Grey-box model for tool wear monitoring in gear hobbing

Tool wear is a decisive factor in gear hobbing for process reliability, component quality, and cost-effectiveness. Conventional wear detection methods are often based on visual inspections of the tool or indirect monitoring based on workpiece quality at fixed intervals and are therefore either inaccurate or delay the manufacturing process. In this work, a combination of physical models and data-based methods for tool wear monitoring was systematically investigated to enable early and reliable detection of wear conditions. The objective was to determine the current tool condition and enable a prediction of the remaining tool life. For this purpose, various process signals were recorded during the gear hobbing process. The raw data was preprocessed and converted into characteristic features. These characteristic features were then used to develop black-box models that monitor the wear status of the tool. The investigation included both classical approaches like threshold methods and artificial intelligence models such as random forests and neural networks. In addition, the potential of a grey-box model was analyzed. The grey-box model combines the physical fundamentals of the gear hobbing process such as cutting forces with data-based black-box approaches. The use of data-driven models, especially in combination with physical models, has shown that it is possible to distinguish between different degrees of wear with high accuracy. In addition, the use of self-learning algorithms offers the possibility of adapting the system to different tools, materials, and process conditions. The study highlights the potential of data-based approaches for wear monitoring in gear hobbing and provides a solid foundation for further development and industrial application.