Xiaohang Fang

In this study, two-dimensional simulations are performed to investigate hydrogen–air detonation interactions with a semi-cylindrical obstacle. A comprehensive regime map covering propagation, reinitiation, and quenching behaviors is established by varying the blockage ratio (BR) and channel height (H). The results show that while the classical gap-size to cell-size ratio (d/λ) criterion provides a useful first-order classification of detonation regimes, but it fails to capture the highly sensitive and non-monotonic behavior observed near critical conditions. Near the propagation/reinitiation boundary, boundary-induced shock attenuation is shown to weaken the leading shock. Given the Arrhenius dependence of the reaction rate on post-shock temperature and pressure, this increases the local induction length and produces transient shock–flame decoupling. Subsequent transverse wave interactions recompress the mixture and recouple the shock and reaction front, enabling anomalous detonation survival. Near the reinitiation/quenching boundary, an anomalous quenching “dead zone” is identified, where neither bottom-boundary nor top-boundary reflection mechanisms provide sufficient compression to sustain reinitiation, resulting in expansion-dominated attenuation and global failure. The channel height is further shown to alter the near-limit dynamics. For identical gap sizes, larger channels can reduce curvature-induced expansion and enhance upstream pre-compression, thereby suppressing shock and reaction zone decoupling and broadening the reinitiation regime. Thus, the present results demonstrate that near-limit detonation behavior is governed by a delicate interplay between local geometric effects and multidimensional transverse wave dynamics, and cannot be fully characterized using a single global scaling parameter.

Interpreting complex flows, which may include transient features, non-linearity, and high dimensionality, is challenging because averages may not represent any individual flow field. This work evaluates the variability of dimensionally-reduced flow field data measured using Particle Image Velocimetry (PIV) compared to an ensemble mean, using several novel vector comparison metrics, namely; Weighted Relevance Index (WRI), Weighted Magnitude Index (WMI), and a modified Combined Magnitude and Relevance Index (modified-CMRI). Three dimensionality-reduction techniques were assessed, Proper Orthogonal Decomposition (POD), Dynamic Mode Decomposition (DMD) and Sparsity-Promoting DMD (SPDMD) using the proposed metrics. The PIV data were collected using an optically accessible single-cylinder engine under motored conditions at two crank-angle degrees (CAD), 460 CAD (maximum inlet valve lift) and 700 CAD (a typical spark-timing angle). The results show that the 0 Hz SPDMD mode preserves the vector magnitude and alignment and thereby better captures intake jet dynamics, aligning more closely with individual snapshots than ensemble averaging. Furthermore, the metrics consistently identify discrepant snapshots in both datasets when compared to the 0 Hz SPDMD mode, outperforming the traditional ensemble mean approach. These findings underscore the utility of the proposed vector comparison metrics in evaluating dimensionality reduction techniques and their potential to enhance the analysis of complex flow fields, ultimately aiding in the identification of cycle-to-cycle variations (CCVs) in large in-cylinder flow datasets. Taken together, the use of SPDMD to extract a physically representative reference flow field and the modified-CMRI to quantify deviations from individual cycles, provides a unified framework for analyzing in-cylinder highly variable flows.

Detonation wave propagation and its interaction with obstacles are fundamental to industrial safety. In this study, detonation attenuation and reinitiation of stoichiometric hydrogen–air mixtures over a semi-cylinder are numerically studied with a calibrated chemical-diffusive model (CDM). Three regimes observed in experiments, including critical reinitiation, delayed reinitiation, and fully quenching, are well captured. It is demonstrated for the first time through a CDM model that the formation of transverse detonation by the Mach reflection behind the obstacle is a key factor for detonation reinitiation. For the critical reinitiation regime, a transverse detonation forms after a hotspot is directly triggered by the regular reflection near the obstacle. For the delayed reinitiation regime, a weak transverse detonation wave is generated by the Mach reflection on the obstacle, and then, it decays into a shock. The Mach shock leads to the detonation reinitiation. When the transverse detonation is not generated, the detonation is fully quenched by the cylinder. A flame diagnostic tool based on chemical explosive mode analysis highlighted the compressibility-induced ignition in the detonation reinitiation process. It is found that transverse detonations do not arise solely from the rapid consumption of the unburned pockets. Instead, they are initiated by the pressure amplification of reaction zones at interfaces between burned and unburned gases. The combustion rates in these zones are intensified through Richtmyer–Meshkov instabilities caused by the passage of transverse shock waves, which subsequently evolve into detonations.

We present EngineBench, the first machine learning (ML) benchmark designed for engine in-cylinder flow research. The benchmark data consist of curated particle image velocimetry (PIV) measurements previously gathered from the Transparent Combustion Chamber (TCC-III) by the General Motors University of Michigan Automotive Cooperative Research Laboratory. ¹ The dataset is then leveraged in order to benchmark the performance of four ML methods for a flow reconstruction (inpainting) task. We propose large gaps at the edges of the field of view as the benchmark task in order to reflect realistic scenarios in which data are harder to obtain closer to walls, and to challenge the ability of the models to predict the turbulent flow motion with limited access to surrounding data points. Pixel-wise, vector-based and multi-scale performance metrics are used to provide broad evaluations of the models. We find that models which utilise skip connections show significantly improved performances at this task on both small and large gap sizes, due to their enhanced ability to leverage contextual information. The benchmark proposed in this paper supports the development of ML models for engine design problems, as well as PIV data enhancement more generally. In addition, the ML model comparisons allow for more informed selection of models for problems in experimental flow diagnostics. All data and code are publicly available at https://eng.ox.ac.uk/tpsrg/research/enginebench /.

Parametric large eddy simulations (LES) of a supersonic lifted hydrogen flame are reported. The emphases are on two aspects: impacts of (1) Lewis number (Lei of the ith species) and (2) turbulent Schmidt and Prandtl numbers (Sct and Prt) on supersonic turbulent flame and flow structures. Five cases are considered: species-specific Lei, Sct=Prt=1.0 (C0); unity Lei, Sct=Prt=1.0 (C1); species-specific Lei, Sct=0.5, Prt=1.0 (C2); species-specific Lei, Sct=1.0, Prt=0.5 (C3); and species-specific Lei, Sct=Prt=0.5 (C4). Numerical results of instantaneous and/or time-averaged species mole fractions, mixture fraction, heat release rate, flame base location, and mixed modes of premixed and diffusion combustion are compared between cases C0 and C1. Differences in auto-ignition locations and strengths and flame structures and stabilization specify the impacts of Lewis number. They are triggered by different predictions of species mass and thermal diffusions at fuel-coflow and/or coflow-ambient air mixing layers. These differences are rationalized by a scale analysis of mass/thermal diffusion and convection for case C0, which suggests the relatively low but non-negligible former against the latter. Cases C0 and C2–C4 barely see differences in terms of instantaneous and/or time-averaged temperature, velocity, and mixed combustion modes except for further downstream areas where combustion occurs. Both Sct and Prt impose less significant influences than Lewis number, as sub-grid scale mass/thermal diffusion is subordinate to its resolved counterpart according to their scale analysis for case C4.