New parameter identification method based on bidirectional compression in the parameter space and time domain
New parameter identification method based on bidirectional compression in the parameter space and time domain
复制标题
DOI：
10.1360/sst-2020-0384
复制
发表时间：
2021-05
期刊：
影响因子：
--
通讯作者：
Y. Liang;Xiaowei Gao;M. Cui;Baojing Zheng
中科院分区：
文献类型：
--
作者：
Y. Liang;Xiaowei Gao;M. Cui;Baojing Zheng
极端温度下热工参数的实时监测对保证结构的安全稳定性具有重要意义。然而，高温高压环境增加了测量设备的失效风险，给现场测量带来困难。针对上述关键技术问题，本文研究了数据库获取和时域代理模型参数辨识方法。提出了一种通过压缩未知参数空间和时间域数据来测量热力参数的有效方法，该方法包括两个方面：（1）代理模型的构造和（2）参数的反演。一方面，采用拉丁超立方采样方法，在参数空间中获得合适的采样点密度;采用自由单元法离线采集采样点的瞬态温度信息，建立空间-参数-时间三维数据库。其次，通过在参数空间和时域中使用适当的正交分解方法来分解数据库。在时间序列中，通过对空间参数数据切片进行压缩，生成与POD基相关的参数;在参数空间中，对包含参数公共信息的POD基进行二次压缩，生成时间。然后，结合径向基函数代理模型，建立了一个重构的参数模型PSTDC代理模型。另一方面，从在低温区域中的位置处进行的瞬态温度测量获得的附加信息用于在最小二乘范数的最小化的基础上估计未知参数。所选用的优化算法是遗传算法，遗传算法是一种具有全局性和随机性的随机算法。通过这种方式，可以减小输入和输出的规模，并且可以实现未知参数与时域物理场之间的映射。另外，将计算量大的正演模拟离线进行，一旦建立了空间未知参数-时间数据库，在线参数辨识的效率大大提高。两个数值例子，即，通过对二维平板导热系数随温度变化和三维燃烧室对流换热系数随温度变化的分析，验证了该方法的正确性、稳定性和准确性。与CGM和LM算法相比，PSTDC代理模型的效率提高了一到两个数量级，并保证误差小于0.1%。通过对测量点位置和噪声的分析，证明了该方法的稳定性。
The real-time monitoring of thermal parameters under extreme temperature is important to ensure structural security and stability. However, high-temperature and high-pressure surroundings increase the failure risk of measurement devices and bring difficulties to in-situ measurement. To address the above key technical problems, this paper investigates database acquisition and time-domain surrogate models for parameter identification. An efficient method of measuring thermal parameters by compressing data in unknown parameter spaces and time domains is presented; the method consists of two aspects: construction of (1) the surrogate model and (2) the inversion parameters. On the one hand, Latin hypercube sampling is used to obtain an appropriate density of sample points in the parameter space. The transient temperature information of sample points is acquired offline via the free element method, which is used to establish a space-parameter-time 3D database. Next, the database is decomposed by using a proper orthogonal decomposition method in the parameter space and the time domain. In the time series, the parameter related to POD bases are generated by compressing the space-parameter data slice; in the parameter space, the POD bases, which contain common information about parameters, are secondarily compressed, and time is generated. Next, in combination with the radical basis function surrogate model, a reconstructed parameter model called PSTDC surrogate model is established. On the other hand, the additional information obtained from transient temperature measurements taken at locations in low-temperature areas is used for the estimation of the unknown parameters on the basis of the minimization of the least squares norm. The selected optimization algorithm is the genetic algorithm, which is a stochastic algorithm with globality and randomness. In this manner, the scale of input and output is reduced, and the mapping between unknown parameters and the time-domain physical field can be achieved. In addition, forward simulations with high computational cost is arranged offline; once the space-unknown parameter-time database is established, the efficiency of online parameter identification is greatly improved. Two numerical examples, i.e., 2D plate to temperature-dependent thermal conductivities and 3D combustor to coefficients of heat convection, are analyzed to verify the correctness, stability, and accuracy of the proposed method. Compared with the CGM and LM algorithms, the PSTDC surrogate model is more efficient by an order of magnitude or two, and it ensures an error of less than 0.1%. The stability of the proposed method is proven by the analysis of the location of measurement points and noise.