Aerodynamic mechanisms and artificial neural network modeling of tandem cylindrical biomass particles with high aspect ratios

Against the backdrop of the “dual-carbon” strategy and the ongoing energy transition, biomass co-firing in coal-fired boilers has gained wide attention due to its combined advantages of emission-reduction potential and engineering feasibility. However, the aerodynamic behavior of typical high-aspect-ratio straw particles during pre-furnace transport and mixing remains insufficiently understood, particularly the influence of particle–particle interactions on multidimensional force characteristics. This study examines a tandem configuration of two finite-length cylindrical particles and, for the first time, systematically incorporates variations in incidence angle (0° ≤ θ ≤ 90°) into particle-resolved direct numerical simulations (PR-DNS). Across a broad Reynolds number range (10 ≤ Re ≤ 2000) and multiple spacings (2 ≤ L/Deq ≤ 8), we comprehensively analyze the evolution of the drag (CD), lift (CL), and torque (CT) coefficients for Front and Back. The results reveal a characteristic “shielding–recovery” behavior of Back under different orientations and spacings and elucidate the mechanism by which increasing Reynolds number triggers three-dimensional wake instabilities and shifts force peaks. Furthermore, a lightweight surrogate model based on artificial neural networks (ANN) is developed, which demonstrates excellent robustness across the full parameter space. By training on PR-DNS datasets, the ANN achieves rapid predictions of CD, CL, and CT with deviations typically below 3%, even under highly nonlinear flow conditions where conventional correlations fail. Sensitivity analysis shows that ANN captures the coupled effects of Reynolds number, incidence angle, and spacing with superior accuracy, highlighting its capability to generalize beyond the sampled conditions. This work not only fills a gap in aerodynamic databases for non-spherical particles in tandem configurations but also proposes a hybrid modeling approach that couples physical mechanisms with data-driven learning. The combined insights from PR-DNS and ANN provide a scalable force-closure strategy for Eulerian–Lagrangian simulations of multiphase flows, enabling both physical interpretability and computational efficiency. The findings offer practical guidance for improving co-firing efficiency and emission reduction in coal-fired boilers.