University of Wollongong

Research Online
Faculty of Engineering and Information Sciences -
Faculty of Engineering and Information Sciences
Papers: Part A

2011

Offline programming for short batch robotic

welding
Nathan Larkin
University of Wollongong, nlarkin@uow.edu.au

Aleksandar Milojevic
Kranendonk Production Systems

Zengxi Pan
University of Wollongong, zengxi@uow.edu.au

Joseph Polden
University of Wollongong, jpolden@uow.edu.au

John Norrish
University of Wollongong, johnn@uow.edu.au

Publication Details
Larkin, N., Milojevic, A., Pan, Z., Polden, J. & Norrish, J. (2011). Offline programming for short batch robotic welding. 16th Joining of
Materials ( JOM) conference 2012 (pp. 1-6).

Research Online is the open access institutional repository for the University of Wollongong. For further information contact the UOW Library:
research-pubs@uow.edu.au
Offline programming for short batch robotic welding
Abstract
Robotics based automation has many advantages over both manual operation and dedicated hard automation.
In practice however the robot programming overhead and high cost of dedicated jigging are serious
impediments to its use for short production runs. Offline programming packages offer an attractive alternative
to reduce programming time and robot down time, however many of the available packages are fairly complex
and treat the geometric programming and welding operation separately. The paper will describe the
development of alternative approaches and the evaluation of an integrated offline robotic welding package
(RinasWeld) which addresses these issues and makes low volume robotic welding of complex fabrications
viable.

Keywords
programming, offline, short, robotic, batch, welding

Disciplines
Engineering | Science and Technology Studies

Publication Details
Larkin, N., Milojevic, A., Pan, Z., Polden, J. & Norrish, J. (2011). Offline programming for short batch robotic
welding. 16th Joining of Materials ( JOM) conference 2012 (pp. 1-6).

This conference paper is available at Research Online: http://ro.uow.edu.au/eispapers/551

 Offline Programming for Short Batch Robotic
Welding
Nathan Larkin1, Aleksandar Milojevic2, Zengxi Pan1, Joseph Polden1 John Norrish1
1. University of Wollongong, Wollongong, NSW, 2530, Australia
2. Kranendonk Production Systems, 10 Sir Rowland Hillstraat Tiel, JT, 4004, Holland

