A Fast Tomographic Reconstruction Method for Flame Temperature Distribution Measurement Based on Direct Solution Algorithm

Biao Zhang, Wei-Jian Peng, Jian Li, Zhihao Li, Chuanlong XuFrontiers in Energy Research 2021 · DOI

Abstract

The rapid and accurate measurement of the flame temperature distribution is of great significance to the structural design and health diagnosis of the engine. Aiming at the low reconstruction efficiency of traditional flame temperature distribution reconstruction algorithms, a Direct Solution algorithm for flame temperature distribution reconstruction is proposed in this paper based on the structural characteristics of the reconstruction equations. By setting several numerical cases, the performance of the Direct Solution algorithm and some commonly used traditional algorithms, such as Simultaneous Algebraic Reconstruction Technique (SART), Least Squares QR-factorization (LSQR) algorithm, Non-Negative Least Squares QR-factorization (NNLS) algorithm, is compared in the reconstruction of the flame temperature distribution. The results show that the efficiency of the Direct Solution method is 169.4, 7.4, and 3483.3 times higher than that of the SART, LSQR, and NNLS algorithms under the condition of 40 × 40 grids. In addition, with the increase of the number of grids, the growth rate of the reconstruction time of the Direct Solution algorithm is much lower than that of other algorithms. The overall reconstruction accuracy of the Direct Solution algorithm is better than that of SART and LSQR algorithms. This shows that it has an excellent comprehensive performance and has a great application prospect in the rapid reconstruction of the temperature distribution.

Introduction

In modern aero-engines, increasing the combustion chamber's outlet temperature is key to boosting the thrust-to-weight ratio. However, these extreme temperatures put immense thermal stress on turbine blades, leading to ablation and failure. To prevent this, engineers need to monitor the flame's temperature distribution in real-time.

The standard tool for this is optical tomography, which uses cameras to capture flame radiation from multiple angles and then "reconstructs" the internal temperature field. Mathematically, this involves solving a massive system of linear equations. Traditional iterative algorithms like SART (Simultaneous Algebraic Reconstruction Technique) or NNLS (Non-Negative Least Squares) are computationally expensive. As resolution increases, the time required to solve these equations explodes, making real-time diagnosis nearly impossible.

Methods

The researchers propose a Direct Solution algorithm that moves away from the "solve everything at once" iterative approach. Their method relies on two main components:

Forward Model

They use a generalized source term finite volume method to describe how light travels through the flame (the radiative transfer equation). This creates a coefficient matrix ($A$) that relates the measured light intensity to the unknown temperature at each grid point (voxel).

Schematic of the forward model showing light ray propagation through the flame voxel grid.

The "Solving-Substitution" Strategy

The breakthrough comes from analyzing the structure of matrix $A$. Because most light rays only pass through a few voxels, the matrix is incredibly sparse. The algorithm identifies rows with the fewest non-zero elements—essentially finding the "easiest" parts of the puzzle to solve first. It solves these small sub-problems using a standard solver (LSQR), then substitutes those known values back into the remaining equations. This process repeats, gradually eliminating unknowns until the entire field is reconstructed.

Key Insights and Ideas

The core innovation lies in treating sparsity as a feature rather than a bug. While most algorithms struggle with large, sparse matrices, the Direct Solution algorithm uses that sparsity to decompose a massive problem into a sequence of tiny, manageable ones.

Characteristics of the coefficient matrix A, highlighting the sparsity exploited by the Direct Solution algorithm.

By solving and substituting locally, this method avoids the "global" trap of iterative methods, which often spend significant time calculating updates for the entire field simultaneously. This approach also avoids the "semi-convergence" problems that plague traditional solvers like LSQR when applied to large-scale, ill-conditioned problems.

Key Results

The performance was tested across various resolutions, including $10 \times 10$, $20 \times 20$, $40 \times 40$, and $80 \times 80$ grids using a simulated 2D "four-peak" flame. On a $40 \times 40$ grid, the results were transformative:

  • Speed: The Direct Solution was 169.4 times faster than SART, 7.4 times faster than LSQR, and a staggering 3,483.3 times faster than NNLS.
  • Absolute Time: While NNLS took over 8,000 seconds (over 2 hours) to reconstruct the flame, the Direct Solution finished in just 2.31 seconds.
  • Accuracy: The Direct Solution outperformed SART and LSQR in accuracy. While slightly less accurate than NNLS (which uses strict physical constraints), the trade-off in speed makes it far more practical for engineering applications.

Limitations and Future Work

While the results are promising, the researchers identified several paths for future development. Integrating physical constraints (like non-negativity) into the Direct Solution without sacrificing speed is a priority. Additionally, while the logic holds for 3D voxels, the complexity of the "solving-substitution" sequence will increase significantly. Finally, moving from numerical simulations to real-world experimental validation with high-speed camera data from combustion test rigs will be the ultimate proof of the algorithm's utility in aero-engine health diagnosis.

Summary updated

Abstract

The rapid and accurate measurement of the flame temperature distribution is of great significance to the structural design and health diagnosis of the engine. Aiming at the low reconstruction efficiency of traditional flame temperature distribution reconstruction algorithms, a Direct Solution algorithm for flame temperature distribution reconstruction is proposed in this paper based on the structural characteristics of the reconstruction equations. By setting several numerical cases, the performance of the Direct Solution algorithm and some commonly used traditional algorithms, such as Simultaneous Algebraic Reconstruction Technique (SART), Least Squares QR-factorization (LSQR) algorithm, Non-Negative Least Squares QR-factorization (NNLS) algorithm, is compared in the reconstruction of the flame temperature distribution. The results show that the efficiency of the Direct Solution method is 169.4, 7.4, and 3483.3 times higher than that of the SART, LSQR, and NNLS algorithms under the condition of 40 × 40 grids. In addition, with the increase of the number of grids, the growth rate of the reconstruction time of the Direct Solution algorithm is much lower than that of other algorithms. The overall reconstruction accuracy of the Direct Solution algorithm is better than that of SART and LSQR algorithms. This shows that it has an excellent comprehensive performance and has a great application prospect in the rapid reconstruction of the temperature distribution.
BibTeX
@inproceedings{zhang2021fast,
  author = {Biao Zhang and Wei-Jian Peng and Jian Li and Zhihao Li and Chuanlong Xu},
  title = {{A Fast Tomographic Reconstruction Method for Flame Temperature Distribution Measurement Based on Direct Solution Algorithm}},
  booktitle = {Frontiers in Energy Research},
  year = {2021},
  doi = {10.3389/fenrg.2021.790581},
}
Figure