UdZ 01.21

SPECTRUM – APPLIED RESEARCH 42 / UdZ 2-2021 Figure 1 : Use Case DFA Tulip Input Decibel Input Energy Input RTLS Input Laser cutting process Database Application of data fusion Output of the results Inputs of the start and stop times of the operation C urrent megatrends such as globalization and digitization make great demands on manufacturing companies: Real-time and efficient production planning and control (PPC) as the basis for sound and short-term decision-making is one of them. Necessary conditions for reliable decision-making continue to be high data quality and application-oriented information availability. The central challenge in increasing data quality lies in the investment costs for implementing appropriate measures. Small and medium-sized enterprises (SMEs) are particularly affected by a limited ability to invest. One way to reduce investment costs is to increase the data quality of a dataset by merging data or combining different data sources. The research project ‘DaFuER’ took up exactly this point. The aim was to ensure data quality, especially for production management and production controlling, by applyingdatafusionmethods.Theprojectdevelopedaguideline for the application of data fusion in the context of operational feedback data, which serves as a decision-making aid for user companies and developers of operational application systems. The structure of the guide is based on the model developed in the research project for the application of data fusion in the context of operational feedback data. In a first step, the users define the current use case. They first determine the most important information needs; this is done on the basis of an overview showing the information relevant for PPS. The users select the information that they consider unreliable due to poor data quality. They determine the information availability based on the available data. These are derived from the data sources available to the company. In the second step, the data sources to be merged are determined. For each piece of information identified as relevant, an overview lists potentially related data sources from which this information can be extracted. Users can thus assign concrete, available data sources to their information needs. In addition, they qualitatively determine for each data source to what extent it fulfills the various data quality criteria. Based on these quality characteristics, users are able to select those available data sources for fusion that, complementing each other, fulfill the data quality characteristics. Finally, the appropriate methods of data fusion are identified. Based on a classification of the considered data sources, the users first developed a morphology to describe a data source generically. By combining all possible types of data sources or their morphological characteristics, process-typical errors during fusion can be derived and error classes can be formed. During the project, it was determined towhat extent amethod for applying data fusion is resistant to a corresponding type of error. If it is known which data sources are to be fused, it is possible for the users to derive concrete methods of data fusion based on this assignment. Those can thus best address the key challenges in combining the selected data sources. In order to test the model developed in theory in practical application, a use case was carried out at the Demonstration Factory Aachen (DFA), among other things. The background to this use case was irregularities in the comparison of start and end times of individual operations of a laser cutting machine. The existing data quality is currently inadequate, especially with regard to accuracy, completeness and up-to-dateness.