Original article

Proc IMechE Part E:

J Process Mechanical Engineering
A gantry robot system for cutting single 0(0) 1–11
! IMechE 2020
Y-shaped welding grooves on plane Article reuse guidelines:
sagepub.com/journals-permissions
workpieces DOI: 10.1177/0954408920960425
journals.sagepub.com/home/pie

Xiaobing Hu1, Xi Deng1 and Qingxiang Zhao2

Abstract
To guarantee the strength and precision of the final welding assemblies, it is necessary to cut welding grooves before
welding thick workpieces. General methods to cut welding grooves on plane workpieces need much manual assistance,
and some even need manual operation purely. Therefore, this paper proposed a robot system for cutting Y-shaped
welding grooves with full automation. Flame cutting technology has been adopted, requiring no jig to fix workpieces,
which also causes no direct vibration to robot structure. Vision-based sub-system firstly captures the edges to be cut,
which are composed of continuous points, and a laser range finder (LRF) starts to obtain the thickness of the edges
precisely. To convert these edges into the trajectory of flamer, Least Square Method and Hermite Interpolation are
respectively utilized to fit lines and curves. Robot system subsequently computes the motion-related parameters
according to the position of the edges and geometric parameters of the desired welding grooves. The inverse kinematics
of this robot is solved by geometry methods, which decreases computation burden and saves much time compared with
traditional algebra method. Another core novelty is that a velocity planning method combining optimization algorithm
has been put forward, which, we think, is not only useful in this gantry robot but also benefits other motion axes with
heavy load. This further reduces vibration. Finally, the simulation and experimental results both prove the feasibility of
this system. To date, no available robots or machines tool can finish this process with full automation (to the best of our
knowledge).

Keywords
Gantry robot, cutting welding grooves, inverse kinematics, velocity planning, automation
Date received: 28 February 2020; accepted: 19 August 2020

Introduction
beam cutting4 and flame cutting.5 The principle of
Welding is an important process in manufacturing the former two is that remove extra materials directly
industry, whose automation degree has been via mechanical cutting, while that of the latter two
improved greatly with the development of robotics. ways is making use of high temperature to fusion
However, one of its pretreatments, cutting welding and remove extra materials, in which tools do not
grooves, still requires to be automated urgently. To touch workpiece directly so that jigs are unnecessary
ensure the strength of the final welding assemblies,1 to to fix workpieces. The disadvantage of flame cutting
ensure thorough welding, and to guarantee high weld- technology is also evident. It is not as easily to control
ing shape tolerance, it is necessary to cut precise weld- as others, causing relatively lower precision. This can
ing grooves, especially for workpieces with big be accepted in rough manufacturing stage. Herein, we
thickness. select flame cutting technology, considering its
There are generally 4 categories of welding
grooves: I-shaped, Y-shaped, X-shaped and
J-shaped welding grooves,2 in which the Y-shaped is 1
School of Mechanical Engineering, Sichuan University, Chengdu,
the most widely applied one. Its shape and corre- Sichuan, China
2
sponding parameters are shown in Figure 1(b), Department of Mechanical Engineering, Hong Kong Polytechnic
University, HKSAR, China
which is composed of two single Y-shaped welding
Corresponding author:
grooves located on two workpieces separately. Qingxiang Zhao, Department of Mechanical Engineering, Hong Kong
Currently, cutting a single Y-shaped welding groove Polytechnic University, Hung Hom, Kowloon, Hong Kong.
has generally four ways: milling,3 planning, laser Email: scdy_zqx@163.com
2 Proc IMechE Part E: J Process Mechanical Engineering 0(0)

