Datenpaket: A Multimodal Dataset for Process Monitoring and Anomaly Detection in Industrial CNC Milling

Der vorliegende Datensatz enthält etwa sechs Stunden Prozessdaten, die bei Fräsbearbeitungen auf einer DMC 60 H von Deckel Maho aufgezeichnet wurden. Um die Vielfalt und Variabilität realer Fräsbearbeitungen widerzuspiegeln, wurden branchenrelevante Bauteile, Werkzeuge und Bearbeitungsstrategien verwendet. Insgesamt sind 33 Fräsversuche enthalten, die drei Bauteiltypen und acht Anomalietypen abdecken. Der Datensatz umfasst NC-Programme, CAD-Modelle und eine detaillierte Dokumentation, um eine vollständige Reproduzierbarkeit und eine Anwendung als Benchmark zu ermöglichen. Während der Bearbeitung wurden die Signale der SINUMERIK 840D Steuerung über die SINUMERIK Edge App „Analyse MyWorkpiece/Capture” erfasst. Diese ermöglicht einen Datenexport mit 500 Hz im JSON-Format (.json). Zusätzlich werden die von externen Sensoren aufgezeichneten Kraft- und Beschleunigungsdaten im MATLAB-Dateiformat (.mat) bereitgestellt. Jeder Versuch umfasst zudem eine MATLAB-Datei, in der alle aufgezeichneten Signale synchron abgespeichert sind.

This dataset contains process data from around six hours of milling operations performed on a three-axis horizontal milling machine (DMC 60 H, Deckel Maho). To reflect the diversity and variability of real-world milling scenarios, components, tools and machining strategies relevant to the industry were used. A total of 33 milling experiments are included, covering three component types and eight anomaly types. To support full reproducibility and benchmarking applications, the dataset includes NC programs, CAD models and detailed documentation. During processing, signals were acquired from the Siemens SINUMERIK 840D CNC system via the SINUMERIK Edge app 'Analyse MyWorkpiece/Capture', which enables the export of high-frequency data at 500 Hz in JSON format (.json). Additionally, force and acceleration data recorded by external sensors is provided in MATLAB format (.mat). Each experiment also includes a combined MATLAB file in which all recorded signals are synchronized.

This dataset contains process data from around six hours of milling operations performed on a three-axis horizontal milling machine (DMC 60 H, Deckel Maho). The data was collected in a laboratory setting using industry relevant components, tools and machining strategies in order to reflect the diversity and variability of practical milling scenarios. Controller-side process signals were acquired from the Siemens SINUMERIK 840D CNC system via the SINUMERIK Edge app "Analyze MyWorkpiece/Capture", which enables high-frequency data export at 500 Hz. Additional force measurements were recorded using a force measurement platform (Kistler Type 9255C), and accelerations were captured using a sensor mounted on the main spindle (PCB Type 356A33). The force platform signals were amplified using a charge amplifier (Kistler Type 5015A1000 K) before being acquired at a sampling rate of 10 kHz useing a data acquisition card (Data Translation DT9836; referred to as DAC). The acceleration sensor was connected to the same DAC via a Kistler coupler (type 5122). Raw data from the force and acceleration sensors is provided in the form of MATLAB timetable files (.mat). The process signals from the Siemens SINUMERIK Edge are stored in JSON format (.json). Preprocessed Edge data is also available as structured CSV files:

• hfdata.csv: high-frequency controller signals (e.g., position, current, torque) for the X, Y and Z axes as well as the main spindle
• hfblockevent.csv: executed NC code extracted from the machine log
• header.csv: recording metadata In addition, each experiment includes a consolidated MATLAB file (.mat), in which all available Edge and DAC signals are synchronized and stored as a single timetable object. To achieve this, the 500 Hz Edge signals were interpolated using the PCHIP (Piecewise Cubic Hermite Interpolating Polynomial) method to match the 10 kHz resolution of the sensor data. This unified file allows signals to be compared directly across all sources and simplifies further analysis. A MATLAB figure (Sync_Plot.fig) is also included to visualize and confirm the successful synchronization between the sensor and Edge signals. A total of, 33 milling experiments are included, covering three component types and eight anomaly types. NC programs, 3D CAD models (.stp), and detailed documentation are provided to support full reproducibility and benchmarking applications.

Documents:

• Dataset: All data files organized by component type and trial ID
• Descriptive:
  • Design of Experiment (process parameters and configuration details for each experimental trial)
  • Cutting Tool information
  • Part description: NC programs executed during the experiments (G-code) and 3D CAD models (.stp), organised by component type Experimental Setup:
• Machine: DMC 60 H (retrofitted 3-axis horizontal milling center)
• Materials: S235JR (unalloyed structural steel), Al2007 T4 (aluminum alloy), POM-C (polyoxymethylene copolymer)
• Cutting Tools:
  • HSS-Co8 end mill (TiAlN coating, Ø 20 mm, k10 tolerance)
  • Solid carbide end mill (TiSi coating, Ø 10 mm, f8 tolerance)
  • Solid carbide end mill (uncoated, Ø 5 mm, e8 tolerance)
• Workpiece blank dimensions: 150x75x50mm