺1 共 y 1 min /r 兲 (7)
quiring closer spacing between the lines, which of course results
in slower calculation time. With faster computers and improve- where r is the current axial disc radius. The entrance and exit
ments in the efficiency of modern solid modelers there is some angles for the other three quadrants can be found in a similar
resurgence of interest in that approach 关14兴. manner.

3.2 Modeling 3-Axis Cutting. An extended Z-buffer

method is selected for use in the geometric model 关20,24兴. The
workpiece is broken into a set of evenly distributed parallel lines
or discrete vectors, called z direction vectors or ZDVs, with the
length of each vector representing the depth of the workpiece, and
a cross sectional area associated with each vector 共Fig. 2兲. Mul-
tiple values of z can be stored for one vector, allowing extension
to three-axis machining with multiple setups and to five axis ma-
chining. The spacing of the ZDVs is chosen to obtain a desired
level of accuracy in the contact area calculation during a tool
move 关25兴. The toolpath envelope is then modeled as a set of
geometric primitives. This greatly simplifies calculation of the
intersections between the cutting tool and workpiece, reducing it
to a set of simple line-geometric primitive intersection calcula-
tions for 3, 4 or 5 axis cutting. Regions of intersection along the
lines composing the Z-buffer with these primitives are then ‘‘re-
moved’’ from the model 共Fig. 2兲. Intersection information, along
with ZDV location and height, is then used to determine the maxi- Fig. 3 y min and y max are saved for each of the four quadrants
mum and minimum radial depth of cut for each cutter disc, and of each disc. A tooth of a tool rotating in the clockwise direc-
from this, the entrance and exit angles for each disc. tion will enter at the points labeled 1, 3, 5, 7 and exit at 2, 4, 6, 8.

216 Õ Vol. 123, MAY 2001 Transactions of the ASME

 3.3 Discrete Vector Spacing. ZDV point spacing for the to include feedrate for each tool move. The program uses the
work piece geometry is determined based on a ‘‘worst case sce- geometric and mechanistic models to calculate the feedrate for
nario’’ approach. Two distinct cases must be considered, one each tool movement so that the process constraints are satisfied.
where the tool movement is extremely small, and another where The output of the selection program is a new part file containing
the tool movement is much larger than the point spacing. The the calculated feedrate, a file that contains the workpiece geomet-
former case can be a problem because a given tool move may not ric model after the cuts are simulated, and a part file that has the
intersect any ZDVs. This can be handled by using a slightly larger deflected tool positions. Storing the new geometry file enables the
tool radius in the intersection calculations. If the radius is in- selection program to calculate feedrate for another cutting pass,
creased by the ZDV point spacing, any small tool movement will e.g. finish pass, over the workpiece. The deflected tool part file
result in an intersection, providing there is material to be removed. enables us to compare the desired surface geometry to the simu-
Two sets of intersections are now required to maintain an accurate lated cut surface, including static cutter deflection, to check the
model. The geometric model uses the actual radius of the tool to part tolerance.
accurately estimate the cut workpiece geometry, while the en- A block diagram of the feedrate post-processor program is
larged tool is used to calculate the contact angles for each tool shown in Fig. 5. The program is broken into three main sections.

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

movement. The first section reads in the tool path information from the NC
Now consider a large tool move in the X direction, as shown in part file, and prompts the user for the required tool dimensions,
the slot cut of Fig. 4. In the worst case the tool envelope barely workpiece material constants, acceptable tool deflection, desired
misses some vectors, indicated by rectangles in the figure. This geometric model accuracy, cut type 共rough, semi-finish or finish兲,
leads to error in the angle computation of Eq. 共7兲 where y max is and spindle speed. A geometric model consisting of z-direction
the current tool disc radius r minus the vector point spacing a p . vectors is then developed for the entire workpiece block. The
An acceptable level of angle error in percentage, e, can be defined spacing of the vectors is determined by the user defined model
as: accuracy.

冉 冊
The second section of the program takes a single tool move-
100⫺e ment from the part file and finds a feedrate based on a chosen
␣ cal ⫽ ␣ real (8)
100 constraint, for example, maximum tool deflection. The geometric
where ␣ real is the actual angle, i.e. ␲/2 for a slot cut, and ␣ cal is model is used to calculate the tool/workpiece intersection infor-
the calculated angle from the intersection information. Substitut- mation that includes the entrance and exit angle of each discrete
ing from Eq. 共7兲 for ␣ max and defining y 1max as (r⫺a p ) gives: disc of the cutter. The constraint force is then used in an iterative
loop 共In Range block of Fig. 5兲 to calculate a feedrate for this
sin⫺1 冉 冊冉
r⫺a p
r
⫽
100⫺e ␲
100 2 冊 (9)
particular tool move. Note that a precise model of the CNC ma-
chine controller 共machine model block of Fig. 5兲 is necessary to
accurately determine the relative motion between the tool and the
Solving for a p yields: workpiece.

