Direct Toolpath Generation Based on Graph Theory for Milling Roughing

Abstract

The paper presents a tool path generation methodology for roughing operation based on the oriented graph theory. The cutting areas are identified using an original method that is based on a bicolor and binary map. The toolpath is generated using the searching Dijkstra algorithm inside a graph in order to find the single-source shortest path. The method was employed in order to be applied on ordered and/or unordered point clouds. The entire algorithm was implemented into a mathematic calculus solution which allows the import of point clouds and its processing until the final NC code is generated.

Full text

Procedia CIRP 25 ( 2014 ) 75 – 80
Available online at www.sciencedirect.com
2212-8271 © 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Peer-review under responsibility of The International Scientific Committee of the 8th International Conference on Digital Enterprise Technology - DET
2014 – “Disruptive Innovation in Manufacturing Engineering towards the 4th Industrial Revolution”
doi: 10.1016/j.procir.2014.10.013
ScienceDirect
8th International Conference on Digital Enterprise Technology - DET 2014 – “Disruptive Innovation in
Manufacturing Engineering towards the 4th Industrial Revolution
Direct toolpath generation based on graph theory for milling roughing
Daniela Popescua, Florin Popistera*, Sorin Popescua, Calin Neamtua, Mircia Gurzaua
Technical University of Cluj-Napoca, Blvd. Muncii 103-105, Cluj-Napoca, 400641, Romania
* Corresponding author. Tel.: +40 742 394980. E-mail address: florin.popister@muri.utcluj.ro
Abstract
The paper presents a tool path generation methodology for roughing operation based on the oriented graph theory. The cutting areas are
identified using an original method that is based on a bicolor and binary map. The toolpath is generated using the searching Dijkstra algorithm
inside a graph in order to find the single-source shortest path. The method was employed in order to be applied on ordered and/or unordered
point clouds. The entire algorithm was implemented into a mathematic calculus solution which allows the import of point clouds and its
processing until the final NC code is generated.
© 2014 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Scientific Committee of the “8th International Conference on Digital Enterprise Technology - DET
2014.
Keywords: direct toolpath, graph theory, cloud points, roughing
1. Introduction
More and more researchers are studying the problem of
direct tool path generation from point clouds because the
devices capable of creating such geometry are becoming more
efficient. These point clouds can be generated by 3D scanning
of a real object (contact or non-contact) using Computer
Tomography (CT) or by using photogrammetry.
Machining toolpath generation directly on point cloud
shortens the reverse engineering cycle of a component and
can reduce costs by eliminating the scan processing costs.
Thus, the filtering stage, the mesh generation stage and the
surface generation stage can be completely eliminated.
Usually the software of the scanning equipment (either by
palpation, or laser, or optical) allows a primary processing of
the point cloud so that it can be filtered or cropped to remove
unwanted points.
As show in [1], rapid generation of contour from massive
point cloud data is the crucial step for feature-based model
reconstruction. Based on the outline of a point cloud a clear
distinction can be made between the area where the material
is to be chipped and the area where it must remain intact.
In [2] the authors propose an algorithm for generating
roughing and finishing tool paths on measured data directly
by using Moving Least-Squares (MLS) method. In their
algorithm the roughing is performed in a slice-by-slice
manner to remove material rapidly. They generate an offset
point cloud for machining allowance, and slicing data points
were fitted by using MLS method to generate offset curve for
ball-end cutter.
Using the same technique for slicing the cloud point in
order to extract the machining regions (contour curves and
their inclusion relationships), [3] uses a boundary extraction
algorithm for determining the machining region. To avoid the
computational difficulties, they develop an algorithm to
calculate the finishing tool-path based on well-known 2D
geometric algorithms, such as 2D curve offsetting and
polygonal chain intersection algorithms.
In [4] the authors present a strategy for direct tool path
generation based on B-spline surfaces constructed using
extracted point data form initial cloud as control points. They
generate the adaptive (region-based) tool paths that overcome
the inherent drawback (redundant tool paths).
