دانلود رایگان مقاله انگلیسی طبقه بندی ریزساختار پیشرفته با روش داده کاوی - الزویر 2018

ABSTRACT

The mechanical properties of modern multi-phase materials significantly depend on the distribution, the shape and the size of the microstructural constituents. Thus, quantification and classification of the microstructure are decisive in identifying the underlying structure-property relationship of a specific material. Due to the complexity of the microstructure in modern materials, a reliable classification of microstructural constituents remains one of the biggest challenges in metallography. The present study demonstrates how data mining methods can be used to determine varying steel structures of two-phase steels by evaluating their morphological parameters. A data mining process was developed by using a support vector machine as classifier to build a model that is able to distinguish between different microstructures of the two-phase steels. The impact of preprocessing and feature selection methods on the classification result was tested.

5. Conclusion

The results of the present work demonstrate the feasibility of an objective classification of different structures in steel on the basis of morphological parameters by data mining methods. The process in Rapid Miner with the SVM as classifier showed good classification results for the three classes martensite, pearlite and bainite. We were able to show how to develop a process for the classification of microstructures. Furthermore, data preprocessing and feature selection could at the same time improve the classification results in order to make the model less complex and increase the generalization. Additionally, we found significant differences in the results between shuffled data split and sample-wise data split. The integration of substructure parameters in the classification process has shown high accuracy using fewer parameters. Comparable results for the morphological microstructural parameter were only possible with reproducible etching and segmentation methods. In order to further improve the accuracy, characteristic parameters of the second-phase objects and the substructure could be combined and more data should be produced in order to get statistically sufficient results and improve generalization.