再 冋冉
a p ⫽ 1⫺sin
100⫺e ␲
100 2 冊 册冎 r (10)
The third section checks the selected feedrate by using the re-
sulting forces and deflections to check for tooth failure, machine
power limitations, and surface finish. If the selected feedrate fails
This gives the point spacing, a p , for a given acceptable percent- any of the constraint limits, it is reduced and the checks are
age error e and the tool radius r. Note that this formula is based on reapplied.
a slot cut because it yields the most conservative, i.e. smallest, In the final section of the program, the selected feedrate and its
point spacing of all possible entrance and exit angles. associated x and y deflection values are stored on the appropriate
For a tool move in the world Y-direction the point spacing based tool move line, but in separate part files. The completed feedrate
on percentage error is identical to that for X. For tool moves with part file can be subsequently used to cut the part on a CNC ma-
both X and Y components, the same percentage error results in a
larger point spacing.

4 Model Integration for Feedrate Selection

4.1 Selection Program Description. The feedrate selection
program integrates the geometric and mechanistic models along
with service subroutines such that an NC part file can be revised

Fig. 4 Entrance and exit angle calculation errors resulting

from the discrete z direction vector spacing „slot cut… Fig. 5 Flowchart of feedrate selection

Journal of Manufacturing Science and Engineering MAY 2001, Vol. 123 Õ 217
chine. The deflection file can be used to check for part tolerance values are solved for a rotation angle of 0 to 360 deg in incre-
as described earlier. After storing the information the program ments of 1 deg. The smallest feedrate is chosen for this particular
loops back to analyze the next tool movement. tool movement.
4.2 Constraints and Feedrate Calculations. Constraints
associated with machine power, shank failure, tooth failure, and
surface finish are described in Choi and Jerard 关20兴, Spence and 5 Results
Altintas 关12兴, and Gagne 关26兴. The force constraint, based on an 5.1 Experimental Validation. Experimental verification of
allowable tool deflection can be derived from a cantilever beam the feedrate selection program is made by ball end milling the
deflection model 关27兴: sculptured sine surface shown in Fig. 6共a兲 in eight passes. The
part file for each pass is initially run through the selection pro-
F y ⫽ 共 6EI ␦ y 兲 / 共 2L 3 ⫺3L 2 C f y ⫹C 3f y 兲 gram so that a feedrate can be assigned to each tool move. This
file is then used to cut the sine surface. An evaluation of the
F x ⫽ 共 6EI ␦ x 兲 / 共 2L 3 ⫺3L 2 C f x ⫹C 3f x 兲 (11) selection program is made by comparing the desired resultant

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

where C f x and C f y are the center location of the cutting force on force F r , to the experimentally measured force. Any difference
the tool, in the z direction, and L is the free length of the tool. between these two forces would be indicative of inaccuracies in
Currently, the feedrate algorithm uses the end of the tool as the the geometric/mechanistic simulation models. Experimental cut-
force centers, resulting in an acceptable but conservative deflec- ting forces are also measured for a ‘‘standard practice’’ constant
tion. This can be extended to include calculation of C f x and C f y as feedrate cut so comparisons can be made between variable and
a function of the tool location ␪. The moment of inertia 共I兲 for the fixed feedrate cuts. The standard practice feedrate that is used in
cutter is: this testing was provided by an industrial company that machines
complex sculptured parts.
␲ r 4e The fourth and sixth pass sine surface files were processed by
I⫽ (12) the selection program using a maximum tool deflection of 0.25
4 mm. This constraint results in a cutting force of 133 N if the force
where r e is an effective radius which accounts for the area reduc- is assumed to act at the end of the tool, obviously a worst case
tion because of the flutes. A suggested value for r e of 0.8r is condition. The other process constraints associated with feedrate
given by Kops and Vo 关28兴. selection were not considered for this cut so the models could be
For most tool moves, the surface perpendicular to the move will evaluated. The conditions of the machining, along with the re-
be of most concern in terms of tolerance, leading to a conservative quired model constants, are shown in Table 1.
constraint on the F y cutting force. However, there are instances in Eight passes are used to reduce the block to the surface shown
contour milling where the F x force must be limited. An example in Fig. 6共a兲. The first five passes follow a sine curve in the Y and
would be cutting down the backside of a convex surface. If a large Z directions with a step over of 1.2r, resulting in continuously
tool deflection occurred in the opposite direction of travel, the varying axial and radial depths of cut. In the sixth pass, the sine
material behind the cutter could be gouged 关29兴. frequency is changed such that the cutter enters and exits the
As seen in Fig. 5, an iterative loop is used to find the feedrate workpiece several times by intersecting previous passes. The last
resulting in the desired cutting force. A bisection method is used two passes have significant changes in the depths of cut, but the
to adjust the feedrate during each interaction, providing speed and cutter does not exit the workpiece. All cuts were limited to a
robustness. Additional algorithm efficiency is obtained by reduc- maximum feedrate of 16.9 mm/sec. Force measurements were
ing the mechanistic model equations to constant terms and terms made for each tool pass with a Kistler dynamometer and a PC data
that vary with tool angular position. The constant terms only need acquisition board using 1000 Hz sampling. A tool runout of ap-
to be calculated once for each tool move as described below. proximately 0.00635 mm was measured with a dial indicator. Cuts
The average chip thickness for a given tool move can be deter- were made using a Boston Digital 505 five-axis CNC machine.
mined from the contact angles of each axial disc. Equation 共2兲 is Figure 6共b兲 shows the resultant forces during the fourth pass of
integrated over the contact angles of a disc element, and divided the cut. The axial depth of cut for this pass varies from 0.6 mm,
by the total angle of contact, to obtain the average chip thickness where cutting takes place only on the ball, to 2.98 mm, where
of a disc. Since the cutter is sliced into axial discs, the overall chip cutting takes place on the ball and cylindrical portion of the cutter.
average is obtained from summing all the individual disc aver- The feedrate program generated values ranging from 8.3 to 11.1
ages, and dividing by the number of discs, as shown by the fol- mm/sec for this pass, and this resulted in a maximum cutting force
lowing equation: very close to the desired value of 133 N.
The experimental results for the sixth pass are shown in Fig.

