Generalizability of an Identification Approach for Machine Control Signals in Brownfield Production Environments

Abstract

Digital transformation has been a central aspect of optimizing processes in manufacturing companies for several years now. A basic prerequisite of successful transformation is the vertical integration of all machines and machine tools to capture data at all levels. This can create further applications that enable more sustainable and resource-saving processes. At the same time cost- and quality-optimizing analyses such as failure detection, predictive maintenance or general process optimization represent major incentives for companies. While the necessary interfaces are now integrated in state-of-the-art machine tools, companies with older legacy machines face the problem that no such interfaces are readily available. Brownfield machine tools feature outdated technology that does not allow direct networking connectivity without further effort. To participate in the technological progress, a system was developed that allows to extract machine control signals from machine tools and identify them automatically as time series. This is compatible with several communication protocols (e.g., OPC UA) to be as universally applicable as possible. Since machine control signals are often not interpretable for the user due to different naming conventions, the extracted time series are analyzed by machine learning and analytical rule bases, these are based on expert knowledge, and assign a specific signal type in each case. With regard to a cross-machine generalization capability, several aspects have to be considered. Due to different data sources, the identification system must still function reliably with varying sampling frequency. Another challenge is the diversity of different types of machines and production equipment. Therefore, this publication investigates the different influences of data sources and machine types on the machine control signal identification system.