THE ORTHOGONAL DECOMPOSITION METHOD IN THE POST-PROCESSING OF

REDUNDANT MACHINING OPERATIONS

Hojjat Valipour and Luc Baron

Department of mechanical engineering École Polytechnique de Montréal

P.O. box 6079, succ. CV, Montréal, Qc, H3C 3A7
Email: hojjat.valipour@..., luc.baron@polymtl.ca

Abstract: This paper presents an orthogonal decomposition procedure to resolve the

redundancy by using a direct calculation of the generalized inverse of the Jacobian matrix
considering the joint limits of the machine tool. An optimization procedure has been
implemented in a post-processor module and tested with the architecture of the milling
center Huron KX8-Five. Different shapes of free form surfaces have been successfully
machined with the module without exceeding joint limits. The tasks have been validated
with VERICUT (a commercial simulator software) before machining them with the
milling center.

Keywords: Post-processor, Inverse kinematics, Five-axis milling, Orthogonal

decomposition, Redundancy solution.

1. INTRODUCTION Iterative numerical methods involve a large number of

The sculptured surfaces (free-form surfaces) have found
extensive industrial applications, such as in automobile
bodies, ship hulls and aerospace parts. Five-axes machines
performs efficient and accurate machining for complicated
sculptured surfaces in order to achieve optimum cutting
conditions.
There are many problems that can occur during the
machining process, such as worse quality of machined
surface, redundant machining and interference between the
tool and surrounding objects. Hence, there is a high demand
for the computer aided manufacturing (CAM) software to
achieve better surface quality, better efficiency and no
interference between the tool and the part surfaces in
machining, while avoiding the joint limits.
In general, CAM systems can be divided into three stages.
The first is to generate cutter contact points (CC points), the
second stage is to generate cutter location data (CL-data) and
the last stage is to convert CL-data to machine code. The
interface that links the CAM systems and NC machines is
known as post-processor [1]. The tool positions and
orientations relative to the part frame require further
transformations by the post-processor before encoding it into
G-codes file. The transformation requires knowledge of the
architecture of machine tool in order to solve the inverse
kinematics, i.e., transferring the tool positions and
orientations (operation space) into machine axes positions
(joint space).
The inverse kinematics for five-axis machines can be found
either as a closed-form solution (analytical solution) or as an
iterative numerical solution. In general, analytical methods
are faster than numerical ones. It is also easier to apply the
rules for choosing one among several possible inverse
solutions, but finding an optimal one is rather complicated
for redundant case.
numerical operations [2], although they are powerful in finding
the optimal solution in redundant cases for machine tools as it
has been demonstrated in finding the optimal solution for
robots.
The redundancy of machining operations may arise in many
different situations, where the dimension of the task space
(included in operation space) is smaller than the dimension of
the joint space. It means that the degree of freedoms (DOF)s
required by the tool to perform the task is less than the number
of axes of the machine, e.g., a surfacing task with a ball-end
mill can be considered as a redundant operation. This task
requires only 3-DOFs, and hence, can be realized with only a
3-axis machine-tool. However, if the task is realized with a 5-
axis machine-tool, there are 2-DOFs of redundancy, and hence,
additional specifications are allowable and desirable to avoid
joint limits [3], for example.
2. KINEMATIC MODEL
In machine-tools, the direct kinematics (DK) problem is to find
the position and orientation (pose) of the tool for given
machine joint positions. The DK solution can be obtained
through several mathematical tools. In this paper, the Denavit-
Hartenberg (DH) convention [4] is chosen as the DK of the
milling center Huron KX8-Five.
The positive directions of joint motions are shown in fig. 1.
Coordinate frames P and T are attached to the top of the rotary
table and centred to the sphere of the ball-end mill tool,
respectively. The kinematic chain of the Huron is illustrated in
fig. 2. Finally, associating the coordinate frames from the
workpiece to the tool, according the DH parameters is shown
in fig. 3, while the DH parameters are summarized in table 1.
The homogeneous transformation matrices of each coordinate
frame (from P to T) can be computed from table 1. Multiplying
 and N = [i j k] (2), where,

i = -1/2 Cc- 2 /2 Sc SA+1/2 Cc CA,

j= -1/2 Sc+ 2 /2 Cc SA+1/2 Sc CA ,
k= 1/2 CA+1/2,

where, SA, CA, Sc and Cc denote the sin (A), cos (A), sin (C)
and cos (C), respectively. Moreover, l is the tool length, i.e.,
the distance between the gauge plane and the ball center, while
d is defined as the distance between the surface of table and the
rotating axis-A. Moreover, the joint limits of the machine are
as follows:
-315 mm ≤ X ≤ 335 mm
-350 mm ≤ Y ≤ 350 mm
0 mm ≤ Z ≤ 450 mm (9)
Fig. 1. Position joint motion of the Huron KX8-Five. -180 D ≤ A ≤ 45 D
-99 999.999 D ≤ C ≤ 99 999.999 D
d = 75 mm