Abstract: Robotics based automation has many several OLP software packages is provided. Common
advantages over both manual operation and dedicated packages such as RobotStudio™ from ABB Robotics and
hard automation. In practice however the robot Delmia™ from Dassault Systems are compared, along
programming overhead and high cost of dedicated jigging with a Matlab™ based OLP system. RinasWeld™ from
are serious impediments to its use for short production Kranendonk Production Systems, a state-of-the-art OLP
runs. Offline programming packages offer an attractive software package developed for single piece production in
alternative to reduce programming time and robot down the shipbuilding industry is also dicussed.
time, however many of the available packages are fairly
complex and treat the geometric programming and 2. The OLP process
welding operation separately. The paper will describe the
development of alternative approaches and the evaluation The OLP process can be broken down into several steps:
of an integrated offline robotic welding package 1. 3D computer model generation. Typically
(RinasWeld) which addresses these issues and makes low generated during the design stages of a product.
volume robotic welding of complex fabrications viable. 2. Tag generation, where process start and end
points are identified.
1. Introduction 3. Path planning, where robot motion paths are
planned and the reachability and potential for
Compared to manual welding or automated welding using collision of a robot is assessed.
dedicated machines, robotic welding has many 4. Process planning, where each individual process
advantages; Robotic welding is highly flexible, it is sequenced and optimized.
achieves consistent welding results, and heat input can be 5. Post processing, where the required process I/O
precisely controlled. In practice however, the is added and the program is converted into the
programming overhead and cost of dedicated jigging limit native programming language of the robot.
the viability of a robotic welding system for all but the
most repetitive tasks. In some cases, the programming In addition to these steps, the program will usually require
time can be 360 times the execution time [1]. calibration to account for differences in the real world
geometry of elements compared to the nominal geometry
For industrial welding applications, there are two main of the elements modelled in the software package. OLP
programming methods; online programming, where a packages often offer a simulation system, where programs
teach pendant is used to move the robot to the required can be simulated to verify them before execution on the
position and orientation for each step of the program and physical robot system. A block diagram of the typical
Offline Programming (OLP), where the program is OLP process is shown in Figure 1.
developed in a 3D computer environment.
In most OLP software packages that are currently
The concept of the online programming method is simple, available, a robot programmer completes each of these
however it is requires significant time and effort to steps manually, with perhaps a small amount of computer
generate robot programs for even the simplest processes assistance. For example, tags can be generated along a
and geometries. The OLP process is more complex than complex path by selecting the modelled geometry, rather
online programming methods and typically does not than being generated individually. To allow cost effective
significantly reduce the programming time. The main robot programming for low production runs, these manual
advantages of the OLP method result from freeing the steps need to be automated wherever possible.
work cell for production while programming is taking
place on a computer system. The common OLP packages 2.1 Tag Generation
are Not designed for single or low volume production and
the high cost of proprietary OLP software is normally In robotic welding, the tag generation can be as simple as
only justified for mass production applications. identifying geometrical elements that are in contact and
defining the start and end locations as process start and
This paper describes the issues facing the current end tags. However, since several welding process
generation of OLP software packages for use in variables are linked to geometry, it is convenient for the
programming robots and investigates some of the possible tags to contain extra geometric information. This tag data
solutions available to improve the viability of robotic can then be used at a later stage to automatically generate
welding for low production volumes. An evaluation of optimal process variables. For example, if the plate
 Figure 1: Block diagram of the OLP process.
thickness and material is known, welding speed, arc required limits the viability of this programming method
voltage and wire feed rate for a Gas Metal Arc Welding for low volume part production. Minimising the user
(GMAW) process can be generated. input, or programming time, required by OLP software is
crucial if smaller batch sizes are to be considered. Of the
Tags are also required for weld position calibration. OLP process steps described, the tag generation and
Position calibration is important in robotic welding. A motion planning steps are the most suitable for
small error between the robot path and intended weld automation. The selection of weld process parameters
location can lead to poor weld quality. Typically, robotic during post processing can also be automated, provided
welding systems utilise touch sensing capability built into weld geometry is known. Process planning requires
modern welding power sources, or a structured light 2- detailed knowledge of weld procedures that are not
dimensional profile sensor to correct such positional typically handled by OLP software.
errors.
Delmia™ provides several tools that allow efficient tag
2.2 Path Planning generation of weld seams. However, each seam requires
several operator inputs and can only be programmed
Path planning, involves determining the paths that the individually. Whilst this facility reduces the programming
robot will take to get from each process point to the next. time, it is not ideal for low volume production.
This process takes into account robot reachability, robot RobotStudio™ has similar tools that allow paths to be
configuration, and collision avoidance to generate robot generated from geometrical features in the model. The
position steps that will lead the robot to weld start and end tags generated in these programs do not contain weld
points, any calibration points required, and other locations geometry information, and weld settings need to be
in the cell such as wire cutting or touch cleaning stations. generated manually.