advantages: no jig required, cost friendly and tiny The cutting velocity of tool also highly involves the
shock to robot itself. quality of welding groove. Smooth and even velocity
One conventional method to cut welding grooves solution benefits welding groove and causes little
in small lot production is that handler firstly sets the vibration to robot itself. This indeed is a problem of
inclination angle of flamer (equal to the angle of weld- velocity planning in multi-axis machines. It is also a
ing groove), where the flamer is fixed on a mobile problem of velocity planning for motion axes. One
vehicle, and then handler moves this vehicle running traditional velocity planning method is trapezoidal
along the edges to cut welding grooves. This is a cost velocity profile.14 It purses the shortest time but
friendly way but the quality depends on these factors: cause serious shock and vibration. Some researchers
velocity of the vehicle, precision of the angle, and proposed other related algorithms7,9,15,16 to address
handle’s ability. Therefore, to achieve full automation this concern. Sidobre and Desormeaux7 used cubic
of cutting welding grooves, two steps are really polynomial to generate trajectories for robots and
important: 1). Precisely set the pose of tool (flamer then planed its velocity with constrains. It requires
or other cutters), relative to workpiece, and 2). accu- that velocity must follow the change of trajectory
rately move the tool along the edges, which is also function, which may not be practical in application.
means generating accurate trajectory for tool. The derivative of trajectory function may cause
The first step is easier to realize compared with the sudden change. In Alzaydi,15 minimal motion time
second step, where tools like flamer does not touch and vibration-based velocity planning method was
with workpiece directly. The second step is more com- proposed for on-the-fly laser drilling. In this work,
plicated for different workpieces have different authors adopted trapezoidal acceleration to satisfy
shapes, which should be converted to trajectory of travel distance of motion axis, where the jerk is con-
flamer accurately. A common solution of this is that stant. This method will cause serious vibration to
generate trajectory on computer and cut welding robot. It can be seen that planning velocity should
grooves on a CNC machine tool or a robotic manip- generally find the trajectory function firstly and
ulator.6–9 This is definitely feasible and can get high- solve its derivatives. Therefore, the precision of the
quality welding grooves under one condition that the fitted trajectory function will directly affect velocity
3 D digital model of the workpiece has been obtained planning, which also brings computational burden. A
in advance. Besides, it requires much professional common sense learned from these works is that jerk
manual assistance. For example, Liu et al.10 adopted should be taken into account for reducing vibration.
a 6-DoF serial manipulator to cut Y-shaped grooves Thus, an inverse way is put forward in this work.
on sphere-pipe intersection whole, in which the digital Trajectory is a condition to be met rather than an
model of spherical workpiece is known before opera- important factor determining whole velocity
tion. By mapping the joint configuration of robot and planning.
pose of tool, they modeled the trajectory of robot end Learning from previous achievements, this paper
effector. This work is particularly for cutting spheri- desires to propose a robot system for cutting Y-
cal single Y-shaped welding groove, so the trajectory shaped welding grooves on plane workpieces with
is essentially smooth enough, and the process of solv- full automation. Our target is processing workpieces
ing trajectory is mathematically complex. Liu et al.11 with bigger size, so gantry robot is suitable since it has
used plasma cutting technology to cut welding high precision of motion and is cost friendly, com-
grooves on surface of a pipe, which considered the pared with its series counterpart. Before cutting, a
geometric model of pipe hole and tool compensation vision system is applied to capture the horizontal
mechanism. Complex computation is also an obvious position and one LRF is installed to solve the vertical
shortcoming. In Shi et al.,12 authors took plasma data of edges (thickness). Considering the irregularity
beam and series manipulators to cut welding grooves of the edges, two mathematic methods are employed
on pipe hole. Zhang et al.13 proposed a trajectory to fit lines and curves respectively, solving the trajec-
planning method based on the change rate of V- tories where flame should cover. Then combining the
shaped groove, but computing the change rate is position of edges (composed of continuous discrete
sometimes challenging since the curve is uncertain. points) and the geometric parameters of the desired
A direct neural network-based trajectory generation welding groove, robot system computes the pose of
method was proposed in Li et al.,6 where a Markov flamer, where the flame is regarded as a cylinder. To
decision process was applied for controlling the path solve the corresponding inverse kinematics (a
of CNC tool. Reinforcement learning including: common concern in robotics) in operation, we com-
Artificial Neural Networks (ANN), environment, pare traditional algebra method and geometry-based
action, agent, and reward function were also built. method in terms of computational burden. The result
Results show the feasibility of this method, but shows that the geometry-based method solves the
obtaining large number of datasets is impractical. inverse kinematics rapidly and precisely. Taking
Different learning algorithms will bring different tra- velocity planning into account is another core part
jectory precision. It took too much time to train the of this work, which is a factor should be considered
model. for all motion axes with heavy load. Based on the
Hu et al. 3