© 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/3.0/).
Peer-review under responsibility of The International Scientifi c Committee of the 8th International Conference on Digital Enterprise
Technology - DET 2014 – “Disruptive Innovation in Manufacturing Engineering towards the 4th Industrial Revolution”

76 Daniela Popescu et al. / Procedia CIRP 25 ( 2014 ) 75 – 80
In [5] the authors present a method based on generating
iso-planar piecewise. The primary direction of the generated
iso-planar tool paths is derived from the projected boundary
of the discrete points. A projected cutter location net (CL-net)
is then created, which groups the data points according to the
intended machining error and surface finish requirements.
In [6] a point-based conformal map is employed to build
the parameterization cloud point based on discrete Laplace
equation. They use also Iso-parametric curve determined in
base of normal curvature.
Surfel is the term used by [7] to describe the association of
“surface” and “element” in an entity which is used for
rendering the scanned data. Direct tool path generation is
made using a point-based model which is created using an
elliptical Gaussian re-sampling filter that is based on a signal
re-sampling algorithm. They use a re-sampling filter based on
Nyquist criterion to transform the input data (random cloud
point) into a continuous surface and then re-sample the
continuous surface. Because the re-sampled data can
accurately represent the original surface, tool paths can be
generated using CL. The CL point is found by lowering the
cutter in the direction normal to the projection plane until
contacts with the re-sampled points are made.
Using an algorithm based on 3D biarc fitting technique it is
possible to generate a 5-axis machining tool-path as shown in
[8]. They construct a three-dimensional (3D) triangular mesh
from the digitizing data with the mesh points considered as
the tool contact locations. The tool axis orientations which
must be determined in 5-axis tool-path are calculated using
3D biarc fitting.
[9] Since such planning falls into the iso-parametric
category, it intrinsically depends on the parameterization of
point clouds. Accordingly, a point-based conformal map is
employed to build the parameterization. Based on it, formulas
of computing path parameters are derived, which are much
simpler than the conventional ones. By regularizing parameter
domain and on the basis of the previous formulas, boundary
conformed tool-path can be generated with forward and side
step calculated against specified chord deviation and scallop
height, respectively. Experimental results are given to
illustrate the effectiveness of the proposed methods.
As shown in the literature, it has not been reported yet any
research aimed directly at tool path generation based on graph
theory. The majority of developed methodologies use slicing
and various algorithms and methods to reduce the number of
points and to generate the tool path.
2. Methodology for Direct tool path generation in case of
roughing
Current digitization technology allows scanning free form
type surfaces and directly obtaining the point cloud, Fig. 1, or
in some cases directly obtaining the 3D meshes [10, 11].
The algorithm presented in this paper can be applied in
case of roughing and finishing with commercial milling tools
in case of both ordered point clouds and unordered point
clouds.
Fig. 1. Ordered point cloud (top) and unordered (bottom)
Fig. 2. General algorithm
Starting from the point cloud direct tool path generation is
done by following the steps below (Fig. 2):
1. Alignment of the point cloud: its orientation in space,
establishing the axes system and determining the upper
and lower limits on the Z axis.
2. Cloud segmentation on Z axis: it is done by slicing on
perpendicular planes to the Z axis, the slicing pitch is
determined taking into account the parameters of the
cutting operation (tool geometry, material, feed rates,
etc.) and the geometry of the component that will be
cut.
3. Point recovery: beginning with the second segmentation
plane all points that are found above the sectioning
plane in which the tool path is calculated are projected
on that plane from top to bottom. This operation aims to

77
Daniela Popescu et al. / Procedia CIRP 25 ( 2014 ) 75 – 80
use all points to generate the tool path correctly and to
avoid collisions with already chipped areas
4. Cloud segmentation on XY plane: for each layer
obtained from the previous operation a grid is made. It
will be generated starting from the x=0 and y=0 point.