2.3 Process Planning In Delmia™ and RobotStudio™, path planning requires

the manual placement of fly-by points to ensure collision
In welding, the order in which each weld is completed is free paths between processing points. For robots with
important to optimise robot utilisiation while adhering to external axes, Delmia™ offers a tool to allow optimal
weld procedure specifications. In some cases it may be robot placement for reachability of weld seams. Like the
acceptable to simply order welds arbitrarily. However for tag generation tools, this requires additional user input.
preheat or interpass temperature control, and sequencing
with other process activities, it may be necessary to order 3. OLP Solutions for Low Volume Production
welds accordingly.
3.1 Customisation of an OLP package
2.4 Post Processing
Some OLP software packages provide the ability to write
Once the welds are identified and the motions and software scripts to automate functions within the package.
processes are planned, a post processing step is Delmia™ is an example of an OLP software package with
completed. This involves adding the required robot I/O, this functionality. This allows a level of software
such as communication to welding power sources, and customisation to minimise the required user input for each
conversion to the robots native language. With the step of the OLP process. These facilities have been used
exception of selecting weld parameters, the post in the current work to assess the feasibility of reducing
processing step is completed with little user input in most programming time.
OLP software available.
To generate tags for process start and end locations, plates
in the model were checked for intersections. A surface
2.5 Gaps in OLP Software was then generated from the intersecting area, and the
perimeter edges treated as weld locations. An analysis of
The current generation of OLP is capable of completing the direction normal to the plates, and the weld edge
the OLP process defined above. However, the user input direction yielded the orientation of the tags. This process
is shown in Figure 2Error! Reference source not
found..

Path planning was treated as two separate steps. The first

step was to determine if the process path is reachable by
the robot, in order to select an appropriate robot
configuration and, the position of external axes. The
second step was to plan the motion path between each of
the process paths.

The first step to check reachability, configuration, and

select the position of external axes, was completed by
generating small robot programs that executed the weld
path, as shown in Figure 3. During execution of the
program, Delmia™ is able to report the reachability of the
weld path and if the robot is in collision. Based on this
output, the robot configuration and external axis position
is modified until suitable positional parameters are found.
On investigation it was found to be too difficult to
implement this step with the scripting functions available
in Delmia™. However, a third party software toolbox, the
Kineo™ Robotic Path Planner, was found for Delmia™
that was able to generate the paths required.

Although this investigation indicated the potential to

reduce programming time by customising the Delmia™
OLP software package, many limitations were
encountered. In particular, the processing speed in
Delmia™ was deemed to be too slow, possibly due to the
generation of simulation graphics during program
execution. Various software ‘bugs’ such as the robot Figure 3: Delmia assessing the rechability of a weld seam.
‘disappearing’, also led to incorrect and inconsistent
results. Delmia™ is a powerful OLP software package, 3.2 Development of an Automated OLP System
however, at this stage it was not considered suitable for
the envisaged low volume production. A key requirement of an OLP software solution for low
volume production is to limit the amount of user input
required. Typical OLP software relies on a 3-dimensional
computer modelled environment for user input and
feedback. In fact, such an interface would not be
necessary for low volume production and a simplistic
OLP system can be developed without the computational
overhead that is required to generate the visual
environment. As automated OLP is calculation intensive,
the Matlab™ software package was chosen as a suitable
platform for an alternative system.

The automated OLP system has two key low level

components on which the higher level algorithms are built
upon. These are; the robot kinematic model which is used
to calculate the various joint angles for a given robot
position and orientation; and, the collision model which
determines if the robot components are in collision with
the other workcell components. Calculation time is an
Figure 2: Plate Analysis for Tag Generation in Delmia. important consideration for an automated OLP system,
particularly for the collision model. To check a weld path,
the collision model can be run up to ten thousand times
for a simple structure or more than one million times, for
a complex structure. To accommodate this, the robots in
the collision model were represented by a collection of
spheres in order to minimise the complexity of the
collision calculation as shown in Error! Reference
source not found..
 compares to an estimated time of 60 minutes using
conventional online programming techniques. However, it
should be noted that this system was designed for a
specific robotic welding application. Further development
is required to make it suitable for programming a generic
robotic welding system.

3.3 RinasWeld

RinasWeld™, developed by Kranendonk Productions

Systems (KPS), is an OLP software package developed
originally for the shipbuilding industry. It is able to
autonomously generate robot welding programs directly
from CAD information with only a small amount of user
input. The software identifies weld seams and generates
weld tags and touch sense calibration points as required.
It also plans collision free motion paths, and selects
optimal robot placement within the external axes.

