Simultaneous identification of transfer functions and combustion noise of a turbulent flame
Simultaneous identification of transfer functions and combustion noise of a turbulent flame
复制标题
DOI：
10.1016/j.jsv.2018.02.040
复制
发表时间：
2018-05-26
影响因子：
4.7
通讯作者：
Polifke, W.
中科院分区：
文献类型：
--
作者：
Merk, M.;Jaensch, S.;Polifke, W.
大涡模拟/系统识别（LES/SI）方法允许从湍流反应流的LES中推断出火焰传递函数（FTF）：通过SI技术对模拟湍流火焰的宽带激发引起的参考速度波动和全局热释放率的时间序列进行后处理，从而推导出火焰动力学的低阶模型，由此可以很容易地推断出FTF。目前的工作研究了已建立的LES/SI方法的扩展：除了估计FTF外，还从相同的时间序列数据推导出燃烧噪声源的低阶模型。通过将这种噪声模型纳入线性热声模型，可以预测不表现自激热声不稳定性的密闭燃烧系统中声压的总体水平和频谱分布。在本研究中，测试了各种用于估计噪声模型的模型结构。比较了这些模型结构的适用性和质量，研究了它们对某些时间序列特性的敏感性。考察了时间序列长度、信噪比和边界条件声反射系数对识别的影响。结果表明，对于同时识别描述FTF和燃烧噪声源的模型，Box-Jenkins模型结构优于更简单的方法。在最合适的模型结构问题之后，讨论了最优模型顺序的选择，特别是噪声模型的最优参数化不明显。运用赤池信息准则和模型残差分析，得出最合适的模型阶数的定性和定量结论。所有的调查都是基于代理数据模型，它允许蒙特卡罗研究跨越一个大的参数空间，适度的计算工作量。所进行的研究为将先进的SI技术应用于实际的LES数据奠定了坚实的基础。(C) 2018 Elsevier Ltd.版权所有。
The Large Eddy Simulation/System Identification (LES/SI) approach allows to deduce a flame transfer function (FTF) from LES of turbulent reacting flow: Time series of fluctuations of reference velocity and global heat release rate resulting from broad-band excitation of a simulated turbulent flame are post-processed via SI techniques to derive a low order model of the flame dynamics, from which the FTF is readily deduced. The current work investigates an extension of the established LES/SI approach: In addition to estimation of the FTF, a low order model for the combustion noise source is deduced from the same time series data. By incorporating such a noise model into a linear thermoacoustic model, it is possible to predict the overall level as well as the spectral distribution of sound pressure in confined combustion systems that do not exhibit self-excited thermoacoustic instability. A variety of model structures for estimation of a noise model are tested in the present study. The suitability and quality of these model structures are compared against each other, their sensitivity regarding certain time series properties is studied. The influence of time series length, signal-to-noise ratio as well as acoustic reflection coefficient of the boundary conditions on the identification are examined. It is shown that the Box-Jenkins model structure is superior to simpler approaches for the simultaneous identification of models that describe the FTF as well as the combustion noise source. Subsequent to the question of the most adequate model structure, the choice of optimal model order is addressed, as in particular the optimal parametrization of the noise model is not obvious. Akaike's Information Criterion and a model residual analysis are applied to draw qualitative and quantitative conclusions on the most suitable model order. All investigations are based on a surrogate data model, which allows a Monte Carlo study across a large parameter space with modest computationally effort. The conducted study constitutes a solid basis for the application of advanced SI techniques to actual LES data. (C) 2018 Elsevier Ltd. All rights reserved.