We develop a fast and effective method of multiscale analysis of composite materials with a complex structure. The complexity of the structure results from the $nano-TiC$ particles of the reinforcing phase being pushed into the interdendritic space during the manufacturing process. In such situations classic image analysis methods are ineffective because they are not able to resolve the high heterogeneity of structure and problems with phase separation. Moreover, even if the image analysis methods are successful, their application to calculate the macroscopic properties is highly problematic because it requires massive data processing. The developed Windows Washing methodology is a multiscale procedure which combines image analysis, statistics, Fourier analysis and analytical Representative Volume Element methods. From the mathematical standpoint we are interested in direct approximate computations of the basic, structural sums. All necessary input concerning the structure of materials is introduced from the real experimental data. There are some inherent limitations on experimental accuracy of determining the positions of particles. But they could be overcome by the complex methodology of Windows Washing. The more efforts we apply the more accurate reconstruction of the positions could be achieved. In fact, instead of real particles we use their images quantified by pixels. The qualitative relationship between the structural image and macro-properties of the $nano-TiC$ composite is found directly on the 2D structural image’s pixel level. The developed method establishes a solid foundation for building a database for classification of the composite structures. The appropriate architecture of data analysis, called Extract, Transform, and Load is included in the developed method.

我们开发了一种快速有效的多尺度分析复杂结构复合材料的方法。结构的复杂性源于$n\mathrm{一种}n○-吨\mathrm{一世}C$ 在制造过程中，增强相的颗粒被推入枝晶间空间。在这种情况下，经典的图像分析方法是无效的，因为它们无法解决结构的高度异质性和相分离问题。此外，即使图像分析方法成功，它们在计算宏观性质方面的应用也存在很大问题，因为它需要大量数据处理。开发的 Windows 清洗方法是一个多尺度程序，它结合了图像分析、统计、傅里叶分析和分析代表体积元素方法。从数学的角度来看，我们对基本结构和的直接近似计算感兴趣。所有关于材料结构的必要输入都是从真实的实验数据中引入的。确定粒子位置的实验精度存在一些固有的局限性。但是它们可以通过 Windows Washing 的复杂方法来克服。我们应用的努力越多，可以实现更准确的位置重建。事实上，我们使用的是按像素量化的图像，而不是真实的粒子。结构图像与宏观性质的定性关系 我们使用按像素量化的图像来代替真实的粒子。结构图像与宏观性质的定性关系 我们使用按像素量化的图像来代替真实的粒子。结构图像与宏观性质的定性关系$n\mathrm{一种}n○-吨\mathrm{一世}C$ 复合是直接在二维结构图像的像素级别上找到的。所开发的方法为建立复合结构分类数据库奠定了坚实的基础。开发的方法中包含适当的数据分析架构，称为提取、转换和加载。