Autonomous Manufacture of Large Steel Fabrications EC Contract: G1RD-CT-2000-00461 'NOMAD'
Keith Herman*, Nicholas Spong ** and Ari Lylynoja***
* Caterpillar Belgium S.A.
** TWI Ltd, Cambridge, UK
*** Delfoi OY, Espoo, Finland
Proceedings of 35th International Symposium on Robotics, 23-26 March 2004, Paris
Introduction
There are many influences affecting the ability of European large-scale fabricators from remaining competitive. Customer preference is shifting from accepting mass-produced 'standard' products to a desire for customised products that more effectively meet their specific needs. There is also a drive to achieve shorter lead-times in supplying custom products.
Increased price competition from fabrication companies in low cost-base geographical areas around the world is also putting pressure on European fabrication companies. A reduction in the number of skilled and trained personnel, particularly welders in the West, is also a considerable challenge.
Traditional European manufacturing companies struggle to respond to the demands imposed by these trends for several reasons. Typically, large-scale fabrications, such as earthmoving equipment and pedestrian bridges, are produced using either manual welding or dedicated automatic (robotic) welding processes. Manual welding is highly flexible in terms of adaptation to size and shape but is characterised by high cost (labour), low production rate, and variable quality. Dedicated automated processes are capable of producing high and consistent quality with high production rates but lack the flexibility required for large structures produced in lower volumes.
Among the constraints preventing automated welding from replacing manual welding in this type of fabrication is the need for complex, dedicated, and expensive fixturing. Not only is it expensive, but it also adds to the lead time for fabricating a part. It is also quite difficult and expensive to purchase a large robotic cell capable of reaching all welds on large structures. Extensive time requirements for robot programming time are also preventative. Although off-line programming saves time on the factory floor, it is still labour intensive and considered too expensive and time consuming for low volume production.
This paper describes progress within the NOMAD project to develop an autonomous, mobile welding robot capable of fabricating large-scale customised structures.
The NOMAD project, which is funded under the European Commission's Framework V program, started in March 2001 and is due to finish in August 2004. The project consortium consists of Caterpillar Belgium SA (Coordinator), TWI Ltd (UK), ESAB AB (Sweden), Delfoi OY (Finland), Reis Robotics (Germany), Fraunhofer Institute for Factory Operation and Automation (Germany), Robosoft SA (France) and Nusteel Ltd (UK).
Approach
The goal of the project is to create a fabrication system that is capable of fabricating small batches of products and even single, unique 'one-off' structures as easily and quickly as large multiples. This is being achieved by eliminating as many of the current constraints as possible. For example, no fixtures will be used and the robot will be brought to the part that has been placed in a work cell. Robot arm programming will be automated. An image of the overall concept can be seen in Figure 1.
In order to achieve this level of automation in a timely and cost effective manner, it was necessary to develop and integrate several key technologies. These technologies include:
- Real-time manufacturing simulation for automated process planning, robot programming and system monitoring.
- Vision based sensor systems that will identify the product's location and orientation within the cell.
- A specially constructed industrially rugged Robot Transport Vehicle (RTV) to enable autonomous robot navigation for high accuracy positioning of a 6-axis robot arm with the attributes required for the welding tasks.
- Use and optimisation of local sensing and guidance systems to compensate for small errors of position and/or work piece movements during welding.
- Electronically available welding data, which gives information about the welds on the structure to be welded as well as a library of welding procedures to be used by the manufacturing simulation system for programming of all-position welding.
The simulation system is based on Delmia's IGRIP, which has been enhanced with many new routines and algorithms required by the NOMAD project. The added functionality has allowed it to function as the centre of operation of the NOMAD system. The user can 'drive' the system from the simulation PC. This system configured to communicate with a product model library, the vision sensing system, a library of welding procedures, the 6-axis robot controller, and theRTV in an automated fashion in order to create programs to weld a part. A screenshot of the simulation system can be seen in Figure 2.
The system starts by retrieving the CAD model of the part to be welded. The use of electronic CAD data is nothing new and is essential to automated processing. The enhancement made by the NOMAD project is the addition of electronic information about the welds needed for the simulation system to specify the requirements for automated weld programming.