The lines and columns of the grid that bound the point
cloud contour are doubled in order to generate a
trajectory that allows correct processing of the contour.
5. Identify the milling area: the grid defined at 3 is
individually analyzed and each unit is assigned a value
of 1 if it contains points (red color) and 0 if it doesn’t
contain points (white color). This step takes into
account both points belonging to the slicing plane and
those obtained at 4 – point recovery.
6. Direct tool path generation: using the properties of
oriented graphs the toolpath is generated.
To illustrate the mathematical apparatus and the
implementation of the processing algorithms, a free form type
Bèzier surface was generated, Figure 3, which was later
physically materialized and scanned to obtain ordered and
unordered point clouds. The surface was generated using the
equation:
ܾሺݑǡݒሻൌσσ൫ଵଶ
௜൯ቀଵଶ
௝ቁ
ଵଶ
௝ୀ଴
ଵଶ
௜ୀ଴ ݑ௜ሺͳെݑሻଵଶି௜ݒ௝ሺͳെݒሻଵଶି௝ܽ௜ାଵǡ௝ାଵ (1)
where ݑאሾͲǡͳሿǡݒאሾͲǡͳሿ
The points’ coordinate matrix was generated to obtain a
Bézier surface having a complex free form type geometry
which contains peaks and valleys and a caldera. According to
the methodology, which is focused on processing complex
surfaces using three – axes milling machines, the surface was
mathematically modelled so that it does not contain
overhangs, i.e. a line parallel to the Z axis must intersect the
surface at one point. The set of parameters was chosen so that
the generated surface fits into a 50 mm side square.
Fig. 3. Bèzier surface used as a standard
The surface was machined on a milling machine and then
scanned using contact and non–contact scanning equipment to
obtain ordered (47817 points) and unordered point clouds
(196057 points).
When machining complex surfaces using 3-axes CNC
machines, the tool path can be described generally as a
function of two variables:
ݖൌ݂ሺݔǡݕሻ (2)
The point cloud can be described by the equation:
ܽൌอݔଵݕଵݖଵ
ݔଶݕଶݖଶ
ݔ௡ݕ௡ݖ௡อ (3)
If we make the following notations:
ܾݔൌอݔଵ
ݔଶ
ݔ௡อǢܾݕൌอݕଵ
ݕଶ
ݕ௡อǢܾൌอݖଵ
ݖଶ
ݖ௡อ (4)
Then, to determine the workspace, we will determine the
maximum values of the vectors bx, by and bz.
In order to optimize the trajectory calculation, the first step
is to create segmentation on the Z axis, thus obtaining planes
parallel to the XOY plane as shown in Figure 4.
Fig. 4. Sectioning the cloud point
Setting then slicing pitch is based on quality of the
geometry of the machined surface, quality of the point cloud
(density, ordered, etc.), milling speed, quality of the milled
material, etc. For proper results the authors recommend that
the distance between the sectioning planes (dz) is not greater
than:
݀ݖൌ݀כͲǤ͵Ͷ
Τ (5)
where d is the diameter of the mill.
To determine the number of planes with which the cloud
point is sectioned on the Z axis, it must not be a fractionary or
mixed number, thus the integer will be extracted from the
following mathematical expression:
݊ൌ௭೘ೌೣି௭೘೔೙
ௗ೥ (6)
where, zmax, zmin – represent the minimum and maximum
values of the matrix b, dz – represents the distance (depth of
cut) between the planes determined with formula 5.
If the result of equation 5 is not an integer then it is
rounded positively, the counting of the section planes begins
with zmin. The section planes are denoted with zi where ziܰכ .
Point recovery
The third step of the algorithm is the point recovery, Figure
5, it involves the projection of all points that are found above

78 Daniela Popescu et al. / Procedia CIRP 25 ( 2014 ) 75 – 80
a sectioning plane. The projection of the points from the upper
layers (from Zmax to Zmin) enables rapid determination of tool
paths through the oriented graphs method taking into account
the upper layers already machined. This operation is done by
changing the Z coordinate of the points that are above the
current sectioning plane. The projection is done on the
sectioning planes beginning with the sectioning plane at zmax.
