Synthetic data and Artificial Intelligence: forming process control using real-time data acquisition and AI trained with simulation datasets; M. Schmiedt, W. Rimkus, S. Feldmann (Hochschule Aalen), J. Lenz, (Hochschule Würzburg-Schweinfurt); 13th European LS-DYNA Conference 2021 Ulm and online, 2021
von Wolfgang, Rimkus
So far, training an Artificial Intelligence (AI) required a high testing effort or access to large datasets, making it unsuitable for small and medium-sized companies. The innovation of the research project “SimKI” is based on training an AI by means of several datasets: a) simulation data from a parameter study, b) live data from forming experiments, c) a combination of both. Therefore, process parameters and the respective quality information of deep drawn components are recorded both inline as well as in real time and transferred to the cloud-based IoT platform “ThingWorx”. The quality information is determined by an optical measurement system (GOM Aramis) and laser triangulation (Gocator line scanner). In order to support small and medium-sized companies with a toolchain applicable to their existing machinery, an older forming press is retrofitted with digital interfaces. The simulation model created contains a thermal material model of the formed blank (LS-DYNA MAT36) and the parameters of the real forming process such as the press speed. The collected data is correlated and evaluated using an Artificial Intelligence algorithm. In contrast to conventional training with scanned components and the corresponding genuine datasets (82.96 % accuracy), the testing results show a comparable validation accuracy of the DNN-based quality assessment by using simulation data alone (83.84 %) or a hybrid dataset (84.38 %). This allows to significantly minimize both the implementation effort as well as the testing effort in production and thus to increase the industrial applicability.