Abstract
In computational fluid dynamics (CFD) for combustion, Flamelet-Generated Manifolds (FGM) are used to reduce chemical complexity by indexing flame states in a lookup table. However, when accounting for preferential diffusion, flamelets in an unnormalized progress variable ($PV$) subspace often have mismatched lengths. This research evaluates the resulting interpolation errors and tests several mitigation strategies, including additional normalization, optimized mixture fraction spacing, and perspective transformation mapping. Findings suggest that geometric remapping and normalization are essential for maintaining physical accuracy in complex fuel blends like $NH_3/H_2$.
Introduction: The Challenge of Preferential Diffusion
In combustion modeling, researchers often utilize Flamelet-Generated Manifolds (FGM) to manage the high computational cost of chemical kinetics. FGM acts as a pre-computed lookup table where the flame state is indexed by two primary variables: the mixture fraction ($Z$, representing the fuel-to-air ratio) and a progress variable ($PV$, representing reaction completion).
While $PV$ is typically normalized from 0 (unburnt) to 1 (fully burnt), accounting for preferential diffusion—where lighter molecules like Hydrogen diffuse faster than others—complicates this. In such cases, the mixture fraction is no longer constant along a flamelet, making it difficult to define consistent boundaries for $PV$. Consequently, many researchers interpolate in an unnormalized space, which introduces significant numerical errors.
The "Mismatch" Trap
The core issue arises when flamelets have different lengths in the progress variable subspace. If one flamelet ends at $PV=0.12$ and an adjacent one ends at $PV=0.14$, interpolating between them at $PV=0.13$ forces the model to average a "finished" flame with one that is still actively reacting. This leads to non-physical results and massive errors in predicting the reaction source term.
Methodology
The authors utilized a 1D FreeFlame model in Cantera to generate baseline data for an Ammonia/Hydrogen/Nitrogen ($NH_3/H_2/N_2$) fuel blend. They evaluated four strategies to mitigate interpolation errors:
- Additional Normalization: A lookup step to identify specific boundaries of the nearest flamelet before interpolation.
- Mixture Fraction ($Z$) Spacing: Increasing table density by reducing the interval between mixture fractions.
- Multi-Objective Optimization: Using a Genetic Algorithm (GA) to find a mathematical definition of $PV$ that is both monotonic and possesses uniform boundaries.
- Perspective Transformation Mapping: Applying image-processing techniques to "warp" the distorted $Z-PV$ quadrilateral into a uniform, normalized grid.
Key Results and Insights
The study revealed that standard linear interpolation in unnormalized space produces source term errors so severe they fail to resemble the original flamelet data. While decreasing the spacing between mixture fractions ($Z$) helps, it offers diminishing returns; accuracy gains became negligible once spacing dropped below 0.013.
Furthermore, the Genetic Algorithm failed to find a $PV$ definition that was both monotonic and perfectly uniform across boundaries for this complex fuel blend. This suggests that a single "perfect" progress variable may not exist for multi-fuel systems. However, Perspective Mapping and Additional Normalization proved highly effective, successfully bringing interpolated source terms back in line with 1D reference data.
Conclusion
Interpolation errors in unnormalized $PV$ subspaces are a critical bottleneck for accurate FGM modeling in the presence of preferential diffusion. While brute-force table densification is limited, geometric remapping and normalization steps provide robust solutions. Future work will focus on implementing these methods in full 3D turbulent combustion simulations to evaluate their impact on engine and burner performance predictions.
Summary updated