Využití směsových modelů pro identifikaci lomových ploch v kovových materiálech

Abstract

The presented thesis introduces the application of cluster algorithms in the brittle and ductile fracture identification in a steel structure. The fracture surfaces of eight DWTT broken samples are analysed using a new surface evaluation concept. The analysed data source comes from 3D scan. Beside formerly used fractal geometry and roughness approach, new concept based on normal vector characteristics is used. The implementation of a box counting estimate is verified on generated fractals and applied on the real DWTT specimens. Thereafter, normal vector characteristics, fractal dimensions and roughness characteristics are analysed by K-means++ and Gausssian mixture clustering. Applications of clustering algorithms improve a correct detection significantly, so that achieved clusters highly correspond to the real distribution of the ductile and brittle fracture areas on the DWTT surface. At the end of the thesis, the achieved results are enlarged by support vector machines technique. The presented methods are useful tools for the objective evaluation of the ductile fracture percentage.