Fig. 5. Point recovery
The identification of the points is made by comparing the
point’s Z coordinate with the height of the sectioning plane.
The index function for each layer is defined, which is 0 if
the layer doesn’t have any points and it receives the value 1 if
there are points in the layer.
For each point a[[j]] ( j gets values from 1 to Np = number
of points) is calculated (i = the smallest integer higher or
equal to (the difference between the maximum value z1 and
the value of the point a[[j,3]]) divided at the height of a layer
(dz) and if this number is 0, then the value 1 is assigned:
݅ൌቒ௭ଵି௔ሾሾ௝ǡଷሿሿ
ௗ௭ ቓǢܫ݂ሾ݅ൌͲǡ݅ൌͳሿ; (7)
Cloud segmentation on XY plane
To determine the milling area, the oriented graph theory is
used. In the first phase a constant pitch grid, Figure 6, is
generated on both axes (x and y), the size of two grid lines
can be approximated as being equal to the radius of the mill.
In the second phase it identifies the areas of the grid where
points are found. This is done by comparing each point in that
layer with a series of discrete values representing the grid
Fig. 6. Creating the grid
nodes. In this operation are taken into account all the points
obtained after the recovery operation as can be seen in the
figure below.
Each space in the grid that contains at least one point is
numbered 1 (red) and will be part of the area that will not be
machined for that layer. Spaces containing no points are
numbered 0 (white) and will be part of the area for which the
milling trajectory will be determined as shown below.
Using tandem numbering and coloring enables the creation
of a binary and bi-color map, Figure 7, whose nodes are used
to generate the trajectory.
Fig. 7. Binary and bi-color map
Direct tool path generation
Trajectory calculation is made based on the contact point,
CC point, which can be found on the grid. The distance
between the grid lines was chosen to be ½ of the milling
tool’s diameter, thus allowing the use of different milling
geometries using a single algorithm: oriented graphs theory.
Define the following oriented graph:
ܩൌሺܺǡܷሻ (8)
where X is the set of vertices, in our case represented by
the set of grid nodes, and U is the set of arcs, represented in
our case by the set of points.
Thus any arc u Є U will be represented through
ݑൌሺݔǡݕሻݓ݄݁ݎ݁ݔǡݕאܺݏ݅ݔ്ݕ (9)
To determine the trajectory of the milling tool the chain
sequence of arcs:
ܥൌሾݑଵǡݑଶǡڮݑ௡ሿ (10)
With the condition that any arc ui from this array shares an
extremity with ui-1 , and the other extremity is shared with ui+1
for any i=1,...,n-1.
If all the arcs from the same L chain have the same
orientation that is given by direction from x0 to x
n then the
chain is called a road.
The road in an oriented graph G(X,Y) is an array of peaks:
ܶൌሾݔଵǡݔଶǡڮݔ௡ሿ (11)
With the property that (x0, x1), (x1, x2), ….,(xn-1, xn) Є U
(are arcs of the graph G).
The peaks x0 and xr are called extremities of the road D. If
the peaks x0 ,x1 , ... , xr are distinct, the road D is called
elementary. Thus, an oriented graph G is called connected if

79
Daniela Popescu et al. / Procedia CIRP 25 ( 2014 ) 75 – 80
for any two distinct peaks x and y there is a chain of
extremities x and y in G.
In order to determine the shortest path in an oriented graph,
one can use Dijkstra’s algorithm or the Roy-Floyd algorithm.
Dijkstra’s algorithm is a search in a graph algorithm that
solves the single-source shortest path problem for a graph
with non-negative edge path costs, producing a shortest path
tree [12].
Generating the peaks, v, of the grid using the resulted grid
is achieved determining the indexes of the points that can be
touched by the milling tool generating minimal paths in order
of increasing lengths.