h avg⫽
f
⍀n t n z
nz

兺
i⫽1
冋 共 cos ␾ ext i ⫺cos ␾ ent i 兲 cos ␪ 2 i cos ␤
␾ ent i ⫺ ␾ ext i
6共c兲. The measured forces show that the mechanistic model does a
conservative and adequate prediction of the cutting forces except
when the cutter is undergoing significant transient conditions, e.g.

册
entering or leaving the workpiece. The peak forces are up to 50
percent above the desired value for these instances. This model
⫺sin ␪ 2 i sin ␤
error can be tolerated if a conservative approach is used in select-
ing the desired force. An iterative deflection model is needed for
h avg⫽ f K (13) more accurate prediction of these transient forces.
The results for the eighth pass are shown in Fig. 6共d兲 for vari-
The integration of Eq. 共2兲 is simplified because ␪ 2 and ␤ are able feedrates, and in Fig. 6共f 兲 for a constant feedrate of 6.35
constant for each disc. This equation can be combined with Eqs. mm/sec. The desired force for this pass has been increased to 156
共3兲 through 共6兲 to yield force expressions that are a product of N. Once again the feedrate program does a good job of maintain-
constants times the summation of sine and cosine of the flute ing the force over varying depths of cut. Peak force errors of 30
position angles of each disc element, and the feedrate magnitude f. percent do occur during several transients. Figure 6共e兲 shows the
The calculation described above is for one particular tool loca- feedrate generated by the selection program that results in a 31
tion in the workpiece and for one angular position of the cutter. second cut. The standard practice cut is shown in 共f 兲, which re-
During a full rotation of the cutter the flute location in the contact sults from a constant feedrate of 6.35 mm/sec. The peak force is
area may vary substantially. To insure that the peak force location around 200 N, the same as for the variable feed cut, but the entire
is determined for all cases, e.g. ball end mill with runout, feedrate pass

218 Õ Vol. 123, MAY 2001 Transactions of the ASME

 Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025
Fig. 6 Experimental cutting force and feed velocity for sine surface

takes 40 seconds. The feedrate program saves 23 percent cutting ing of an oil bottle mold 共Fig. 7兲. The 127⫻203⫻50.8 (mm) alu-
time over standard practice while maintaining desired peak force minum mold was machined in three phases: roughing, semi-finish
levels. and finish cutting. Toolpaths were generated using a commercially
It is evident that the machining time can be reduced, dramati- available CAM system, FeatureCam, developed and marketed by
cally in some cases, and that the cutting force magnitude can be Engineering Geometry Systems of Salt Lake City, Utah.
constrained by automatic selection of the feedrate. This capability FeatureCam includes built-in tables 关30兴 for a wide variety of
is essential for fast feedrate planning and cutting while still main- materials and cutting tools, and will automatically calculate the
taining acceptable quality. correct spindle speed and feedrate. The table selected cutting con-
ditions for the mold, along with the allowable cutting force are
5.2 Test Case. In order to determine the probable benefits shown in Table 2.
of optimization, we tested them on the sculptured surface machin- Cutting times, maximum and average forces for the three ma-
chining phases are shown in Table 3. The table selected feedrates
used were those shown in Table 1. Overall, the cutting time for
Table 1 Cutting conditions for ball end-milling of sine surface the 31,202 line program was reduced from 89 minutes to 77 min-
utes, a 13 percent reduction. Our algorithms took a little more