3. OPTIMIZATION PROBLEM

In the field of robotics, the subject of redundancy-resolution

Fig. 2. Kinematic chain of the Huron Kx8-Five
has been studied at both joint-position level [5] and at joint-rate
level [6]. Here, we adopt the joint-rate formulation because the
problem of redundancy-resolution is linear at that level. In the
case of surfacing operation, the task requires only the three
DOFs of position, and hence, the Jacobian of the machine-tool,
namely J, maps the joint-rates θ into the tool velocity p , as
follows:
∂p
J θ = p , J ≡ (10)
∂θ
where the solution is well-know to be

Fig. 3. DH coordinate frames of the Huron KX8-Five θ = J + p + (1 − J + J )h (11)

(from workpiece to tool through joint C A Y X Z)
with h being a secondary task and J + the right generalized
θi αi inverse of J, i.e., J + = J T ( JJ T ) −1 . Moreover, (1 − J + J ) is an
i ai bi i+1
orthogonal complement of J projecting h onto the nullspace of
P (C)-90 0 -d 45 1
1 (A)+90 0 0 90 2 J. The first part of eq.(11) presents minimum norm solution
2 45 0 (Y) -90 3 and the second part the eq.(11) presents a homogenous
3 -90 0 (X) -90 4
4 0 0 (Z)+d 0 5 solution. The first part of eq.(11) does not control the tool
5 90 0 -l 0 T orientation. h in the second part must be consisted, at least a
frame degree mm mm degree frame term of cutting tool orientation, i.e., the normal of the milling
Table 1. DH parameters of the Huron KX8-Five surface Of. With this term, it is possible to optimize tool
orientation O, in order to approach it toward Of with some
The first three rows of the last two columns of the DK model acceptable deviation. In fact, the axis of the tool in any point of
readily provides the position, namely P and the orientation, the tool path would be within a cone of axis with the angle of
namely N, vectors of the tool with respect to the part. desired maximum error. The first part of eq.(11) cannot control
P = [x y z] (1), where, the joint limits of the milling machine. Therefore, approach of
θ to the mid-joint position θ can be imposed. So the
x=Y Sc CA-1/2 Cc (Z+d- l )-1/2 X Cc+ 2 /2 X Sc SA secondary task h is chosen as −∇ z, where z is the following
− 2 /2 Sc SA (Z+d- l )- 1/2 X Cc CA+ 2 /2 Y Cc SA, objective function:

y= 2 /2 Cc SA (Z+d- l )-1/2 X Sc+1/2 Sc CA (Z+d- l) 1

z= (θ − θ )W T W (θ − θ ) + W0 (1 − O T O f ) (12)
-1/2 Sc (Z+d- l )+ 2 /2 Y Sc SA- 2 /2 X Cc SA-Y Cc CA, 2
z= -1/2 X CA+1/2 X+ 2 /2 Y SA+1/2 CA (Z+d- l)
+1/2(Z+d- l ) where W is a diagonal positive definite weighting matrix and
θ is the mid-joint position, while, Of and W0 are, respectively,
the unit vector along the desired final tool orientation and the
positive weighting notation. The objective is to minimize z
so h could be −∇z , hence,
d orientation of the tool frame relative to the part frame. The
h = −W (θ − θ ) + W0 (O T O f ) (13)
dθ
4. LINEARIZATION
When the CL-data are generated by a CAM system, it is
assumed that the tool path between two CL points is a
straight line relative to the workpiece. However, due to the
rotary axes of the machine, the tool path between two blocks
in the NC program will not be linear relative to the
workpiece, reducing the accuracy of the tool path.
Linearization of the tool path in the post-processing module
can solve this problem. In an unlinearized post-processor
module each motion is direct transformed into one command
in the CL-data file, i.e., a G01 command. The number of
lines in the NC program is approximately equal to the
number of lines in the CL-file. Linearization of the tool path
is performed in the post-processor by interpolating new CL-
data point along the ideal tool path, thereby adding new
blocks to the NC program. One command in the CL-file may
result in several lines in the NC file, so the size of the NC
program increases. The final tool path consists of positions
that originate directly from CL-data file, and new positions
that are generated by CL-data interpolation by the post-
processor module.
Linearization does not provide a perfect tool motion, but the
deviation from the ideal tool path is reduced to an acceptable
level. Linearization algorithms must use calculations of both
the forward and the inverse kinematics to calculate
deviations from ideal tool path and to interpolate new CL-
data. A recursive method for linearization, taking into
consideration both the position and the orientation deviation
of the tool, is described by K. Sorby [2].
6. OPTIMAL POST PROCESSOR MODULE
An iterative numerical solution method is used to solve the
inverse kinematics in a similar manner as the resolve motion
rate method introduced by Whitney [7]. The solution
presented in eq.(11) and the secondary task h from eq.(13)
are implemented as an optimal post-processing module
which can be summarized as the following table:
Table 2. The optimal post processing module
Vectors Pf and Of denote the desired final position and
function DKM calculates the actual pose of the tool frame
relative to the part frame. After each “Do” iteration, the next
joint position θ i +1 is computed as the actual joint position θ i
plus a damping factor 0.8 times the joint displacement Δθ , in
order to avoid overshooting of θ i +1 , for ill-condition Jacobian
matrix.
Our algorithm allows a linear interpolation between the actual
and final positions of the center of the ball-end mill, while
leaving the control of the tool orientation to the secondary task.
The latter tries to keep both the joint positions and the tool
orientation at the closest deviation from the preferred tool
orientation. A proper selection of the weight allows a good
balancing between the two contradictory objectives.
7. EXPERIMENTAL TEST
The optimal post processor module is implemented for the
milling center Huron KX8-Five. The CL-data is generated by
CATIA V5, for a free form surface (see fig. 5). The surface is
inscribed in a 4” dia. cylinder. A multi-axis zigzag strategy
with a 0.5” ball-end mill tool is chosen as a finish milling
operation.
Fig. 5. Tool path generated by CATIA V5.
The tool orientation is chosen normal to the surface. After
linearization the APT-file is post-processed for the Huron with
four approaches in order to demonstrate the feasibility of the
operation and to compare the effect of the tool orientation on
the surface quality, i.e.;
1) tool always parallel to vertical; 2) tool always at a constant
orientation different than the vertical; 3) tool always normal to
the surface, and finally, 4) too almost normal to the surface
using the optimal post-processor module in order to avoid
exceeding joint limits.
Joint positions “A” from the third process are shown in fig. 6
(a). According to eq.(9), limits of joint A are
− 180 D ≤ A ≤ 45D , but from fig. 6 (a) the joints A is between
− 82.5D and +82.5 D . Therefore, the joint A exceeds its limits,
and the NC-file generated from this process can not be
accomplished. In order to accomplish this machining task, the
CL-data must be processed by an optimal post-processor
module which can identify the joint limits.
The joint positions A from the optimal post-processing module
are illustrated in fig. 6 (b). This figure, the joint A values are
much lower, i.e., between -78.5 D and 43.5 D . Therefore, the
NC-file generated from this optimal module can be
accomplished by the machine-tool without exceeding the
joint limits.
Fig. 6 (a) Joint A from the third process;
(b) Joint A from the optimal process;
The tool orientation from these two different post-processing
approaches are compared in fig. 7.
Fig. 7 (a) Tool orientation of third process;
(b) Tool orientation of fourth process.
The relative tool orientation between these two processes is
shown in fig. 8.
30
25
Relative tool orientation (deg.)
20
15
10
5
0
0 20 40 60
Fig. 8. Relative tool orientation between processes 3 and 4
6. CONCLUSION
Machining tests has been conducted to demonstrate the
feasibility of the optimal tool path and to compare the
machining quality with different method of post-processing.
To avoid the risk of human or system error, the generated
tool path was validated before actual machining by a
simulation software named VERICUT. Adjusting the feed
rate was implemented to G code-file in order to have an
accurate machining procedure. As it mentioned before, the
task with tool orientation normal to the free form surface
cannot be accomplished with this machine. Hence, the CL-
data of this file is modified and the part was machined up to
A = 43D , i.e., before A-axis reaches its limit.
The machining test result of the four processes from 1 to 4 is
shown in fig. 9.
(a)
(b)
Fig. 9 the four parts machined.
Roughness of surface of the four machined parts is measured
by the Mitutoyo Surftest measurement machine. The results are
(a) summarized in table 3. The results demonstrate that the part 4,
which is machined by process iv) has the least roughness Ra.
Number of part Average of Ra ( μ m)
1 1.54
2 1.45
3 1.5
(b)
4 1.34
Table 3. Average of Ra of four part
REFERENCES
[1] R.S. Lee, C.H. She, Developing a postprocessor for three
types of five-axis machine tools, International Journal of
Advanced manufacture 13 (1997) 658-665.
[2] Knut Sorby, Inverse kinematics of five-axis machines near
singular configurations, International Journal of Machine
Tools and Manufacture may 2006.
[3] Luc Baron, An optimal surfacing post-processor module
for 5-axes cnc milling machines, Third International
Conference on Industrial Automation, Montreal, Canada,
pp.23.17-23.23.
[4] Angeles, J., 2002, Fundamentals of Robotic Mechanical
80 100 120 140 160 180
step
200
Systems: Theory, Methods, and Algorithms, Second
Edition, Springer-Verlag, New York.
[5] Angeles, J., Anderson, K. and Gosselin, C., 1987, An
Orthogonal-Decomposition Algorithm for Constrained
Least-Square Optimization", ASME Robotics,
Mechanisms, and Machine Systems, Design Eng.
Division, Vol. 2, pp. 215-220.
[6] Siciliano, B., 1992, Solving Manipulator Redundancy with
the Augmented Task Space Method Using the Constraint
Jacobian Transpose", IEEE Int. Conference on Robotics
and Automation, Tutorial M1, pp. 5-1-5-8.
[7] Whitney, D.E., 1969, Resolved motion rate control of
manipulators and human prostheses", IEEE Trans. Man-
Machine Syst., Vol. 10, No. 2,pp. 47-53.