After determining the shortest path to the area that will be
machined (the graph that contains type 0 or white nodes) the
scallop area between two successive passes of spherical
geometry mills is calculated based on the formula below:
ܰሾሼʹܥ݋ݏሾݐሿǡʹ൅ʹܵ݅݊ሾݐሿሽȀǤݐ ՜െܲ݅ ͵
Τሿ
(12)
ሾሼʹܥ݋ݏሾݐሿǡʹ൅ʹܵ݅݊ሾݐሿሽൌ
ൌሼʹ൅ʹܥ݋ݏሾ߬ሿǡʹ൅ʹܵ݅݊ሾ߬ሿሽǡሼݐǡ߬ሽሿ
where N-forms the spherical profile
results
ሼሼݐ՜െగ
ଷǡ߬՜െଶగ
ଷሽǡሼݐ ՜గ
ଷǡ߬՜ଶగ
ଷሽሽ (13)
3. Validation
The surface of Figure 8 was machined on CNC equipment
and it was validated in terms of shape deviations by point by
point measuring and comparing with the CAD model using a
multisensor CMM. After the validation, the surface was
digitized, Figure 9, using the contact and non-contact methods
in order to obtain ordered and unordered point clouds (Figure
1).
Fig. 8. Machined standard surface
After the digitization the ordered cloud point contains
47817 points while the unordered one contains196057 points.
For the segmentation on the Z axis a pitch of 0.5 mm was
chosen, that corresponds to the depth of cut, thus resulting 25
layers.
The distance between two consecutive trajectories is ½ of
the milling tool’s diameter, for validation two types of milling
tools were used: ball mill and end mill having diameters of 4
mm with two cutting areas (flutes).
After calculating the trajectory, the NC program is
generated directly in Mathematica according to the
requirements of ISO 6983-1 regarding program format and
definitions of address words.
Using the same parameters for segmentation (both Z axis
and the plane XOY) 36763 program lines were obtained in
case of the ordered point cloud and 38411 lines in case of the
unordered point cloud.
Fig. 9. Equipment used to digitize the master part: contact scanning
equipment (up) and multisensor measuring and scanning equipment (down)
The Figure 10 shows the result of machining with two
different mills (a, b, c and d) in case of using a distance of ½
of the milling tool’s diameter between the trajectories and
constant machining allowance.
Ball end mill
End mill
Roughing based on an ordered cloud using
a) using ball end mill
b)using end mill
Roughing based on an unordered cloud using
c) using ball end mill
d) using end mill
Fig. 10. Results of milling with different cutting tools

80 Daniela Popescu et al. / Procedia CIRP 25 ( 2014 ) 75 – 80
To validate the algorithm and the work method, the
machining process was continued in order to obtain the
finished masterpart. Thus, based on the roughing algorithm,
the finishing methodology was developed using broadly the
same concept and tool support (Mathematica).
The differences between the two procedures (roughing and
finishing) are described below.
To obtain the trajectory for the finishing operation in case
of using the ordered point clouds, the procedure is the
following:
1. The alignment of the point cloud: see roughing
operation
2. Cloud segmentation on Y axis – the point cloud is
sliced in the plane XOZ with a minimum pitch of 0.1 mm
3. Establishment of Control Points: the identification of
which points will be taken into account when generating the
toolpath
4. Direct tool path generation: generating the offset
trajectory using Bezier curves and using CL (Cutter Location)
In case of the unordered cloud points, Control points are
established using the following algorithm:
• Dividing the machined area into 0.2 mm side
squares, the intersection of each of the squares’ diagonals will
be a control point for the Bezier curve
• Check that in the set of points within the square there
is a point which coincides with the intersection of the
diagonals: if there is, it becomes a control point; if there isn’t,
the Z coordinate of the intersection of the diagonals is
determined using all the points belonging to the cloud point
which can be found in a 5x5 matrix centered on the square for
which the z coordinate is determined.
