This study investigates the unsteady wake dynamics of a freely rotating cylinder with a centroid pin and a splitter plate of length & ell;/D is an element of[0.25, 1.0], across Reynolds numbers ( 25 <= Re <= 500). Using an immersed boundary method, the nonlinear interactions between cylinder vortex shedding and plate tip vortices are resolved. The cylinder splitter-plate system exhibits five distinct flow regimes. In regime 1 ( Re<25), the system stabilizes on the centerline with symmetric vortices. Symmetry breaking at Re=25 (regime 2) results in a shift to a stable non-zero mean rotation angle ( theta(M)), enhanced by up to 35% for short plates ( & ell;/D = 0.25). A Hopf bifurcation at Re >= 50 (regime 3) initiates small-amplitude periodic oscillations from vortex shedding. In regime 4 ( 100 <= Re <= 300), ( theta(M)) becomes nearly Re-independent as inertial and wake-induced back-flow forces balance, although oscillation amplitudes vary nonlinearly due to strong wake-structure coupling. For Re>300 (regime 5), multiple shedding frequencies cause a rapid rise in rotation amplitude ( theta(A)) and intensified unsteadiness from secondary vortex interactions. The rotation angle is driven by (i) asymmetric vortex formation on the plate, (ii) turbulent wake fluctuations, and (iii) a feedback mechanism between the wake and the flow separation angle. The birth, growth, and decay of various vortical structures through lifecycle analysis explain non-uniform pressure and velocity distributions, leading to significant drag and lift variations. By linking vortex dynamics, fluid forces, and bifurcation behavior, this study explains how splitter-plate geometry modulates vortex shedding, offering new insights for improved flow-control strategies.
Computer-aided design (CAD) is the digital construction of 2D and 3D objects, and is central to a wide range of engineering and manufacturing applications like automobile and aviation. Despite its importance, CAD modeling remains largely a time-intensive, manual task. Recent works have attempted to automate this process with small transformer-based models and handcrafted CAD sequence representations. However, there has been little effort to leverage the potential of large language models (LLMs) for sequential CAD design. In this work, we introduce a new large-scale dataset of more than 170k CAD models annotated with high-quality, human-like descriptions generated with our pipeline based on GPT-4.1. Using this dataset, we fine-tune powerful code-LLMs to generate CAD sequences represented in a JSON-based format from natural language descriptions, demonstrating the viability and effectiveness of this approach for text-conditioned CAD generation. Because simple metrics often fail to reflect the quality of generated objects, we introduce geometric and topological metrics based on sphericity, mean curvature, and Euler characteristic to provide richer structural insights. Our experiments and ablation studies on both synthetic and human-annotated data demonstrate that CADmium is able to automate CAD design, drastically speeding up the design of new objects. The dataset, code, and fine-tuned models are available online.
We propose an efficient thermodynamics-informed latent space dynamics identification (tLaSDI) framework for the reduced-order modeling of parametric nonlinear dynamical systems. This framework integrates autoencoders for dimensionality reduction with the newly developed parametric GENERIC formalism-informed neural networks (pGFINNs), which enable efficient learning of parametric latent dynamics while preserving key thermodynamic principles, such as free energy conservation and entropy generation, across the parameter space. To further enhance model performance, a physics-informed active learning strategy is incorporated, leveraging a greedy, residual-based error indicator to adaptively sample informative training data, outperforming uniform sampling at equivalent computational cost. Numerical experiments on the Burgers' equation and the 1D/1V Vlasov-Poisson equation demonstrate that the proposed method achieves up to 2,495x speed-up over the full-order numerical baseline with 1-3% relative errors, as well as significant reductions in training (50-90%) and inference (57-61%) cost. Moreover, the learned latent space dynamics reveal the underlying thermodynamic behavior of the system, offering valuable insights into the physical-space dynamics. Code is available at the repository: https://github.com/xiaolong7/pGFINN-tLaSDI.
Large language models (LLMs) employ safety mechanisms to prevent harmful outputs, yet these defenses primarily rely on semantic pattern matching. We show that encoding harmful prompts as coherent mathematical problems – using formalisms such as set theory, formal logic, and quantum mechanics – bypasses these filters at high rates, achieving 46
This work explores the importance of reaction mechanisms and combustion models on the flame length and emission characteristic prediction by computational fluid dynamics (CFD) simulations of a complex multinozzle combustor configuration, operating under CH4/H2 blend variations. For the study, both RANS and LES turbulence models are explored. Test data used for the analysis is taken from work published by KAIST University, on the investigation of combustion dynamics and NOx/CO emissions from lean-premixed multinozzle CH4/H2 blended flames. The combustion domain consists of densely distributed small-scale multitube injectors called Micromixer nozzles. This setup provides insights into the collective behavior of small-scale multinozzle flames and resultant emission rates. Test data for different inlet compositions, keeping a thermal power condition of 78 kW, are considered for evaluation. Results from simulations for OH* chemiluminescence, OH concentrations, NOx, and CO emissions are compared against the test data. Reduce model fuel library (MFL) mechanism with relevant NOx pathways along with flamelet generated manifold (FGM) model found to predict the trend of flame length and emissions concentration with change in fuel composition reasonably well, compared to detailed chemistry combustion model, as well as test data. However, for capturing the impact of local nonunity Lewis number effects, the detailed chemistry model is found to be better for the low turbulent flow conditions, as considered in the referred experimental data.
