Determination of Transient and Nonlinear Tool Wear Development in Orthogonal Turning of AISI 304 Steel

Sustainable manufacturing requires reliable strategies to extend the service life of cutting tools and reduce the consumption of critical raw materials such as cobalt. When machining difficult-to-cut materials like AISI 304 stainless steel, low thermal conductivity leads to high thermomechanical loads and accelerated tool wear. Conventional white-box models provide insights into chip formation and thermomechanical loads. Such models are capable of predicting linear and continuously progressing tool wear behaviour. However, they have limitations in capturing the highly nonlinear tool wear development under real cutting conditions. To address this gap, this paper presents black-box modelling approaches as a basis for a grey-box model that combines white-box and black-box concepts to predict previously unpredicted nonlinear tool wear. Experimental wear data are analysed using black-box-based image segmentation to derive wear curves throughout the tool’s service life. Additionally, an automated machine stop is implemented on a production lathe to enable ex-situ analysis at critical loads. A Field Programmable Gate Array (FPGA)-based script processes the recorded forces in real-time, while a rule-based detection algorithm evaluates maximum and mean cutting forces to stop the machine tool upon reaching critical loads. Furthermore, a Long Short-Term Memory (LSTM)-autoencoder was implemented as a black-box approach for data-driven anomaly detection. By learning characteristic patterns in the cutting force signal, it enables adaptive decisions without predefined thresholds. Although transient wear could not yet be distinguished more reliably than by rule-based monitoring, the method highlights the potential of black-box modelling to capture complex process dynamics using data-driven approaches.