After the final machining operation, the master part was
scanned and compared to the initial 3D model. The
differences are highlighted in the color map, Figure 11.
Fig. 11. Deviation analysis between CAD model and machined surface in
case of: ordered cloud point (up) and unordered cloud point (bottom)
4. Conclusions
Direct tool path generation using the oriented graphs’
properties is the first approach in literature as was shown in
the introduction to this article and allows rapid generation of
machining trajectories based on an algorithm that computes
the shortest path between two vertices.
The presented working methodology has been
implemented in a mathematical program for general use and
has been validated on the surface of a master part generated
by a mathematical equation (1). For the validation process
even the finishing operation was done using a methodology
derived from the roughing process in order to be able to
compare the master part with the CAD model using a
multisensory CMM.
The presented methodology can be used both for ordered
and unordered cloud points. Even though between the number
of points there is a difference of four times, the number of
lines of the NC program is about equal, however at this time
the NC program has quite many lines. Using the oriented
graph proprieties the interference that can appear during tool
path generation process directly from measured or scanned
data points allows generating the shortest path provides a
trajectory optimization to minimize air movement of the
milling tool. The method for identifying the area to be
machined is original and highlights for the first time in binary
and bicolor map the areas that will generate the toolpath. Even
if the implementation in Mathematica enables automatic
generation of NC code, the number of program lines is very
large; this should be optimized in future.
5. REFERENCES
[1]. Cheng, X., D. Jia, and Y. Liu, A fast contour generation algorithm of
massive point cloud data. Tongji Daxue Xuebao/Journal of Tongji
University, 2012. 40(10): p. 1559-1563.
[2]. Han, Y.A. and Y.J. Zhang, Tool path generation on measured data using
moving least squares method. Jisuanji Jicheng Zhizao Xitong/Computer
Integrated Manufacturing Systems, CIMS, 2011. 17(3): p. 638-642.
[3]. Park, S.C. and Y.C. Chung, Tool-path generation from measured data.
Computer-Aided Design, 2003. 35(5): p. 467-475.
[4]. Yin, Z., Adaptive tool path generation from measured data. Proceedings
of the Institution of Mechanical Engineers Part B-Journal of Engineering
Manufacture, 2004. 218(1): p. 103-111.
[5]. Feng, H.Y. and Z.J. Teng, Iso-planar piecewise linear NC tool path
generation from discrete measured data points. Computer-Aided
Design, 2005. 37(1): p. 55-64.
[6]. Zou, Q. and J.B. Zhao, Iso-parametric tool-path planning for point
clouds. Computer-Aided Design, 2013. 45(11): p. 1459-1468.
[7]. Yau, H.T. and C.Y. Hsu, Generating NC tool paths from random
scanned data using point-based models. International Journal of
Advanced Manufacturing Technology, 2009. 41(9-10): p. 897-907.
[8]. Chui, K.L., W.K. Chiu, and K.M. Yu, Direct 5-axis tool-path generation
from point cloud input using 3D biarc fitting. Robotics and Computer-
Integrated Manufacturing, 2008. 24(2): p. 270-286.
[9]. Zou, Q. and J. Zhao, Iso-parametric tool-path planning for point clouds.
Computer-Aided Design, 2013. 45(11): p. 1459-1468.
[10].Popişter, F., D. Popescu, and D. Hurgoiu, A new method for using
reverse engineering in case of ceramic tiles. Quality - Access to Success,
2012. 13(SUPPL.5): p. 409-412.
[11].Morovic, L. and P. Pokorny, Optical 3D Scanning of Small Parts, in
Automation Equipment and Systems, Pts 1-4, W.Z. Chen, et al., Editors.
2012, Trans Tech Publications Ltd: Stafa-Zurich. p. 2269-2273.
[12].Mehlhorn, K. and P. Sanders, Algorithms and Data Structures: The
Basic Toolbox. 2008: Springer.