smoothness and continuity of ‘Sigmoid’ function, a It should be noted that parameters including hi , lf ,
velocity planning sub-system is designed constraining df and di are input into robot system manually, and
velocity, acceleration and jerk within allowable range. then other related parameters are automatically com-
Constrained optimization algorithm is employed to puted by robot system. These parameters jointly
solve the optimal parameters controlling velocity- determine the pose of flamer, relative to the local
time function. Finally, the simulation of cutting machining coordinate system. Firstly, the
trajectory proves the feasibility of this trajectory gen- horizontal distance Di and vertical distance Hi can
eration method. Machining several workpieces on this be solved by (1).
gantry robot furtherly shows that the robot system is
(
useful and feasible, in which the welding grooves are Di ¼ lf sinðhi Þ
smooth and with low tolerance. (1)
Hi ¼ lf cosðhi Þ  di þ di
This rest of this paper is organized as follows:
machining requirements and overview of the robot
are detailed in the next section, where we detail the This only positions flamer’s tip, and the angle
necessities for cutting a welding groove using flame between flamer and Z axis is ð180  hi Þ, which
cutting technology. In the ‘Methodologies’ section helps to decide the posture of flamer. Secondly, the
covers methodology including: how to identify the posture of flamer can be solved by the position of
type of an edge, method to generate trajectory for cutting point Pi and its slope. In addition, to ensure
flamer’s tip, the geometrical method to solve the thorough cutting (completely remove the blue area in
inverse kinematics and how to plan velocity. The pen- Figure 1(b)), we require: df  ðdi  di Þ  sinðhi Þ. The
ultimate section shows the trajectory and velocity parameters about flame ðdf ; lf Þ can be changed by
testing results on a typical workpiece. Several work- adjusting the flow rate of combustion gas (ethyne).
pieces were machined on this robot, to actually vali-
date this robot system. The final section concludes
this work. Overview of the gantry robot
Analyzing aforementioned requirements, five degrees
Machining principles and design of robot of freedom (DoFs) are needed for this robot in
In this section we detail the basic requirements for terms of motion. Covering three-dimension (3 D)
cutting a Y-shaped welding groove on plane work- position of cutting points needs three prismatic
pieces using flame cutting technology, according to DoFs, and satisfying edges whose tangent line direc-
which a brief digital model of a gantry robot is tion vary from point to point (as shown in Figure 1
designed. (a), the red flamers), needs one revolute DoF.
Different hi needs another revolute joint. The over-
Machining requirements view of this gantry robot is shown in Figure 2.
As shown in Figure 1(b), the parameters of a welding To obtain the coordinates of points making up
groove at point Pi ðxi ; yi ; zi Þ include: angle of groove edges to be cut, a calibrated vision system (two
hi , depth of groove ri , root face di and length Li . The stereo cameras, D1000-IR-120/Color) functions to
goal of this robot is to remove extra material (blue obtain the horizontal position data, i.e. ðxi ; yi Þ.
area in Figure 1(b)) automatically and precisely using Considering the varied thickness and deformation
flame cutting technology. There are two types of existing at edges, which leads to different height, i.e.
edges on a plane workpiece: interior edge (blue edge zi , we take one LRF (optoNCDT 1420 Micro-
in Figure 1(a)) and exterior edge (red edge in Figure 1 Epsilon) to find this value. Once getting the horizon-
(a)). Flamer locates at different side for the different tal data, robot treats the LRF as its end effector and
type of edges, as shown in Figure 1(a). Therefore, this moves along each point Pi , to get accurate zi ¼ ri . In
robot should be able to identify the type of an edge this paper, we focus on the system involved in cutting
automatically. Furthermore, the edges are often com- process, assuming that accurate edge data has already
posed of: lines and curves, so robot system should been obtained.
also convert the edges into the trajectory of flamer.
During cutting process, we assume that the flame is a
cylinder denoted by length lf and diameter df . The
Methodology
flamer should be perpendicular to the tangent line
of the cutting point Pi , and its projection in In this section, the key technologies about this robot
Xw Ow Yw should also be coaxial with the normal are detailed. They consist of: identifying the type of
direction Ow Yw , which can be mathematically speci- an edge, converting the points into trajectory of the
fied by Zf  Xw Ow Yw and Zf ?Ow Xw , as shown in flamer, and geometry-based method to solve the
Figure 1(b). Ow Xw and Ow Yw respectively point the inverse kinematics of robot. Another core is velocity
tangent direction and normal direction of Pi . planning algorithm.
4 Proc IMechE Part E: J Process Mechanical Engineering 0(0)

Figure 1. Basic illustration of cutting Y-shaped welding groove on a plane workpiece. Here we regard the flame injected from flamer
is a cylinder, denoted by df and lf . (a) When cutting interior edges and exterior edges, the flamer should locate at outside and inside
the edge respectively. To ensure the quality of final cut welding grooves, flamer must be perpendicular to the tangent line of the point
being cut, as the red flamer shows. (b) Necessary parameters about a welding groove which also involves in cutting process, are
denoted. In addition, local machining coordinate system and tool coordinate system at one cutting point Pi are also respectively
illustrated, which helps to intuitively solve the pose of flamer. The coordinate of each cutting point relative to the universal coordinate
system is obtained by vision sub-system and LRF.

Vision
System
LRF Workpiece
Y Axis
X Axis

Flamer1
Flamer2
Z Axis

U Axis

V Axis

Figure 2. Overview of the gantry robot and its corresponding digital model. Generally, gantry robot has higher motion precision
than series manipulators, and is cost friendly. In actual robot, there are two flamers (flamer 1 and flamer 2), and the flamer 2 that has
only three DoFs, is always perpendicular to workbench. Flamer 2 is only for quickly cutting off plane workpieces, which is often a
previous step of machining welding groove. So, herein we only take flamer 1 to cut welding grooves, which has 5 DoFs in total. It is not
a redundant robot, which brings two characterizations: (1) Singularity can somehow be avoided and solving the inverse kinematics
would be less complicated, and (2) Robot with five DoFs also suffers from lack of enough flexibility in operation.

Coordinate systems Joint coordinate systems (Ol  Xl Yl Zl ,

l ¼ f1; . . . ; 5g): locates at the center of each motion
Before operation, it is indispensable to set coordinate
axis.
systems. Based on the mechanical structure and
common principles, universal coordinate system,
Flamer coordinate ! system (Of  Xf Yf Zf ): locates at
the tip of flamer. Of Zf points to the injection direction
local cutting coordinate system, flamer coordinate
of flame (see Figure 1(b)).
system and joint coordinate systems are set respec-
tively, as shown in Figure 3.
Universal coordinate system (O  XYZ): locates at Identification of edge type
the origin of workbench and keeps immobile, as can Binary image of a workpiece, obtained from vision
be seen from Figure 1(a). sub-system, helps to identify the type of edge.
Local machining coordinate system Figure 4 shows the binary image of a typical work-
(Ow  Xw Yw Zw ): locates at cutting point, and the piece. The pixel value of workpiece is 1, while that is 0
three axes direct to normal, tangent, and vertical for background. In Figure 4, the edges where point A
directions respectively, as shown in Figure 1(b). and B locate, are exterior edges and interior edges
Hu et al. 5