Downhand, vertical and overhead welds are identified and

can be handled differently, along with intricate weld path
Figure 4: Sphere Model of an ABB IRB4400 Robot. details such as those required around plate edges.
RinasWeld also links weld geometry with weld settings,
Weld seams were identified in the Matlab™ system by further streamlining the OLP process.
searching for plate edges that align with other plate
surfaces. The start and end locations of the intersection As RinasWeld has been developed specifically for
edges along with the normal direction to the surface were industrial applications, it differs from existing OLP
used for tag placement and orientation. Once the weld software. Robot type(s), position, configuration of
tags were generated calibration points were added along external axis, and tooling are all ‘hard coded’ into a
the weld direction, and in the case of a corner, on other specific version of the software supplied for the end user.
required geometries. Although RinasWeld has some software limitations
around workcell customisation, the OLP functionality is
Linear paths between the weld and calibration points were well developed, and is the only commercially available
then checked, and optimal robot configuration and software currently suitable for generating robot programs
external axes position selected. Points were also added specifically for low volume production. Kranendonk
using a probabilistic Roadmap (PRM) planner for robot claims that each hour of software use can generate 15
paths that are difficult to access. Error! Reference hours of robotic weld programs [3].
source not found. shows a robotic path planned using the
PRM planner.

Figure 5: Path Planned using a PRM

Figure 6: Simulation of a weld Program in RinasWeld.
After the robot motions are planned, the system converts
4. Application Examples
the generated paths into code that can be executed by the
robot controller. I/O is also added for the ancillary
4.1 Welding Research
functions that are required by welding, torch cleaning and
calibration systems.
Welding research requires the precise control of weld
parameters and consistent, repeatable welds to ensure
This system is very effective at producing robotic welding
optimal results. Although this can be obtained using hard
programs. It takes approximately 1 minute of user time to
automation, the flexibility offered by a robotic
produce a weld seam for a complex structure, this
manipulator significantly improves welding capability. automatically split the construction into smaller segments
Although programmed weld paths can be repeated several for further efficiency.
times, weld lengths are short compared to industry and
programming time is a significant overhead. Also, Along with the identification of and generation of weld
conventional robot programming techniques limit the start and end points, joint properties are also recorded.
utilisation of the facility to only those with robot User configurable rules then allow certain welding
programming skills. parameters to be defined. For example the bead size of the
weld can be set according to plate thickness, weld
Automated OLP software reduces the programming time orientation and other variables. Once rules are set for one
to a few minutes per weld path and unlike online product, they can be reused directly on other products. In
programming methods complex geometries and difficult this way welding “knowledge” is accumulated in the
to navigate obstacles do not add significant programming software system allowing the automatic OLP process to
time. Figure 7 shows a simulation of a robot navigating be applied on variations of the original product. The
around a clamped plate for welding. OLP software, such RinasWeld™ system also has high level weld
as RinasWeld™ has now been applied to many welding identification and manipulation features. In many cases,
research tasks at the University of Wollongong. It has welds are identified which do not require robotic welding.
been applied to fabrication of test samples for For this purpose “assembly level” rules can be setup to
metallurgical analysis, temperature profile measurements identify which welds do not require robotic welding.
on welded plates and weldability studies on coated steels. Furthermore, manual tools allow the user to directly
It allows researchers not skilled in robot programming to delete, cut, or trim individual welds. This eliminates the
create usable robot programs. need to manipulate CAD drawings specifically for the
OLP process, minimising the changes to current drafting
practices.

In RinasWeld™ the work preparation can be carried out,

and results saved, for multiple products and the robot
program generation can be initiated as a batch process for
a number of these products. This allows the user to
prepare the work during normal working hours, and then
let the computer perform the unattended robot program
generation outside of working hours.

