|
6251 |
|
Thursday, February 13, 2020, Raum Hesse 10:00 AM Aus- und Weiterbildung |
|
Automatische Bahnplanung für Schweißroboter: Stand der Technik und neue Lösungsansätze zur Offline-Programmierung |
|
Akhil S Anand / Fraunhofer IPA, Germany Gauthier Hentz* / Fraunhofer IPA, Deutschland Christian Landgraf/ Fraunhofer IPA, Deutschland Johannes Stoll/ Fraunhofer IPA, Deutschland |
|
Robotic welding is well established in large-scale enterprises whereby programming of the process follows a manual online teach-in method. This method implies immobilization of the robot cell during programming of a new product batch or adaptation of the program to products with geometric deviations. Small and medium-sized enterprises (SMEs) have to deal with much smaller lot sizes and high product variety leading to undesirably high programming time. Alternatively, using an offline programming (OLP) environment allows to improve productivity as the robot cell is uninterrupted during programming. Still, the current industrial OLP solutions demands high programming time because of the use of virtual teach-in approach. These OLP solutions further demand online adaptation of the path in a so-called after-teach-in process. Fraunhofer-IPA develops innovative OLP components to address both low programming time and after-teach-in requirements. We recently developed a framework by integrating those components and successfully tested it in robotic cells at two SMEs. Firstly, a 3D laser-sensor with cognitive software components allow precise localization and measurement of assembly deviations of the real product. The virtual environment is adapted to match the real workpiece. Secondly, cognitive functions based on a description of the process parameters allow to automatically suggest process features to the programmer on the product, as e.g. welding seams. Offline programming is further facilitated by an automatic path optimization approach using RRT* on a sampling space defined by the process constraints. It generates an optimal collision-free path considering various constraints such as, reachability, singularity, safety and other process constraints. In many robotic-welding use-cases, optimal collision-free path-planning in a cluttered environment is a central issue. Despite of major efforts to automate welding robots as in the case of all types of industrial robots, automatic path planning is still not solved completely. Motion planning is a fundamental research area of robotics with applications in mobile robots, robot manipulation, etc. Among various methods emerged in past 5 decades, sampling-based methods are currently the state-of-the-art for path planning in high-dimensional spaces. Sampling-based path planning methods have proven to be effective and robust in many test use- cases. However, it demands very high computation effort for complex seams and collision geometry requiring more than three DOFs. This makes it impractical in more advanced industrial scenarios. Therefore, in this paper we aim at extending our current OLP framework by exploring novel machine learning and algorithmic based approaches. In this work, we first present a comprehensive review of existing path planning methods for welding robot with focus on the state-of-the-art sampling based path planner developed at Fraunhofer-IPA. Then we explore future directions to increase efficiency and robustness of the path planner in complex welding scenarios making use of algorithmic as well as machine learning approaches. |