Failure Prediction in Robotic Spot-Welding Applications - Challenges in Data Management
Publikation aus Robotics
Titanilla Komenda, Mathias Brandstötter, Jakob Ginera, Tatevik Gharagyozyanc, Andreas Pichler, Werner Liemberger
15th CIRP Conference on Intelligent Computation in Manufacturing Engineering, Procedia CIRP , 7/2021
Even though the automotive industry is a highly optimized production sector, robotic spot-welding processes are still characterized by downtimes caused by tool contamination, cable breaks, communication errors of safety elements or errors due to clamping devices -- to name a few. With the increasing developments in digitization and IIoT integration, a large amount of data can be recorded to identify the causes of the mentioned errors. However, data management is a highly challenging task -- mainly due to the reasons of data quality, volume and harmonization. This paper presents an overview of the different data sources of modern production facilities and challenges in data understanding. The article focuses especially on the challenges for the needed IT infrastructure to harmonize and use different data sources and volumes within a data management pipeline for applying machine-learning models in order to be able to predict production downtimes in robotic spot-welding applications based on available data and data sources.
Keywords: failure prediction, data management, spot-welding, industrial robot, machine learning