Figure 3. Cartesian coordinate systems of this robot. The blue lines represent configuration of joints, and the green lines are rods
connecting two adjacent joints.

respectively. Moreover, any edge belongs to a closed

geometry figure. Several closed geometry figures
make up the contour of a workpiece. For each
closed geometry figure, its geometric center locates
inside the figure. By amplifying the binary image,
the type of edges can be identified. Firstly, amplify
every closed geometry figure with u times around its
geometric center, which in Figure 4 means (2),

kA0 C1 k kB0 C2 k
¼ ¼u (2)
kAC1 k kBC2 k

where k  k is Euclidean distance and u is a positive

number. Then, take the coordinate of the amplified
points (A0 ; B0 ) into the unamplified image. The type of
edge is identified by:(1) if the pixel value of the ampli-
fied point is 1, then this edge is interior;(2) if the pixel Figure 4. Binary image of a typical workpiece. Two cutting
value of the amplified point is 0, this edge is exterior. points ðA; BÞ locating on exterior edge and interior edge
The algorithm to identify an edge is mathematically respectively, are amplified around its corresponding geometry
center C1 ; C2. As can be seen, the amplified two points ðA0 ; B0 Þ
detailed in Algorithm 1. fall into the black and white area of the unamplified binary image,
respectively. This principle helps to identify the type of a contour.
Algorithm 1. Steps to identify the type of an edge.
Given: Line and curve fitting. We correct the points only in XOY
binary image I; edge m; plane, because the variation in vertical direction is
point A ¼ ðx; yÞ; A  m; Iðx; yÞ ¼ 1. small enough. Least Square Method (LSM) is used
Note: Iðx; yÞ is pixel value of pixel ðx; yÞ. to get an ideal line. To get smooth curves, we use
Steps: Hermite Interpolation (HI) method,17,18 which not
1). Resize I with u ¼ 1:1 times around geometric only fits control points but also fits the first derivative
center, and get the amplified point A0 ¼ðx0 ; y 0 Þ with of control points. HI method is only detailed for brief.
using (2). Every continuous 12 points are adopted to fit a
2). If Iðx0 ; y 0 Þ ¼ 1, edge m is an interior edge. segment of curve, as shown in Figure 5, in which 4
Otherwise, edge m is an exterior one. points from vision system are control points (marked
as red). So, the fitting can be denoted as finding
Flamer trajectory generation proper ai in (3).

Welding grooves on plane workpieces is usually a X 5

yðxÞ¼ ai xi
rough machining step, which means that both flatness
tolerance and straightness tolerance are low.  i¼0 (3)
yðxk Þ ¼ yk ; k ¼ 1; 2; 3; 4
Therefore, robot system needs to correct the points y0 ðxj Þ ¼ y0 j ; j ¼ 1; 4
deviating its subordinate edge. One more important
step is to convert the corrected points into the pose of
flamer, considering the parameters of welding groove Pose of flamer. It can be seen from Figures 1(b) and 5
and type of its edge. that the pose of flamer’s tip is jointly decided by the
6 Proc IMechE Part E: J Process Mechanical Engineering 0(0)

So, the bridge is 1w T, in which the interior edges and

exterior edges are totally different. Flamer works
anticlockwise, as the arrow in Figure 5 shows. For
interior and exterior edges, 1w T can be solved by (8)
and (9) respectively,

1
wT ¼ translðxi ; yi ; zi Þtrotzðai Þ (8)

1
wT ¼ translðxi ; yi ; zi Þtrotzð180 þ ai Þ (9)

where ai is jointly determined by angle between the

positive direction of tangent line and X axis, as shown
Figure 5. Illustration of cutting process in XOY plane. The red in Figure 5.
arrow refers to the machining sequence: anticlockwise direc- Since ai is the directional angle that X axis rotates
tion. A segment of curve edge consisting of 12 points from to the w X, it cannot be calculated just by slope of the
vision system is fitted. When cutting any point, the centerline tangent line, as Figure 6 shows. There are four cases
of flamer should be coaxial with the normal line of this point.
Using HI method can effectively obtain smooth curves com-
for the rotation
!
angle. By considering
!
the dot product
pared with rough unfitted obtained points which may cause of vector Ow Xw and vector OX, ai can be calculated.
sudden change of slope. In addition, we can easily get the slope The method to solve ai is detailed in Algorithm 2.
of each cutting point via the fitted function, which is conducive So far, the pose of flamer’s tip is built, which has
for solving the configuration of the 4th DoF. considered the machining requirements and all
dimensional parameters. This is an indispensable
bridge of solving robot inverse kinematics.
type of edge, parameters of welding groove, and
dimension of flamer. Therefore, the pose of flamer
Algorithm 2. The method to calculate ai .
relative to Ow  Xw Yw Zw can be denoted firstly via
homogeneous transformation matrix wf T in (4). 1. Get the slope k ¼ y0 ðxi Þ of the tangent line at
point Pi , get!the angle q ¼ atanðkÞ.
 