Fig. 7 Bottle mold during semi-finish cutting with a 12.7 mm

ball-end mill, and after the finish pass

Journal of Manufacturing Science and Engineering MAY 2001, Vol. 123 Õ 219
than eleven minutes on a Pentium II 266 MHz to calculate the
optimized feedrates, a rate of around 46 lines/second.
The table selected feedrates used in rough cutting yield cutting
forces that are fairly constant. On the other hand, during the semi-
finish cutting the steps left by the rough cutting present large
differences in chip load and therefore wide variations in the cut-
ting force. Maximum cutting forces are reduced by 53 percent
while still decreasing the overall cutting time by 16 percent. Op-
timized finish cutting yielded a very modest improvement of 13
percent in cutting time, but a significant decrease of 70 percent in
maximum forces. This improved control of cutting force will cer-
tainly result in closer tolerances and a superior surface finish.
The results shown in Table 3 indicate that optimized feedrates
can offer significant improvements in both cutting time and cut-

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

ting forces. Experimental measurement of actual cutting forces
was accomplished using a Kistler load cell at a sampling rate of
1000 Hz. Good correlation was found between the model pre-
dicted 共simulation兲 and experimentally measured forces for the
sixth level roughing pass, as seen in Fig. 8. In Fig. 8共a兲 the ex-
perimental peak resultant forces are slightly less than the pre-
dicted, i.e., desired, forces. Factors such as cutting fluid, tool
runout and process dynamics cause some inaccuracy in the model.
Figure 8共b兲 is a comparison of the forces for the finish cutting of
the bottom of the mold. The experimental peak forces are slightly
higher than the desired and much higher during two transients.
The accuracy of the process simulation was somewhat diminished
during large process transients, but overall performed reasonably
well.
The effect of changing the constant feedrate to a value which
results in acceptable force levels is illustrated in Table 4. We
choose constant feedrates low enough to greatly decrease the
maximum cutting forces. For rough cutting the cutting conditions Fig. 8 Comparison of experimental and simulation „desired…
peak resultant cutting force for roughing and finishing of the oil
are fairly constant and there is no significant advantage to a vari-
bottle mold
able feedrate program; therefore Table 4 does not include rough
cutting.
Table 4 shows how low the feedrates must be set for the semi- 6 Conclusions
finish and finish cases in order to decrease the peak cutting forces
to the desired levels. The resulting cutting times are significantly A geometric model has been combined with an inverse mecha-
longer and the difference between constant feedrate and optimized nistic model to automatically select feedrates for three-axis end
cutting times is dramatic. The constant feedrate total cutting time mill machining. The feedrate selection is based on a desired cut-
共including rough cutting兲 is 292 minutes versus 77 minutes for the ting force that is calculated from an allowable tool deflection.
optimized case, a difference of 74 percent with peak forces that Multiple pass NC part files can be run through the selection pro-
are still lower than the constant feedrate case. gram so that optimum feedrates can be placed in the file for each
tool move.
The program was tested experimentally using a three-axis CNC
machining center to cut a contour sine surface in a 6061-T6 alu-
Table 2 Table selected cutting conditions minum block. Results showed that the desired peak resultant cut-
ting force was maintained except under large transient conditions.
Machining time was shown to be reduced by 23 percent when
compared to the times for standard practice cuts. Simulation com-
parisons showed significant improvement in machining time and
peak forces as compared to standard practice for an oil bottle
mold that requires 31,202 tool moves. Gcodes processing was at a
rate of 46 lines/sec using a 266 MHz PC.
Important future work includes extension of the algorithms to
five-axis cutting, inclusion of process dynamics, and prediction
and measurement of surface error using static and dynamic force
Note that the allowable force was determined using 50% of the tolerance. models.

Table 3 Comparison of model predicted cutting time, maximum and averages

forces for traditional table selected and optimized feedrates

1兲
simulations run with an assumed tool runout of 0.00254 mm
2兲
z-buffer spacing is 0.254 mm
3兲
workpiece material is 6061 T6 aluminum

220 Õ Vol. 123, MAY 2001 Transactions of the ASME

Table 4 Comparison of model predicted cutting time, maxi- Appendix A
mum and averages forces for constant feedrates and optimized
feedrates for semi-finish and finish machining 1. Define tool local coordinate system
Initial and final tool positions, Pi and Pf are defined in terms of the
XYZ World Coordinate System 共WCS兲 with unit vectors i,j,k. 共see
Fig. 9兲 The tool xyz Local Coordinate System 共LCS兲 has respec-
tive unit vectors n,o,a. defined by:
a⫽k⫽ 共 0,0,1 兲 (A1)
0⫽ 共 a⫻D兲 / 兩 a⫻D兩 (A2)
Acknowledgments n⫽ 共 o⫻a兲 / 兩 o⫻a兩 (A3)
This research is supported by the National Science Foundation where D is the vector defining the tool motion direction
under contracts DMI-9622612 and DMI-9872575. Their support

