Hybrid neural network framework for predicting tool tip dynamics via receptance coupling

Tool tip frequency response functions (FRFs) are fundamental to predicting stability lobe diagrams and mitigating chatter in machining operations. This study introduces a hybrid framework that integrates physics-based modeling with data-driven learning to reduce approximation errors in tool holder–tool geometries and mitigate uncertainties in their contact parameters. The tools and tool holders are modeled using a Timoshenko beam-based finite element formulation and assembled as free-free structures via receptance coupling substructure analysis (RCSA). Uncertainties in the elastic modulus, Poisson’s ratio, and density of the tool and holder materials are minimized by aligning the measured and simulated natural frequencies of representative tool and holder samples. Neural network models are pre-trained using simulated FRFs with approximate contact parameters and subsequently fine-tuned through a limited number of experimental free-free impact tests on holder–tool assemblies. The optimized contact parameters are then archived in the database for each holder type. The finite element models of the tools and holders are coupled using the tuned contact parameters and subsequently assembled with the stored spindle model via RCSA. The proposed hybrid approach is experimentally validated through impact testing of diverse holder–tool configurations mounted on machine tools. The resulting methodology contributes to the establishment of a robust digital machine tool database, thereby facilitating more reliable stability predictions and enabling enhanced productivity in NC part programming within CAM systems.