2. Set *vector OX ¼ ð1; 0; 0Þ, and calculate vector
w w
w
¼ f R pf
(4)

f T Ow Xw ¼ ðxiþ1  xi ; yiþ1  yi ; ziþ1 

013 1 *zi Þ.
*
3. Compute dot product: re ¼ dotðOX; Ow Xw Þ
4. get ai by:
where wf R; w pf respectively represent the rotation
if re > 0
matrix and position relative to Ow  Xw Yw Zw . Once
if 0 < q < 90 then ai ¼ q
getting 1w T, we can find the pose of flamer relative to
else ai ¼ q þ 180
O  XYZ (detailed later). Before computation, it
else if re < 0
should be noted that tranlsðÞ and trotxðÞ respectively
if 0 < q < 90 then ai ¼ q þ 180
represent translation matrix and rotation matrix in this
else ai ¼ q
paper. The rotation matrix wf R can be solved by (5),
else if re ¼ 0
if yiþ1  yi > 0 then ai ¼ 90

f R¼trotxð180 hi Þ
w
 trotzð90 Þ (5)
else ai ¼ 270

and w pf ¼ ½w xf ; w yf ; w zf T can be calculated by (6).

Kinematics of robot
8
>
>
w
x ¼0 Kinematics, including forward and inverse kinemat-
<w f
yf ¼ Di ¼lf sinhi  df coshi ics, targets at mapping joint configuration to end
(6)
>
> w zf ¼ lf coshi  ðzi  di Þ þ df sinhi effector configuration in robotics. Two approaches
:
2 for solving the inverse kinematics: 1) algebra-based
method using forward kinematics and 2) geometric
Solving the trajectory of flamer means getting the method, are compared.
continuous pose of flamer for every point Pi relative The configuration of joint i is defined as qi , and the
to the universal coordinate system, which is expressed distance between joint i and joint i þ 1 is defined as
in (7). li (see it in Figure 3). The forward kinematics can be
easily solved from Denavit-Hartenberg (D-H)
1
fT ¼ 1w T wf T (7) method, and denoted as 1f T ¼ 12 T 23 T 34 T 45 T 5f T, which
should be equal to the result in (7) in order to achieve
Hu et al. 7

Figure 6. Rotation angle from X axis to Xw axis. (1) and (2) ai ¼ atanðkÞ; (3) and (4) ai ¼ atanðkÞ þ 180 . For a closed figure (both
exterior and interior) to be cut, the posture of flamer should always follow the change of slope of cutting points. Since the cutting
sequence is required to be anticlockwise, the relative angle between Xw and X has four cases. The joint configuration of the 4th DoF
can then be solved.

precise cutting. Namely, the two poses of flamer rel- The joint configuration Qi ¼ ½q1    q5 T for any
ative to universal coordinate system both from work- cutting point Pi is solved. It is also obvious that the
piece and forward kinematics are equal. The joint singularity of this robot exists when q5 ¼ 0. No matter
configuration is derived from (10), but solving this how U axis rotates, the posture of flamer will not
equation set is challenging for it includes sinðÞ and change.
cosðÞ items.
Velocity planning
1
fT ¼ 1w Twf T ¼ 12 T 23 T 34 T 45 T 5f T (10)
Velocity planning is highly involved in precision and
motion stability. Axes with heavy load need smooth
To ease the computational burden, geometry- velocity solution, such as the X axis of this robot. In
based method is proposed to directly solve the inverse this subsection, we propose a ‘Sigmoid’ function-based
kinematics, which is validated by the forward kine- velocity planning algorithm, which concentrates on the
matics in (7) and (10). velocity of flamer’s tip relative to universal coordinate
As shown in Figure 1(b), the angle of groove hi is system, and the velocity of each joint can be accord-
only determined by the angle displacement of joint 5 ingly obtained by
(V Axis), so we can get q5 ¼ hi . The direction *of X4

always points to the opposite direction of Ow Xw , so O
Vf ¼ ½v13 ; x13 T ¼ J1 ðQÞQ0 (13)
q4 ¼ ai  90 . To get the values of other three joint
configurations (q1 ; q2 ; q3 ), the position of flamer’s tip
where v13 ; x13 ; J1 ðQÞ and Q ¼ ½q1 ;    ; q5  are
f Pðxf ; yf ; zf Þ relative to universal coordinate system is
1

helpful, which is calculated by (11). Then, the values of linear velocity, angular velocity, the pseudo-inverse
q1 ; q2 ; q3 via geometry relationship is solved by (12), of Jacobian matrix and joint configuration,
respectively.
Considering the fact that the axes with large
1
fP ¼ ½xf ; yf ; zf ; 1T ¼ 1f T½0; 0; 0; 1T (11)
motion range and velocity change are X and Y axes,
8 our velocity planning algorithm is only applied on
< q1 ¼ xf  l5 sðq5 Þcðq4 Þ  l4 cðq4 Þ  l2  l3 them. Others are actuated with Proportion-
q2 ¼ yf  l5 sðq5 Þsðq4 Þ  l4 sðq4 Þ (12) Integration-Differentiation (PID) algorithm to con-
:
q3 ¼ l1  zf  l5 cðq5 Þ trol position and velocity. The velocity solution is
formulated using a velocity-time function (14),
where sðÞ; cðÞ represent sinðÞ; cosðÞ respectively. which is composed of three piecewise functions
8 Proc IMechE Part E: J Process Mechanical Engineering 0(0)