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

is gratefully acknowledged. D⫽ 共 Pf⫺Pi兲 ⫽ 共 Dxi⫹Dyj⫹Dzk兲 兩 D兩 (A4)
the tool movement vector can be expressed in terms of the LCS
Nomenclature
d⫽ 共 dxn⫹dza兲 兩 D兩 (A5)
a ⫽ axial depth of cut
ap ⫽ ZDV point spacing dx⫽D.n/ 兩 D兩 (A6)
b ⫽ radial depth of cut dz⫽DC.a/ 兩 D兩 (A7)
Cfx ⫽ center location of the x cutting force acting on the
cutter note that dy⫽0 since by definition d lies on the x-z plane of the
Cfy ⫽ center location of the y cutting force acting on the LCS.
cutter 2. Find the intersection of the swept envelope (SWE) of the tool
D ⫽ unit vector in tool feed direction with the ZDVs
E ⫽ modulus of elasticity of the end mill The Z-buffer is used to determine the contact area between the
f ⫽ feedrate cutting tool and the workpiece. Each ZDV is tested to see if it
Fx ⫽ x component of force acting on cutting tool intersects the SWE. Intersection calculations for ball-end, flat-end
Fy ⫽ y component of force acting on cutting tool and round-end tools are derived in Jerard, Drysdale and Hauck
⌬F t ⫽ tangential force per unit length 关31兴 and Choi and Jerard 关28兴. Five axis intersection calculations
⌬F r ⫽ radial force per unit length are derived in Jerard et al. 关32兴.
h ⫽ unperformed chip thickness As shown in Fig. 10, as the tool moves along its path it will
h avg ⫽ unperformed average chip thickness intersect the ZDV removing the segment between p1 and p2. The
I ⫽ moment of inertia of the end mill ZDV is shown as a volume to indicate the fact that each ZDV
K ⫽ cutting constant for each tool move that relates aver- represents a discrete area determined by the spacing between the
age chip thickness to feed per tooth ZDVs. The length of p1⫺p2 multiplied times the area associated
Kt ⫽ tangential specific cutting force coefficient with the ZDV is the volume removed by the SWE. The summa-
K tc ,p 1c⫽ tangential force constants for cylindrical section of tion of the volumes of all the intersected ZDVs is the total volume
cutter of material removed by this SWE.
K ts ,p 1s ⫽ tangential force constants for spherical section of cut- The point p1 is the lowest intersection between the ZDV and
ter the SWE. Point P0 is the base point on the X-Y plane (X 1 ,Y 1 ,0).
Kr ⫽ radial specific cutting force coefficient Point p2 is the current position of the end of the ZDV with z 2
K rc ,p 2c⫽ tangential force constants for cylindrical section of corresponding to the z value of the surface of the workpiece at that
cutter x-y position.
K rs ,p 2s⫽ radial force constants for spherical section of cutter The x-y coordinates of p1, expressed in terms of the LCS are:
L ⫽ free length of the cutter
x 1 ⫽ 共 P0⫺Pi兲 .n (A8)
nt ⫽ number of cutter flutes
nz ⫽ number of cutter axial slices y 1 ⫽ 共 P0⫺Pi兲 .o (A9)
Ns ⫽ surface normal vector
dS ⫽ tooth segment length
r ⫽ radius of axial slice of tool
re ⫽ effective tool radius for moment of inertia calculation
y max ⫽ maximum radial depth of cut, located from entrance
position of cutter to x tool coordinate
y min ⫽ minimum radial depth of cut, located from exit posi-
tion of cutter to x tool coordinate
xyz ⫽ local coordinate system
XY Z ⫽ world coordinate system
␣ cal ⫽ calculated intersection locating angle
␣ real ⫽ actual intersection locating angle
␤ ⫽ angle between feed direction D and x axis
␦x ⫽ x deflection of the cutting tool
␦y ⫽ y deflection of the cutting tool
⍀ ⫽ spindle speed
␾ ⫽ angle location of tooth segment
␾ ent ⫽ entrance angle location of cutter
␾ ext ⫽ exit angle location of cutter
␪2 ⫽ angle location on the ball of the end mill
⌿ ⫽ helix angle of tool cylindrical section Fig. 9 World and local coordinate system definitions

Journal of Manufacturing Science and Engineering MAY 2001, Vol. 123 Õ 221
 z 2 ⫽ 共 P2⫺Pi兲 .a (A15)
共note that x 2 ⫽x 1 and y 2 ⫽y 1 ).
If any of the following conditions are true then quit the calcu-
lation and process the next ZDV of interest:
兩 y 1 兩 ⬎r (A16)
u 1 ⬎ 兩 D 兩 or u 1 ⬍0 (A17)
z 2 ⬍z 1 (A18)
Equations 共A16兲 and 共A17兲 are true if the ZDV does not fall under
the shadow of the SWE 共the initial and final stationary tool posi-
tions must also be checked兲. If 共A18兲 is true then there is no
intersection between the ZDV and the SWE, i.e. the material has

