Enhancing cell agglomeration with dynamic clustering and dimensionality reduction

High-fidelity combustion simulations require detailed chemical kinetics to accurately capture reactive flow behaviour, but their computational cost remains prohibitive due to the large number of species and stiff chemistry. To address this challenge, this study introduces a novel integration of Principal Component Analysis (PCA) with the Cell Agglomeration (CA) framework through a Dynamic Multi-Zone (DMZ) clustering algorithm. Previous work by Stock et al. (2024) used PCA to construct a compact thermochemical space, relying on a prescribed clustering grid, which however fixes cluster boundaries a priori and limits adaptability to evolving flow conditions. In contrast, the present work removes this reliance on static, pre-defined grids by embedding PCA directly inside the dynamic agglomeration loop, allowing cluster boundaries to emerge and evolve naturally from the instantaneous thermochemical state. The proposed PCA-DMZ framework identifies thermochemical similarities in a reduced-dimensional principal component space and dynamically constructs clusters that minimise chemistry evaluations while preserving predictive fidelity. The methodology is evaluated for two benchmark configurations of the Adelaide Jet in Hot Coflow (AJHC) burner: (i) unsteady Reynolds-Averaged Navier–Stokes (uRANS) simulations of an n -heptane flame with a reduced mechanism (106 species, 1738 reactions), and (ii) Large Eddy Simulations (LES) of a methane-hydrogen flame using GRI3.0 chemistry. Compared with the standard CA approach, the PCA-DMZ formulation yields more compact and effective cluster structures, achieving approximately 20%–30% fewer ODE system integrations at similar accuracy levels, leading to higher overall speed-up primarily due to improved clustering efficiency. It also significantly reduces the need for manual tuning of the CA tolerances, with a ∼ 10% cluster-to-cell ratio repeatedly emerging as the optimal operating point across both uRANS and LES cases. The proposed PCA-DMZ coupling achieves an overall computational speed-up of approximately 7 × and a chemical integration speed-up of about 10 × , while maintaining high accuracy in temperature and major species predictions. • Fully adaptive and largely case-independent clustering achieved via PCA-based CA-DMZ. • Accuracy preserved while solving 20%–30% fewer ODEs through PC-driven clustering. • PC-based clustering yields higher speed-up than standard clustering at similar fidelity. • Robust, self-adjusting clustering delivered by PCA-DMZ with minimal case dependence.