controlling the velocity of in acceleration period which is often determined by the size of flamer and
½0; t1 Þ, in constant-velocity period ðt1 ; t2 , and in decel- the flow rate of combustion gas.
eration period ðt2 ; t3 . In (14), VI ; VE ; VM are initial
velocity, end velocity and maximum velocity respec- Trajectory and velocity test
tively, and a is a parameter regulating slope. A workpiece for testing is illustrated in Figure 7(a),
8 which is a typical plane workpiece used in portal
>
> V M  VI
>
> VI þ    0 t < t1 crane. The exterior edges include: AB; BC; CA, and
>
> t1
>
> 1 þ exp a t  the machining direction is anticlockwise (as the red
>
< 2
VðtÞ ¼ VM t1 t t2 arrow shows). The interior edge is a circle, and the
>
> V  VE starting and end cutting points are both D. As for the
>
> VE þ  M   t2 < t
>
> t3 welding groove, we set the angle of welding h ¼ 60
>
> t2 þ t3
>
: 1 þ exp a t 
2 and di ¼ 15 mm. To guarantee the stability, we set the
velocity at each intersection point to 0 mm/s, because
(14)
the pose of flamer in these points sometimes need to
be adjusted, namely VA ¼ VB ¼ VC ¼ VD ¼ 0 mm=s.
The reason why we choose ‘Sigmoid’ as basic func- The maximum velocity, acceleration, and jerk are
tion is that any order of its derivative includes ‘exp(x)’ respectively set to 50 mm=s; 100 mm=s2 ; 150 mm=s3 .
item, such that the velocity, acceleration and jerk are all The calculated trajectory of flamer for both exterior
continuous. The final velocity solution should satisfy and interior edges are shown in Figure 7(b) and (c). It
below constraints: (1)the planned maximum velocity, can be seen that the flame is always located at the
acceleration and jerk should be continuous and be desired pose. Two more workpieces are also tested,
smaller than the required one, and (2)the total displace- and their trajectories show that calculated flame pose
ment SðtÞ should be equal to the trajectory length. are all accurate.
Generally, calculus is utilized to compute the total Take the interior edge as an example to test the
length, but it is computationally complex, so we adopt displacement of each joint, and the result is shown
Newton-Cotes formula19,20 instead, as shown in (15), in Figure 7(d). The angle h ¼ 60 is a constant, so
that q5 is also a constant. As for q4 , it changes with
Z b X
4 the rotation angle ai , so there are two linear stages
SðtÞ ¼ fðxÞdx ¼ mk Vðtk Þ (15) and a rapid change period, where the linear velocity
a k¼0
of flamer is 0 mm=s, making it has enough time to
adjust its posture. The other three joints’ displace-
where mk ; tk are coefficients and timestamp. The ment varies following the position of flamer’s tip.
velocity planning aims to find the shortest time to Furthermore, we tested the velocity planning algo-
cut a segment of groove while satisfying motion rithm. The results of AB and CA segments are shown
requirements, so optimization algorithm with con- in Figure 7(e) and (f). Given the required velocity, this
straints is employed to find the optimal parameters. algorithm can obtain the desired solution. It shows
It can be denoted as (16), where we make sure the that velocity, acceleration and jerk in the whole
acceleration and jerk are all constrained. motion process are all continuous and are all within
the constrained range. To reduce the optimization
minimize t3 time cost, we import Particle Swarm Optimization
8 (PSO) besides using MATLAB fmincon toolbox.
< VM Vmax ; V0 ðtÞ V0 max (16) The results show that PSO is far more suitable for
00
s:t: V ðtÞ V00 max
: this velocity planning algorithm, and the time cost
Sðt3 Þ ¼ S is often with 0.8 s, while that of fmincon toolbox is
often more than 2.2 s. The reason why this phenom-
enon may be the latter should try more algorithms,
Test and experiments while the former solves the parameters directly.
Another comparison is about computational cost
In this section, we take one typical workpiece includ- for solving the inverse kinematics. It is obvious that
ing interior and exterior edges to verify the proposed every cutting point needs unique pose so that robot
methods. Firstly, the trajectory of flamer for cutting system needs to compute each of them separately.
each edge is generated. Then the corresponding veloc- Therefore, we compare the computation time between
ity planning solution is derived. algebra method (solving equation set shown in equa-
The length and diameter of an actual flamer is tion (10)) and our geometry-based method. It proves
50 mm and 20 mm, so that lf ¼ 50, and df ¼ 20. In that geometry-based method can directly find the con-
machining scenario, the position of flamer’s tip figuration of each joint rapidly (less than 0.01 s), while
should locate above the workpiece, so selecting the the former needs more than 1.37 s at least. So, if there
flamer and flame should make sure Hi > 30 mm, are thousands of points to be machined, the
Hu et al. 9