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

already been cut lower than this SWE cuts it.
In order to be conservative it may be desirable to increase the
radius of the cutter 共r兲 by an amount equal to the spacing between
ZDVs. For finish machining, where cutting depths are very small,
Fig. 10 The ZDV is intersected by the swept envelope of the this will prevent the simulation from estimating a zero volume
tool at P1 removal when in fact there is contact between the tool and the
workpiece. Equation 共A16兲 should use this expanded radius. Equa-
tion 共A14兲 should be calculated twice; once with the actual radius
As indicated in Fig. 10, the tool moves in the direction d. For the and once with the expanded radius. The expanded radius will be
spherical portion of the cutter the SWE is generated by a circle used in all of the contact area calculations, but the actual radius
perpendicular to the direction vector d 共see Fig. 11兲. At distance will be used to calculate the clipped value of the ZDV for use in
u 2 from the start position the tool comes in contact with point p2. subsequent tool moves.
As the tool moves from u 2 to u 1 the ZDV is intersected, thereby 3. Project p1 and p2 in the ⫺d direction to intersect the cutting
removing the segment p2⫺p1. The projection of p2⫺p1 onto the tool at its initial position Pi
tool is the area of contact and is shown as the segment p3⫺p4 in The contact area is determined by projecting the segment p2
Figs. 10 and 11. ⫺p1 in the ⫺d direction onto the surface of the cutting tool. This
Equations 共A10兲–共A13兲 derive an expression for z 1 using the is equivalent to moving the tool in the ⫹d direction. 共See Fig. 12兲
geometry defined in Fig. 11: The point p3 is found by using the value of u 1 obtained in Eq.
共A12兲
x 1 ⫽u 1 dx⫹h 1 dz (A10)
where x 3 ⫽x 1 ⫺u 1 dx (A19)

h 1 ⫽ 共 r 2 ⫺y 21 兲 0.5 共 see the right side of Fig. 11兲 (A11) y 3 ⫽y 1 ⫽y 2 共 all in the same plane兲 (A20)

therefore the distance traveled by the tool in the direction d when z 3 ⫽z 1 ⫺u 1 dz (A21)
the tool intersects the ZDV at the lowest point is: The point p4 is found by solving for the intersection of a line with
u 1 ⫽ 共 x 1 ⫺h 1 dz 兲 /dx where 0⬍u 1 ⬍⫽ 兩 D兩 (A12) either a cylinder or a sphere. Start by finding the intersection of
the line through p2 with a sphere centered at the LCS origin.
Figure 11 also shows that
x 24 ⫹y 24 ⫹z 24 ⫽r 2 共 equation of a sphere兲 (A22)
z 1 ⫽u 1 dz⫺h 1 dx (A13)
Equation 共A13兲 can be further simplified by eliminating u 1 . p4⫽p2⫺u 2 d (A23)
(u 2 is distance the tool moves when the actual intersection takes
z 1 ⫽ 共 x 1 dz⫺h 1 兲 /dx (A14)
place兲
The value of z 2 in terms of the LCS is

Fig. 12 Projection of p1 and p2 onto the tool determines the

Fig. 11 Swept envelope of the tool clips ZDV at p1, and remov- location of p3 and p4. Segment p3-p4 is contact area between
ing segment p1Àp2 the tool and the workpiece material represented by this ZDV.

222 Õ Vol. 123, MAY 2001 Transactions of the ASME

 if x 3 and x 4 are both negative 共Fig. 13b兲 then the intersection is all
on the backside and therefore:
z upper⫽z 3 and z lower⫽z 4 (A35)
if x 3 is negative and x 4 is positive then the intersection is on both
the back and the front 共Fig. 13c兲 and therefore:
z upper⫽max共 z 3 ,z 4 兲 and z lower⫽⫺r 0 (A36)
where r 0 is the radius of the circle defined by the intersection of a
plane parallel to x-z plane at a distance y 1 . This is:
r 0 ⫽ 共 x 23 ⫹z 23 兲 0.5⫽h 1 (A37)
共the equality with h1 is obvious from observing Fig. 11兲

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

It is necessary to find entrance and exit angles for all the discs
that are intersected. For the cylindrical portion of the tool the
cutting can only take place on the front of the tool 共i.e. x 4 ⫽x 3
⬎0). For each disk maintain a y max and y min by comparing them
to the y 1 value obtained for each ZDV. The angular limits of
contact are then easily obtained. For the spherical portion of the
tool, things get a bit more complicated since cutting can take place
on both the front and back of the tool. The procedure requires
keeping a separate y max and y min for all four quadrants of each
disc as explained in Section 3.2.