In shipbuilding, the products to be welded are defined

many weeks and even months in advance. This allows
preparation of robot programs for welding these products
to happen well in advance of production, thus securing
efficient production with high utilisation of the welding
Figure 7 Robot Path Navigating Around Clamp Obstacle robots.
To create a program, the parts to be welded are first 4.3 Vehicle Hull
clamped into position using a flexible fixturing system.
Once the parts and in position, the fixturing and parts are Although heavy vehicle production is not typically low
modelled in a CAD software package. Each fixture comes volume, automated OLP techniques can be applied to
from a standard library of parts, so this is a simple reduce programming time for product changes. An
process. After the work environment has been modelled, example of this is the production of a defence vehicle
it is imported into the automated OLP software package, hull, where variant changes and customer specific options
where the robot program is generated. are regularly required. This workcell also has other
unique challenges; the hull is tacked together and internal
4.2 Shipbuilding weld seams are only accessible through openings in the
hull; a robot-on-robot manipulator with 12 degrees of
. The large number of long and complex welds required in freedom is utilised to access the weld locations.
the shipbuilding industry are ideally suited to robotic
welding, however the small production volumes limit the
viability of robotic production using conventional
programming techniques. RinasWeld™ is an automated
OLP package which was originally developed for
shipbuilding applications..

Typical applications for RinasWeld™ in shipbuilding are

the welding of webs, stiffeners panels through to double
hull blocks. It supports various CAD formats to allow
product models to be directly imported. As ships can be
very large structures, the import function can
In this case, the OLP software has several tasks:

1. Identify the optimal position of the large 7. References

carrying robot so the small robot can reach the
weld path without collision. [1] Joseph Polden, Zengxi Pan, Nathan Larkin, Stephen
2. Adjust the torch orientation to avoid collision, Van Duin, John Norrish, “Offline programming using
for example into a corner. DELMIA Automation”, 2010 International Conference on
3. Determine the configuration of the smaller Robotic Welding, Intelligence and Automation
welding robot for the weld path along with the (RWIA'2010), Oct. 14-16, 2010, Shanghai, P. R. China
calibration points.
[2] Zengxi Pan, Joseph Polden, Nathan Larkin, Stephen
4. Plan the motion path for the small robot to each
Van Duin, John Norrish, “Recent Progress on
of the required locations.
Programming Methods for Industrial Robots”,
5. Plan the motion path for the carrying robot to
ISR/ROBOTIK, Munich, Germany, June 7-9, 2010.
drive get the welding robot inside the tacked
hull. [3] Anonymous. (2009). "Panel & Web Welding Gantry."
6. Generate the robot code required to complete the Retrieved 15/12/2009, 2009, from www.kranendonk.com.
weld.
Each of these tasks was completed using the Matlab based
automated OLP software package described above.

Using this system, welds can be programmed in a few

minutes compared to the several hours that it takes using
online programming methods. Programs for new weld
seams are also easily implemented into the existing code
structure where small modifications have taken place.

5. Conclusion

Utilising robotics for welding has some clear advantages.

However, for short batch runs, conventional programming
techniques make robotic welding an unviable option. One
solution is the automation of the OLP process, such that
robotic weld programs can be generated directly from a
CAD model with minimal human input.

Three approaches to automating the OLP have been

presented. Adding automation scripts to an existing OLP
software package, developing a Matlab™ based
automated OLP software system, and using a
commercially available automated OLP system. Although
there were problems automating an existing OLP software
package, the Matlab™ based system and the commercial
system were able to generate robot programs cost
effectively for low production volume.

The application examples show how these software tools

are being used for research, in shipbuilding, and in
defence vehicle fabrication. These examples demonstrate
the effectiveness of automated OLP techniques for the
robotic welding process.

6. Acknowledgement

The authors acknowledge the support of the Defence

Materials Technology Centre, which was established and
is supported by the Australian Government’s Defence
Future Capability Technology Centre (DFCTC) initiative.
In addition thanks are due to DMTC partners Thales and
RTA. Assistance in providing training and an evaluation
licence for RinasWeld™ from Kranendonk is also
acknowledged.