Figure 7. Trajectory and velocity planning results of machining a typical workpiece. (a) 3 D model of a workpiece for testing (unit:mm); (b)
3 D view of flame for cutting both exterior and interior edges; (c) Top view of flame’s pose; There is an interior edge (a circle) and five
exterior edges (four lines and an arch) on this workpiece. From the trajectory results, we can find that the pose of flamer satisfies the
machining requirements. Once the flamer is working, extra materials can be removed and ideal welding grooves can be obtained. (d) The
displacement of each joint when cutting the interior edge; (e) and (f) The velocity solution of flame’s tip relative to the universal coordinate
system. Since the angle of welding groove is constant, q5 is reasonably to keep constant. The thickness of this workpiece is also constant, so
q3 is also constant. Since the slope of cutting point changes suddenly near the quadrantal points of circle, q4 should adjust its configuration
to adapt it. The planned velocity is smooth enough, and both acceleration and jerk satisfy given requirements. The total motion time is
longer compared to traditional trapezoidal-acceleration solution, but it is far better in terms of shock and impact to robot.

computation cost is really burdensome for computer.

Although the whole motion time calculated via the
proposed velocity planning algorithm is longer than
traditional constant-acceleration method, it effective-
ly reduces vibration to robot itself.
Note that this robot is not redundant, so each
group of joint configurations is unique for a unique
cutting point. It is acceptable to use geometry-based
method to solve joint configuration.

Results
A customized 5-DoF gantry robot was manufactured
by Aoling Zhineng company (Guangzhou, China)
which provides us with mechanical structure and
motion control system (shown in Figure 2). All the
sub-systems including: image extraction, LRF inte-
gration, solving inverse kinematics and velocity plan- Figure 8. Images of welding grooves. (a) Welding groove on a
ning, run in MATLAB 2019a. The motion control line edge machined by traditional method, which has low pre-
commands are transferred to motion control system cision. (b) and (c) welding groove on a curve edge and a line
by Ethernet Communication. Acetylene and oxygen edge machined by our gantry robot, with smoothness and low
are used as combustion gas, which can generate flame geometry tolerance. The main reason why the welding groove
with 3200  C (enough to cut off steel workpieces). cut by traditional method behaves worse is that motion
This gantry robot can deal with workpieces with max- velocity of flamer is not continuous such that flame cannot
thoroughly remove extra material. One more reason is that the
imum plane size at 2:8 m  3:2 m stably. flamer would unavoidably touch with workpiece, so interaction
We tested several workpieces on this gantry robot, force will lead low quality. In our robot, welding grooves after
as shown in Figure 8, and results show that the weld- removing slag can be utilized for welding directly, which is
ing grooves are of high precision and smoothness. similar to the product of milling.
10 Proc IMechE Part E: J Process Mechanical Engineering 0(0)

It also should be noted that in whole machining pro- Currently, this robot can only aim for cutting
cess, this robot is with full automation except that welding grooves on planes workpieces, where we
inputting parameters of welding grooves needs treat the axis Ow Yw is parallel with the plane XOY.
manual assistance. Compared with the welding The reason behind this is that the robot has only 5
grooves manufactured by traditional half-automatic DoFs, and sensing system is not capable enough to
machine (Figure 8(a)), the products cut by this detect the angle between Ow Yw and plane XOY. We
gantry robot are of higher quality. will solve this problem by introducing another axis on
the robot and another vision system to measure the
Discussion and conclusion angle, enabling this robot to cut welding grooves on
curve faces, which is our further work.
This paper proposed a gantry robot system for cut-
ting Y-shaped welding grooves on plane workpieces Declaration of conflicting interests
with full automation. Only the parameters involved in
welding grooves need to be input to robot system The author(s) declared no potential conflicts of interest with
manually. It changes the traditional method to cut respect to the research, authorship, and/or publication of
welding grooves (CNC or half automatic machine this article.
operated by handlers).
The first novelty is a about the structure of the Funding
robot. Gantry robot has much higher stability than The author(s) disclosed receipt of the following financial
common serial manipulators. In addition, this robot support for the research, authorship, and/or publication
can deal with plane workpieces with maximum size at of this article: This work was financially supported by the
2:8m  3:2m. The limitation of this gantry robot is Sichuan Science and Technology Support Plan
also evident that it cannot deal with workpieces (2018GZ0125), Made in China 2025 Sichuan Provincial
with any shield edges, because it only has 5 DoFs. Action Plan (2017ZZ018), and SCU-ZIGONG Union
Moreover, the beam (Y axis) limits the task space Project (2018CDZG-3).
somehow. This work used vision system and LRF
to get the edge data of workpiece, which is tradition- ORCID iD
ally done by borrowing CAD model. It greatly
Qingxiang Zhao https://orcid.org/0000-0001-7524-5342
reduces the amount of manual assistance and
upgrades the automation degree of cutting welding
grooves. References
The second core novelty is that geometry-based 1. Çevik B. Analysis of welding groove configurations on
method is employed to solve the inverse kinematics strength of S275 structural steel welded by FCAW.
of the gantry robot. The pose of the flamer is solved J Polytech 2018; 0900: 489–495.
2. Sharma V and Shahi AS. Effect of groove design on
by the given parameters of welding grooves and the
mechanical and metallurgical properties of quenched
position of cutting points, where robot system con-
and tempered low alloy abrasion resistant steel welded
trols its joint configuration to ensure flamer can
joints. Mater Des 2014; 53: 727–736.
follow the pose. Then, starting from the structure of 3. Li G, Zhou H, Jing X, et al. Modeling of integral cut-
robot, we make use of geometrical relation between ting tool grooves using envelope theory and numerical
adjacent joints directly to solve the joint configuration methods. Int J Adv Manuf Technol 2018; 98: 579–591.
for all cutting points, which is an effective way to 4. Ma R, Ji L and Yan T. Laser multi-focus precision
solve inverse kinematics. Compared with the tradi- cutting of thick sapphire by spherical aberration recti-
tional method using algebra to solve equation set, fication. Opt Lasers Eng 2019; 126: 105876.
this geometry-based method is less computationally 5. Pond TL and Martin CR. Electrical characteristics of
complex, but it can only solve the inverse kinematics the oxyfuel flame while cutting steel. Exp Therm Fluid
for this robot and this cutting process. Sci 2019; 112: 109985.
As for the motion axes with heavy load and high 6. Li B, Zhang H, Ye P, et al. Trajectory smoothing
stability requirement, ‘Sigmoid’ function-based veloc- method using reinforcement learning for computer
ity algorithm is proposed, considering the velocity, numerical control machine tools. Robot Comput Integr
acceleration and jerk. Thus, strong shock and vibra- Manuf 2020; 61:: 101847.
7. Sidobre D and Desormeaux K. Smooth cubic polyno-
tion caused to robot can be effectively reduced. This
mial trajectories for human-robot interactions. J Intell
method also employs optimization algorithm to solve
Robot Syst 2018; 95: 851–869.
the optimal parameters controlling velocity-time 8. Shen X, Xie F, Liu XJ, et al. A smooth and undistorted
function. As we know that jerk is highly involved in toolpath interpolation method for 5-DoF parallel kine-
vibration, limiting the jerk within acceptable range matic machines. Robot Comput Integr Manuf 2018; 57:
definitely decreases the vibration of a machine. On 347–356.
the other hand, this velocity planning method will 9. Kucuk S. Optimal trajectory generation algorithm for
extend motion time. It can be applied in similar sit- serial and parallel manipulators. Robot Comput Integr
uations where a motion axis has heavy load. Manuf 2017; 48: 219–232.
Hu et al. 11