Fig. 13 Determination of upper and lower z values for the con-

tact area of the tool References
关1兴 Durdag, K., 1994, ‘‘Computer-Generation of Feedrates and In Process Tuning
for Three Axis End Milling,’’ Masters Thesis, University of New Hampshire.
关2兴 Sungurtekin, U. A., and Voelcker, H. B., 1986, ‘‘Graphical Simulation and
x 4 ⫽x 2 ⫺u 2 dx (A24) Automatic Verification of NC Machining Programs,’’ Proc. IEEE Int’l Conf.
on Robotics and Automation, San Francisco, 1, pp. 156–165, Apr. 7–10.
关3兴 Vann, C. S., and Cutkosky, M. R., 1988, ‘‘Closing the Loop in CAD/CAM
y 4 ⫽y 2 (A25) Integration,’’ Proceedings of the USA - Japan Symposium on Flexible Auto-
mation, Minneapolis, Minn., II, pp. 915–922.
z 4 ⫽z 2 ⫺u 2 dz (A26) 关4兴 Wang, K. K., 1988, ‘‘Solid Modeling for Optimizing Metal Removal of Three-
substituting 共A24兲, 共A25兲, and 共A26兲 into 共A22兲 yields: Dimensional NC End Milling,’’ J. Manuf. Syst., 7, No. 1, pp. 57–65.
关5兴 Takata, S., Tsai, M. D., Inui, M., and Sata, T., 1989, ‘‘A Cutting Simulation
au 22 ⫹bu 2 ⫹c⫽0 (A27) System for Machinability Evaluation Using a Workpiece Model,’’ CIRP Ann.,
38Õ1, pp. 417–420.
关6兴 Kline, W. A., DeVor, R. E., and Lindberg, J. R., 1982, ‘‘The Prediction of
where a⫽(dx ⫹dz )⫽1, b⫽⫺2(x 2 dx⫹z 2 dz),
2 2
c⫽x 22 ⫹y 22 ⫹z 22 Cutting Forces in End Milling with Application to Cornering Cuts,’’ Int. J.
⫺r 2 Mach. Tools Manuf., 22, No. 1, pp. 7–22.
关7兴 Takata, S., 1993, ‘‘Generation of a Machining Scenario and Its Applications to
u 2 ⫽ 共 ⫺b⫺ 共 b 2 ⫺4ac 兲 0.5/2a (A28) Intelligent Machining Operations,’’ CIRP Ann., 42, No. 1, pp. 531–534.
关8兴 Lim, E. M., and Menq, C. H., 1997, ‘‘Integrated Planning for Precision Ma-
substitute u 2 into 共A24兲 and 共A26兲 to obtain x 4 and z 4 . If z 4 ⬍0 chining of Complex Surfaces. Part 1: Cutting-Path and Feedrate Optimiza-
then the line intersects the ball-end portion of the cutter; otherwise tion,’’ Int. J. Mach. Tools Manuf., 37, No. 1, pp. 61–75.
it must intersect the cylindrical portion of the tool. In this case use 关9兴 Lazoglu, I., and Liang, S. Y., 1996, ‘‘Feedrate Optimization on Complex
the equation for a cylinder instead of Eq. 共A22兲. Workpieces for CNC Milling Machines,’’ ASME 1996 IMECE Symposium on
the Physics of Machining Process-III, Atlanta GA, pp. 129–138.
关10兴 Altan, T., Shatla, M., Yen, Y., and Akgerman, N., 1998, ‘‘High Performance
x 24 ⫹y 24 ⫽r 2 (A29) Machining for Die/Mold Manufacturing-R&D in Progress,’’ Machining Im-
共equation of a cylinder at LCS origin with long axis in the z possible Shapes, IFIP International Conference on Sculptured Machining, Nov
11, 1998, Chrysler Technology Center, MI, Kluwer Academic Publishers, pp.
direction兲 293–314.
substituting 共A24兲 and 共A25兲 into 共A29兲 关11兴 Stori, J. A., Wright, P. K., and King, C., 1999, ‘‘Integration of Process Simu-
lation in Machining Parameter Optimization,’’ ASME J. Manuf. Sci. Eng.,
au 22 ⫹bu 2 ⫹c⫽0 (A30) 121, February pp. 134–143.
关12兴 Spence, A. D., and Altintas, Y., 1994, ‘‘A Solid Modeler Based Milling Pro-
2
where a⫽dx , b⫽⫺2x 2 dx, c⫽x 22 ⫹y 22 ⫺r 2which can again be cess Simulation and Planning System,’’ ASME J. Eng. Ind., 116, pp. 61–69.
solved using the quadratic formula Eq. 共A28兲. Simple expressions 关13兴 Bailey, T. E., Jenkins, D. M., Spence, A. D., and Elbestawi, M. A., 1996,
‘‘Integrated Modeling for Metal Removal Operations,’’ 1996 ASME IMECE,
for u 2 , x 4 , and z 4 can be derived after a little algebra: Atlanta, Georgia, Proceedings of the ASME Dynamic Systems and Control
Division, 58, pp. 191–198.
x 4 ⫽h 1 (A31) 关14兴 El Mounayri, H., Spence, A. D., and Elbestawi, M. A., 1998, ‘‘Milling Process
Simulation- A Generic Solid Modeller Based Paradigm,’’ ASME J. Manuf.
u 2 ⫽ 共 x 2 ⫺x 4 兲 /dx (A32) Sci. Eng., May, 120, pp. 213–221.
关15兴 Fussell, B. K., Jerard, R. B., and Durdag, O. K., 1995, ‘‘Geometric and
z 4 ⫽z 2 ⫺u 2 dz (A33) Mechanistic Model Integration for 3-Axis CNC Feedrate Generation,’’ ASME
1995 IMECE Joint-Symposium on Concurrent Product and Process Engineer-
4. Determine which axial discs are intersected and store informa- ing, San Francisco, CA, 12 Nov 1995, MED-Vol. 1, pp. 37–51.
tion for entrance/exit angle calculation 关16兴 Kline, W. A., and DeVor, R. E., 1983, ‘‘The Effect of Runout on Cutting
Define z upper and z lower to be the upper and lower values of the Geometry and Forces in End Milling,’’ Int. J. Mach. Tools Manuf., 23, pp.
intersection of the ZDV with the tool. If x 3 and x 4 are both posi- 123–140.
关17兴 Fussell, B. K., Ersoy, C., and Jerard, R. B., 1992, ‘‘Computer Generated CNC
tive then the intersection is entirely on the front of the tool 共Fig. Machining Feedrates,’’ Proceedings of the Japan-USA Symposium on Flexible
13a兲, in which case: Automation, San Francisco, CA, pp. 377–384.
关18兴 Yang, M., and Park, H., 1991, ‘‘The Prediction of Cutting Force in Ball-End
z upper⫽z 4 and z lower⫽z 3 (A34) Milling,’’ Int. J. Mach. Tool Des. Res., 31, No. 1, pp. 45–54.

