Spline-based methods for fluid-structure-contact interaction in orthogonal cutting

Abstract

In high-performance machining, process efficiency and product quality are dependent on tribological interactions at the tool-chip interface, known as the secondary shear zone. While cooling lubricants are employed to mitigate the extreme thermo-mechanical loads in this region, the precise physical mechanisms governing their behavior remain obscured from direct experimental observation. Furthermore, current industrial simulation tools rely on empirical, “black-box” friction models that lack the physical fidelity required for predictive process optimization. This dissertation addresses this fundamental gap by developing a novel, high-fidelity numerical framework for modeling the Fluid-Structure-Contact Inter-action at the microscopic level and exploring approaches for reduced order modeling in this context.This work begins by identifying the specific modeling requirements through experimental investigation. Orthogonal cutting tests reveal that while lubrication significantly reduces loads at low speeds, this effect diminishes at high speeds, highlighting the limitations of current experimental and numerical capabilities. To resolve this, a spline-based numerical framework is proposed and developed. This framework utilizes Isogeometric Analysis for the thermo-elasto-plastic solid and standard Finite Elements for the incompressible fluid, coupled via a partitioned Robin-Neumann scheme to robustly simulate the interaction, including the fully enclosed fluid pockets.The fully coupled, high-fidelity model is employed to investigate the governing tribolog-ical mechanisms at the contact interface. Specifically, this investigation aims to verify the hypothesis that friction in the secondary shear zone is dominated by the mechanical inter-locking of microscopic surface topographies. However, comparisons with experimental data reveal fundamental discrepancies regarding the influence of contact pressure and relative velocity. This outcome challenges the initial assumption of mechanical dominance, demon-strating that mechanical interaction alone is insufficient to explain the observed behavior and indicating that there are other physical mechanisms that play significant roles.Finally, to bridge the gap between high-fidelity simulation and practical engineering application, this work develops two reduced model strategies. First, a scalar surrogate model based on Gaussian Process Regression is implemented, achieving a computational speed-up of over 107 for multi-scale coupling. Second, a methodology for generating full-field reduced models is established. Leveraging deep learning architectures developed in a parallel project, it is demonstrated that Deep Neural Networks can effectively learn compact latent-space representations of continuous functions defined in a spatial domain, offering a promising direction for future real-time, full-field “digital twins”.