1.National Key Laboratory of Transient Physics, Nanjing University of Science and Technology, Nanjing 210094, China 2.Department of Electrical Engineering and Electronics, University of Liverpool, Liverpool L69 3GJ, UK
Fund Project:Project supported by the China Scholarship Council (Grant No. 201906845059), the Young Scientists Found of the Natural Science Foundation of Jiangsu Province, China (Grant No. BK20190439), and the Fundamental Research Funds of the Central Universities, China (Grant No. 30919011258)
Received Date:17 July 2020
Accepted Date:14 August 2020
Available Online:14 November 2020
Published Online:20 November 2020
Abstract:Beam arrangement with limited projections is a critical part of research on tunable diode laser absorption tomography reconstruction for combustion diagnosis. Based on the efforts to regularize this rank-deficient and ill-posed problem with Tikhonov regularization, a novel approach to using the regularization parameter matrix is developed for designing optical component layout and predicting the reconstruction accuracy. Objective function of beam arrangement is established by the rigorous mathematical derivation, and genetic algorithm is adopted to realize the optimization of function to overcome the difficulty associated with the multimodal nature of the problem. Nonuniform distribution properties of matrix elements in physical space relate to location and alignment of the laser/detector pairs, and form a basis for adjusting the weight between measurement and regularization to improve the reconstruction performance. A mathematical model of double Gauss distributions is established in a 10 × 10 element discrete tomography domain, and typically 20 measurement beams scanning the H2O transition at 7185.6 cm–1 are available to probe the domain of interest. The systematic comparison between optimized beam array here and four existing beam arrangements in the literature is analyzed to validate the method. The reconstruction with Tikhonov regularization parameter matrix shows obvious advantages of reducing errors especially under the condition of fewer projections. The validation of reconstruction performance of the optimized beam array is also examined by simulating the laser absorption measurement which is carried out on phantoms generated using a simulation of external flow field of an air-gasoline pulsed detonation engine. The result shows that the optimized beam array consistently outperforms other arrangements reported in complicated fluid field. A demonstration reconstruction experiment is performed on the distribution from small gas burners. Both locations and amplitudes are in good agreement with those in the actual case. This proposed design method will be valuable in broadening the scope of applications of tunable diode laser absorption tomography reconstruction for engine diagnosis and combustion efficiency improvement. Keywords:laser absorption spectroscopy/ tomography reconstruction/ Tikhonov regularization/ singular value decomposition
利用Tikhonov正则化参数矩阵${{\lambda }}$, 根据目标函数(12)式并结合遗传算法获得优化后的激光吸收光谱二维测量光路布置方式. 为验证该光路布置方式重建效果, 分别与以下光路布置方式重建结果进行对比: 方式(a)2投影角度平行光路布置方式; 方式(b)4投影角度扇形光路布置方式; 方式(c)基于Mod设计的光路布置方式; 方式(d)基于单一正则化参数设计的光路布置方式; 方式(e)基于正则化参数矩阵设计的光路布置方式. 光路布置方式及对应投影域如图6所示. 图 6 5种光路布置方式空间分布图与对应投影域图 (a) 2 × 10平行光路布置方式; (b) 4 × 5扇形光路布置方式; (c) 基于Mod设计的光路布置方式; (d) 基于单一正则化参数设计的光路布置方式; (e) 基于正则化参数矩阵设计的光路布置方式 Figure6. Five example beam configuration in the physical space and in Radon space: (a) 2 × 10 parallel beams arrangement; (b) 4 × 5 fanned beams arrangement; (c) beams arrangement designed based on MOD method; (d) beams arrangement designed based on single regularization parameter; (e) beams arrangement designed based on regularization parameter matrix.
在测量区域内建立二维分布模型. 测量对象为燃烧环境下H2O组分, 测量谱线为近红外波段7185.6 cm–1处H2O特征吸收谱线, 重建模型采用双峰高斯分布函数. 将测量区域离散化为10 × 10网格, 将20条光路以图6所述方式进行布置. 为了便于对比, 每种光路布置方式均配合Tikhonov正则化直接计算方法获得重建结果. 其中, 前4种光路布置方式采用单一正则化参数, 正则化参数设定为0.5. 方式(e)为基于正则化参数矩阵设计的光路布置方式, 正则化参数矩阵中元素取值范围为0.1至1. 重建模型与不同光路布置方式重建结果如图7所示. 图 7 二维重建模型与不同光路布置方式重建结果 (a) 重建模型; (b) 2 × 10平行光路布置方式; (c) 4 × 5扇形光路布置方式; (d) 基于Mod设计的光路布置方式; (e) 基于单一正则化参数设计的光路布置方式; (f) 基于正则化参数矩阵设计的光路布置方式 Figure7. Phantom and reconstruction results from different beam arrangement: (a) Phantom; (b) 2 × 10 parallel beams arrangement; (c) 4 × 5 fanned beams arrangement; (d) beams arrangement designed based on MOD method; (e) beams arrangement designed based on single regularization parameter; (f) beams arrangement designed based on regularization parameter matrix.
