In today’s global markets, companies must ensure product quality, transparency, and responsiveness to remain competitive. Efficient and precise milling processes are now more crucial than ever due to the utilization of advanced manufacturing technologies. Deviations of nominal geometry occurring from the milling process can result in time consuming and costly reworking or scrapping. Predicting these deviations before machining has a significant value for optimizing the workflow. Data-based methods have demonstrated their ability to surpass the limitations of classical analysis methods in optimizing machining processes in production. However, their performance depends on an extensive and diverse structured data set. In this paper, we propose a method for fusing all the necessary machine signals collected during the milling process into a large-scale data set suitable for machine learning in milling. This is achieved by recognizing patterns and reference points in various types of sensor systems to transform the signals into a unified data model. Specifically, our data set includes the milling program, internal data of the tool machine, data collected by external accelerometers and spindle sensors, as well as CAD data and 3D scans of the finalized part. Together, this serves as a digital end-to-end representation of the milling process, providing a foundation for machine learning based methods. We apply and evaluate our approach on a test geometry.