After the system has the correct product model, it requests information on the location and orientation of the product in the actual workcell. When the vision system returns this information, the system can calibrate itself and start programming the route of the RTV and the motion for the robotic arm.
By using the electronic weld specification, the part orientation information from the vision system, and the database of welding procedures, the system can also assign welding parameters to the robot motion programs.
When the programming phase is complete, the system communicates with the RTV giving it information to drive to the correct location to make a weld. Constant checking and updating of the RTV position is performed by the vision system to dynamically modify the RTV navigation path. Once the RTV is in place, the simulation system automatically generates the robot arm program, which allows the 6-axis robot arm to carry out the welding tasks.
It is the intention that the robot arm will use a combination of 'touch sensing', through-the-arc seam tracking, and a laser vision camera. This will compensate for the positional errors created by inaccuracies in the vision system and RTV positioning so that precise welding can be carried out.
Results
The project is not due to reach completion until August 2004; however, the results so far are extremely encouraging. Most of the work to date has focused on the development of the individual pieces of technology needed for the overall system. Although the project is in the early stages of the integration of the different technologies, the capability of the system to recognise the location and orientation of a particular component and then drive the RTV and robot arm into position for welding has already been demonstrated. Figure 3 shows this being executed during a demonstration. Not unexpectedly, however, a number of challenges have arisen along the way.
Within the project scope, it was found that welding data could not be attached at the CAD model stage and then directly imported into the simulation system. In order to continue with the development of automated robot programming, a 'welding data file' was created that includes the welding information needed for the simulation system to perform the programming tasks. This includes information on the weld locations, joint preparations, and final weld dimensions. Itis believed that future developments outside this project could allow this data to flow directly from a CAD system to a manufacturing simulation system.
The automation of robot programming for welding tasks has proven to be more challenging than initially expected when considering multi-pass welding. In relative terms, single pass welding is a much easier task to automate. The difficulty associated with multi-pass welding is the application of many different welding layers when they are not all made with one 'stop' of the RTV. To reduce the effects of distortion, it may be necessary to distribute the heat associated with the welding process by only partially welding in one location before moving to another. The fact that the base of the robot arm is not fixed makes it impossible for the robot controller to 'memorise' weld locations so that it can return to finish subsequent welding runs. The approach being taken for making multi-pass welds will utilise a combination of capabilities of the simulation system, the robot controller, and sensing systems attached to the robot arm.
Although the majority of the robot arm programming is to be done automatically, the final system will involve the use of a 'process verification phase' and an 'execution phase' in the simulation system. In the process verification phase, the user will interact with the software to finalise the programming for each weld. The reason that the process verification stage was added is because there are certain decisions that can easily be made by a human that would require considerable development to automate when dealing with structures that have large variations in shape. Decisions that would be made by the user include choosing the touch sensing search points and the type of weld joint tracking that the robot will use. These tasks are quite easy for a human to perform, but automating them proved to be too complex to achieve within the project.
The vision system, which uses information from four cameras to determine the location of parts and the RTV in the workcell, has been tested and evaluated for accuracy. It was determined that the positional error of the system is less than 2mm, which is sufficient for placement of a robotic welding arm when additional local sensing is used by the robot arm. Figure 4 shows the vision system comparing CAD information to image information for position calculation purposes.
Testing of the RTV motion shows that it has the sufficient manoeuvrability to navigate within the workcell and avoid obstacles. During movement, the RTV uses odometry to navigate to the requested position. It was found that use of odometry alone would not provide the required accuracy to place the robot for welding. However, when coupled with the vision system for verification of its location, the RTV is able to achieve sufficient positional accuracy to allow robot welding if the robot arm utilises local sensing (touch-sensing or laser vision).
A new all-positional welding electrode has been developed for use in this project. This metal-cored electrode has been designed to produce very low levels of slag to avoid problems of slag build-up while welding multi-pass butt joints of at least 35mm in thickness. An extensive library of welding procedures has been developed using this electrode that will be used by the simulation system for programming the robot arm for welding.
Conclusion
The project is progressing well and the component parts of the system have now been moved to Caterpillar Belgium to be installed in a dedicated work cell. It is anticipated that the primary technical objective of creating a facility to demonstrate an autonomous robot welding system will be achieved within the timescales of the project.