利用图6(e)中设计得到的光路布置方式, 针对重建模型获取光路投影向量${{b}}$. 通过对投影向量${{b}}$添加不同比例的噪音, 模拟实际测量过程中噪音对于重建精度的影响. 重建误差$\varepsilon $随噪音幅值的变化曲线如图8所示. 图 8 投影数据噪音对重建误差的影响 Figure8. Effect of noise in projections on reconstruction error.
可以看出, 尽管重建误差$\varepsilon $随着投影向量${{b}}$中随机白噪音幅值的增加而增大, 但增长幅度较缓. 在噪音幅值从0提高到10 %过程中, 重建误差$\varepsilon $仅由0.178增加至0.197. 分析其原因, 主要是由于重建过程中Tikhonov正则化通过增加约束方程组控制解的平滑性. 当正则化参数矩阵选取范围适当时, $\left\| {{{\lambda Lx}}} \right\|$将会明显抑制由投影向量${{b}}$中噪音所引起的离散化网格间吸收系数的剧烈变化, 其功能相当于对重建结果进行滤波, 在一定程度上达到抑制噪音的效果, 减小了噪音对重建的影响. 4.针对爆轰管外流场二维重建仿真分析脉冲爆轰发动机(PDE)管外流场变化剧烈, 是验证光路布置方式和重建方法有效性的理想模型, 因此本文中开展了针对PDE管外流场二维分布重建的模拟测量研究. 采用数值计算方法获取爆轰过程外流场分布情况, 以此作为模型考察上述几种光路布置方式对该流场的重建效果. 根据气液两相PDE物理模型(图9), 建立考虑黏性影响下的PDE外流场轴对称控制方程, 采用时空守恒元/求解元(CE/SE)方法进行求解. 模型中PDE长为1.4 m, 口径为0.08 m, 以汽油和空气为燃料和氧化剂, 初始温度压力分别为296 K和1 atm (1 atm=101325 Pa), 填充系数为1. 计算过程中忽略PDE管壁与外界的热交换, 化学反应释放的能量仅被气体吸收. 通过仿真计算得到爆轰过程外流场温度组分分布, 以此作为后续激光吸收光谱二维重建模型. 详细的PDE模型和计算方法见文献[20]. 图 9 气液两相爆轰外流场重建模型 Figure9. Model of simulated tomography measurement in a two-phase detonation flow.
测量区域位于距离管口10 cm处, 为20 cm × 20 cm矩形区域. 将测量区域离散化, 采用20条测量光路建立系数矩阵${{A}}$, 采用7185.6 cm–1处H2O吸收谱线获取投影值并建立投影向量${{b}}$. 利用图6中5种不同光路布置方式分别对PDE爆轰波溢出管口后20 ms时间段内管外流场分布情况进行重建. 重建误差变化情况如图10所示. 图 10 基于图6不同光路布置方式的气液两相爆轰外流场重建结果对比 Figure10. Comparison of reconstruction errors for the external flow field of two-phase detonation based on beam arrangements in Fig.6.
在PDE爆轰波溢出管口后20 ms时间段内, 管外流场温度变化剧烈, 测量区域内最高温度由1480 K迅速降低至570 K, 温度场影响范围由8 cm × 8 cm增加至17 cm × 17 cm. 由图10可以看出, 基于光路布置方式(d)和光路布置方式(e)的燃烧场重建效果较好, 重建误差$\varepsilon $远低于其他3种光路布置方式, 结论与双峰高斯分布模型重建结果一致. 相对而言, 基于正则化参数矩阵的光路布置方式(e)所对应的重建效果优于光路布置方式(d), 平均重建误差降低0.012. 5.燃烧场二维分布重建实验研究以小型燃气炉平面火焰为研究对象, 在实验室内开展了燃烧场二维分布重建实验研究, 测试系统示意图如图11所示. 测量采用Nel公司DFB可调谐半导体激光器, 中心波数为7185.6 cm–1, 通过利用信号发生器驱动半导体激光器以1 kHz频率扫描H2O特征吸收谱线. 激光束由光纤连接至测试区域, 经过分光器后分为20条光路, 以图6(e)所示方式进行布置. 测量光路透射过25 cm × 25 cm矩形测试区域后, 测量得到的投影信号经过光电探测器接收, 由数据采集系统采集记录. 测量过程中波长变化由标准具进行监测. 图 11 激光吸收光谱二维重建实验示意图 Figure11. Experiment on tomography reconstruction based on tunable diode laser absorption.
燃气炉直径为7.6 cm, 测量截面位于燃气炉上方2 cm处, 热电偶测量温度约为1210 K. 根据测量得到的20条光路投影数据, 利用Tikhonov正则化方法分别对测量区域内单个和多个燃气炉燃烧场分布情况进行重建. 图12为计算得到的单个和两个燃气炉分别布置于测量区域内的重建结果. 可以明显看出, 重建结果可以准确反应出燃烧场分布情况, 重建位置与燃气炉位置完全吻合. 对于双炉重建结果而言, 两个光谱吸收系数峰值较为接近, 与实际情况一致, 表明基于Tikhonov正则化参数矩阵的光路设计与重建方法可用于有限投影条件下的燃烧场诊断. 图 12 燃气炉重建结果 (a) 单炉实验结果; (b) 双炉实验结果 Figure12. Reconstruction results of gas burners: (a) single burner; (b) double burners.