在携带颗粒的湍流场中，与两个颗粒的分离距离相关的双颗粒统计量，如表征颗粒聚团的径向分布函数、颗粒相对速度和相对扩散等，强烈依赖于小尺度湍流脉动。在通常的大涡模拟方法中，求解滤波的Navier-Stokes(NS)方程只能得到大尺度湍流结构。小尺度流动结构的缺失导致在欧拉-拉格朗日框架下大涡模拟预测双颗粒统计量在小和中等Stokes数(St<3) 时具有很大误差。这些误差主要来源于(1)对NS方程的滤波运算，(2)封闭被滤波后NS方程的亚格子模型误差。针对这两个误差，我们分别发展基于微分滤波的近似反卷积模型和基于最小多尺度拉格朗日映射方法的亚格子非高斯合成湍流模型。目标是提高解析尺度湍流的模拟准确度和构造缺失的亚格子湍流，发展中等计算成本，并能准确预测湍流中双颗粒统计量的大涡模拟方法。预期成果可以提高湍流二相流大涡模拟预测双颗粒统计量的准确度，选题具有重要的科学意义和应用价值。

The two-particle statistics in particle-laden turbulent flow fields such as the radial distribution function, relative motions and relative dispersion strongly depend on the small-scale motions in such flows. However, the conventional large-eddy simulation (LES) can only resolve large-scale motions by solving the filtered Navier-Stokes (NS) equations which are closed by the subgrid scale models such as the dynamic Smagorinsky model. The missing SGS motions lead to large predictive errors in LES of these two-particle statistics for heavy particles at small and intermediate Stokes numbers (St<3) in the framework of the Eulerian-Lagrangian description. These errors mainly come from (1) the filtering operation on the NS equations and (2) the subgrid scale models used to close the filtered NS equations. To overcome these problems, we will develop the approximate deconvolution method based on the differential filter and the subgrid scale synthetic NON-GAUSSIAN turbulent flow field based on the minimum multi-scale Lagrangian mapping method. The synthetic turbulent flow fields have the dynamic properties such as intermittency of real turbulence to some extent. The project aims at developing the LES method to predict the two-particle statistics accurately at modest computational cost. The purposed models are of significance in improving the predictive accuracy of the turbulent two-phase flows which widely exist in engineering and environmental flows.