10. Liu Y, Liu Y and Tian X. Trajectory and velocity plan- Appendix
ning of the robot for sphere-pipe intersection hole cut-
ting with single-Y welding groove. Robot Comput Integr Notation
Manuf 2018; 56: 244–253. a parameter controlling the slope of
11. Liu Y, Shi L and Tian X. Plasma cutting torch trajec- the planned velocity curve
tory planning for main pipe hole cutting with welding
di root face of welding groove
groove and root face. Int J Adv Manuf Technol 2017;
df diameter of flamer
93: 4329–4343.
12. Shi L, Song R and Tian X. Plasma beam radius Di horizontal distance from edge to
compensation-integrated torch path planning for flamer’s tip
CNC pipe hole cutting with welding groove. Int J Adv DoF degree of freedom
Manuf Technol 2017; 5–8: 1971–1981. ri depth of groove
13. Zhang Y, Lv X, Xu L, et al. A segmentation planning FC centerline of flamer
method based on the change rate of cross-sectional area Hi vertical distance from workpiece’s
of single V-groove for robotic multi-pass welding in upper surface to flamer’s tip
intersecting pipe-pipe joint. Int J Adv Manuf Technol HI Hermite interpolation
2019; 101: 23–38. J Jacobian matrix of robot
14. Klancar G and Blazic S. Optimal constant acceleration k slope of tangent line at point Pi
motion primitives. IEEE Trans Veh. Technol 2019; 68: lf length of flamer
8502–8511.
LRF laser range finder
15. Alzaydi A. Time-optimal, minimum-jerk, and accelera-
LSM least square method
tion continuous looping and stitching trajectory gener-
ation for 5-axis on-the-fly laser drilling. Mech Syst
O  XYZ universal coordinates system
Signal Process 2019; 121: 532–550. O w  Xw Yw Zw Local machining coordinate system
16. Wang H, Wang H, Huang J, et al. Smooth point-to- O f  Xf Yf Zf flamer coordinate system
point trajectory planning for industrial robots with P0 point in amplified binary image
kinematical constraints based on high-order polynomial Pi ðxi ; yi ; zi Þ coordinate of the point to be cut
curve. Mech Mach Theory 2019; 139: 284–293. qi ði ¼ 1;    ; 5Þ joint configuration of robot
17. Phung VM and Tang VL. Bivariate hermite interpola- Q joint configuration
tion on the exponential curve. Period Math Hung 2019; S total displacement of velocity
78: 166–177. solution
18. Shekhtman B. On regularity of generalized hermite u amplification factor
interpolation. Comput Aided Geom Des 2016; 45: VI initial velocity
134–139. VM maximum velocity
19. Aguilar JC. Higher-order Newton-Cotes rules with end
VE end velocity
corrections. Appl Numer Math 2015; 88: 66–77.
20. Sun W and Wu J. Newton-Cotes formulae for the
ai rotation angle from X axis to Xw
numerical evaluation of certain hypersingular integrals. axis
Computing 2005; 75: 297–309. hi angel of welding groove