Analysis of spatial-temporal dynamics of cool flame oscillation phenomenon occurred around a fuel droplet array by using variational auto-encoder
Analysis of spatial-temporal dynamics of cool flame oscillation phenomenon occurred around a fuel droplet array by using variational auto-encoder
复制标题
用变分自动编码器分析燃料液滴阵列周围冷火焰振荡现象的时空动力学
DOI：
10.1016/j.proci.2022.09.047
复制
发表时间：
2022-11
影响因子：
3.4
通讯作者：
Kazuki Iemura;Masanori Saito;Y. Suganuma;M. Kikuchi;Y. Inatomi;H. Nomura;Mitsuaki Tanabe
中科院分区：
文献类型：
--
作者：
Kazuki Iemura;Masanori Saito;Y. Suganuma;M. Kikuchi;Y. Inatomi;H. Nomura;Mitsuaki Tanabe
采用液滴汽化/自燃数值模型，对高温空气中燃料液滴阵列的冷火焰振荡进行了数值模拟。获得了温度和化学物种空间分布的时间序列数据。这些数据用于训练变分自动编码器（VAE），以减少数据的维数。将冷火焰振荡现象映射为由VAE得到的两个隐变量所构成的相平面上的轨迹。振荡现象进行了研究，通过使用VAE，区分温度和物种状态的分布模式。对VAE的解码器输出进行适当的正交分解。振荡机制进行了研究的空间本征函数（模式地图）。三个主导模式的时间本征函数被显示在平面的轨迹上。研究了物理量分布之间的相关性，以研究冷火焰动力学。从以上研究中，确认了平面可以区分物理状态，因为在振荡期间轨迹不相交。将平面视为状态空间。从模式图和时间特征函数中识别出与每个模式相关的物理现象。通过沿轨迹沿着检查时间特征函数来识别相关现象出现的阶段。利用相关图对振荡机理进行了讨论。
Cool flame oscillation of a fuel droplet array in high-temperature air was numerically simulated by using a droplet vaporization/spontaneous ignition numerical model. The time series data of temperature and chemical species spatial distributions were obtained. The data were used to train a Variational Auto-Encoder (VAE) that reduces the dimension of the data. The cool flame oscillation phenomenon was mapped as the trajectory onto a phase plane spanned by two latent variables obtained by the VAE. The oscillation phenomenon was investigated by using the distribution patterns derived by the VAE that distinguishes the temperature and the species states. Proper orthogonal decomposition was carried out on the decoder output of the VAE. The oscillation mechanism was investigated by the spatial eigenfunctions (mode maps). The temporal eigenfunctions of the three dominant modes were shown onto the trajectory of the plane. The correlation among physical variable distributions was evaluated to investigate the cool flame dynamics. From the above investigation, the plane was confirmed to distinguish the physical states, for the trajectory did not intersect during the oscillation. The plane was treated as a state space. The physical phenomena associated with each mode were identified from the mode maps and the temporal eigenfunctions. The phase in which the associated phenomenon arises was identified by checking the temporal eigenfunctions along with the trajectory. The mechanism of the oscillation was discussed with the correlation diagrams.