Model-based Modules for the Self-optimization of Gas Metal Arc Welding Processes
In the field of automatic gas metal arc welding processes, clamping devices are nowadays often used to prevent weld errors by realizing an exact positioning of the plates. Due to a higher individualization of products, time and cost related intensive changes of the set-up become neccessary when a new product is produced. Nevertheless, disturbances to the process can occur leading to process conditions resulting in an aberrated welding result. To react flexibly to such altering conditions, the welding process was enabled to adapt without external intervention by applying intelligent sensor systems combined with process models. For this purpose, process states and their influence on the welding result are recorded and mathematically modelled. By utilizing these models, new parameter combinations can be determined in case disturbances occur, so that the varying conditions can be compensated and a steady welding result can be guaranteed.
Practical Issues
The research activities of the Welding and Joining Institute (ISF) were to focus on a solution for the challenge mentioned above, when using typical applications with thin sheets with hereby commonly used weld preparations. As process type, a pulsed gas metal arc welding process was used which was more applicable to thin sheets due to its great reach of efficiency. Furthermore, this process type was sufficiently flexible to compensate for changing boundary conditions.
Approach
In this section of the Cluster of Excellence “Integrative Production Technology for High-Wage Countries”, a cross-process concept for a model-based self-optimization of production processes has been developed. The transfer of this concept to the gas metal arc welding process includes an automated optimization to guarantee the required weld quality on two levels. On the micro level this is achieved via monitoring and optimization of the process stability with the objective to influence the specific process parameter of the welding power source. On the macro level the actual welding result is monitored and setting parameters are adapted for potential disturbances like variations in the gap width or in the contact tube distance.
Technical Challenges
As well as the development of a global process model which maps the correlation between setting parameters and welding results adequately, the challenge lies in providing a suitable sensor system to monitor the process. In the case of process stability on the lower, micro level a real-time capable sensor module has been realized which detects the mass transfer and energy input of pulsed gas metal arc welding processes. The module uses transient current and voltage data, and detects therein the point of time of the droplet detachment, because this event combined with actual electrical capacity has a significant influence on the homogeneity and the quality of mass transport. The point in time of the droplet detachment is indicated within the voltage data by a short increase of the voltage. An algorithm was therefore written using a special filter method. At a pulse time of about 10 ms, the evaluation results were available approximately 3 ms before the next pulse started. This short evaluation time enabled the start of a control procedure in the subsequent pulse, if the droplet had not detached. For the optimization on the higher macro level an optical sensor system offering the possibility of monitoring the formation of the weld pool was developed. In order to do this, a CMOS-camera had been set up at the welding torch and recorded up to 500 frames per second in full screen mode with a resolution of 1280 x 1024 pixels. The optics had been built according to the special requirements of gas metal arc welding processes, which meant that an optical filter was equipped that limited the spectral range to near infra-red. Furthermore, a removable protection glass was included to shield the sensor from welding spatter. And to reduce radiation intensity, a customized pinhole aperture was added to shield the sensor from welding spatter because of its small diameter. With the help of the newly provided information about the weld pool in combination with observations from the material transfer and the energy input, the monitoring of the welding process was guaranteed and promptly recognized the influences of disturbances on the welding result, indicating a parameter adaption if necessary. The parameter adaption took place with a model-based optimization, targeting an autonomous choice of parameters to react to disturbances or changing objectives. A solution concept for this task will therefore be the usage of surrogate models for an iterative optimization. And for this optimization, different methods were used which range from classical gradient based methods to metaheuristic search techniques. Taking into consideration the specific requirements for the automated welding production, one of the future tasks will be to qualify methods which are usable for an offline or the time-critical online application.