Journal of Manufacturing Science and Engineering MAY 2001, Vol. 123 Õ 223
关19兴 Yucesan, G., and Altintas, Y., 1996, ‘‘Prediction of Ball End Milling Forces,’’ 关26兴 Gagne, D. P., 1990, ‘‘A Method for the Intelligent Control of End Milling,’’
ASME J. Eng. Ind., 118, pp. 95–103. Masters Thesis, University of New Hampshire.
关20兴 Choi, B. K., and Jerard, R. B., 1998, Sculptured Surface Machining: Theory 关27兴 Kolarits, F. M., and DeVries, W. R., 1991, ‘‘A Mechanistic Dynamic Model of
and Application, Kluwer Academic Publishers. End Milling for Process Controller Simulation,’’ ASME J. Eng. Ind., 113,
关21兴 Voelcker, H. B., and W. A. Hunt, 1981, ‘‘The Role of Solid Modeling in pp. 176–183.
Machining - Process Modeling and NC Verification,’’ SAE Technical Paper. 关28兴 Kops, L., and Vo, D. T., 1990, ‘‘Determination of the Equivalent Diameter of
810195 the End Mill,’’ CIRP Ann., 39Õ1, pp. 93–96.
关22兴 Fridshal, R., Cheng K. P., Duncan D., and Zucker, W., 1982, ‘‘Numerical 关29兴 Yang, M. Y., and Sim, C. G., 1993, ‘‘Reduction of Machining Errors by
Control Part Program Verification System,’’ Proc. Conf. CAD/CAM Technol- Adjustment of Feedrate in the Ball-End Milling Process,’’ Int. J. Prod. Res.,
ogy in Mechanical Engineering, March, MIT Press, pp. 236–254. 31, No. 3, pp. 665–689.
关23兴 Wang, K. K., 1987, ‘‘Application of Solid Modeling to Automate Machining 关30兴 Metcut, 1986, Machining Data Handbook, Machinability Data Center, Metcut
Parameters for Complex Parts,’’ Proceedings of CIRP Manufacturing Semi- Research Associates, Third Edition, Cincinnati.
nar, Penn State, pp. 33–37. 关31兴 Jerard, R. B., Drysdale, R. L., and Hauck, K., 1988, ‘‘Geometric Simulation of
关24兴 Van Hook, T., 1986, ‘‘Real-Time shaded NC Milling Display,’’ Comput. Numerical Control Machining,’’ Proc. ASME Int’l Computers in Engineering
Graph., 共Proc. SIGGRAPH兲 20, No. 4, pp. 15–20. Conf., July 31-Aug. 3, San Francisco, 2, pp. 129–136.
关25兴 Drysdale, R. L., Jerard, R. B., Schaudt, B., and Hauck, K., 1989, ‘‘Discrete 关32兴 Jerard, R. B., Hussaini, S. Z., Drysdale, R. L., and Schaudt, B., 1989b, ‘‘Ap-
Simulation of NC Machining,’’ Algorithmica, Special Issue on Computational proximate Methods for Simulation and Verification of Numerically Controlled

Downloaded from http://asmedigitalcollection.asme.org/manufacturingscience/article-pdf/123/2/214/5578960/214_1.pdf by Politecnico Di Milano user on 21 January 2025

Geometry 4共1兲, pp. 33–60. Machining Programs,’’ Visual Comput., 5, No. 6.

224 Õ Vol. 123, MAY 2001 Transactions of the ASME