UdZ 01.21

UdZ 2-2021 / 43 Figure 2: Result of the data fusion (Roll the Dice) Bild 2: Ergebnis der Datenfusion (Roll the Dice) The goal of the practical implementation was to improve the data quality by combining the different information sources using data fusion methods. The information need to be optimized was related to the laser cutting machine operation time. The information availability consisted of the acquired data from a buzzer, an RTLS sensor, an Intelligent Sensor Technology, and a complementary information system. The Intelligent Sensor System included an energy sensor and a noise sensor. The sensors sent the “start”and “stop” signals, respectively, when the energy consumption or noise level exceeded or fell below a certain threshold. The buzzer sent corresponding signals when the employee performed manual actions. The RTLS sensor was able to measure when employees entered or left a predefined work zone. In order to select the best possible data fusion method for the application, it was first necessary to identify potentially occurring defect classes. For this purpose, the morphological characteristics of the data sources were combined. Thus, with regard to the use case, potentially large data volumes, the lack of possibility to apply mathematical operators, and different scale levels, among others, could be identified as errors typical for the process. Based on the analyzed resistances of the data fusion methods to the collected error classes, the method categories decision rules, fuzzy logic, and artificial neural networks were particularly suitable for the fusion of the collected data in the context of this usecase. The corresponding methods were then tested with respect to their applicability. Since, based on the manufacturing concept, the operations on the laser cutting system involved a very large number of different jobs, methods requiring a large data base (e.g. artificial neural networks) were excluded. On this basis, a wide variety of decision rules and a fuzzy logic methodology were implemented in the aforementioned use case. Figure 2 shows an evaluation of the generated data using the data fusionmethod, more precisely the decision rule, “Roll the Dice”. Shown are the start and stop times generated using the aforementioned method for ten different days in the same time period. In addition, the total order duration over the ten days as well as mean values and standard deviations of the start and stop times are shown for comparison purposes. Thus, for the aforementioned decision rule, the calculated total job duration is approximately 8 hours and 17 minutes. Depending on the data fusion method used, different order durations or mean values and standard deviations result. The results of the application of further data fusion methods for this use case can be found on the website of the research project 1 . The research project provided users with two options for the actual implementation of data fusion: one option is to step through the process steps of the data fusion applicationmodel using the application guide. An alternative is the free, interactive online tool created as part of this research project, which guides and instructs users step-by-step through the topic. 1 dafuer-tool.fir.de